A.I. Artificial Intelligence (2001) – IMDb

Edit Storyline

In the not-so-far future the polar ice caps have melted and the resulting rise of the ocean waters has drowned all the coastal cities of the world. Withdrawn to the interior of the continents, the human race keeps advancing, reaching the point of creating realistic robots (called mechas) to serve them. One of the mecha-producing companies builds David, an artificial kid which is the first to have real feelings, especially a never-ending love for his “mother”, Monica. Monica is the woman who adopted him as a substitute for her real son, who remains in cryo-stasis, stricken by an incurable disease. David is living happily with Monica and her husband, but when their real son returns home after a cure is discovered, his life changes dramatically. Written byChris Makrozahopoulos

Budget:$100,000,000 (estimated)

Opening Weekend USA: $29,352,630,1 July 2001, Wide Release

Gross USA: $78,616,689, 23 September 2001

Cumulative Worldwide Gross: $235,927,000

Runtime: 146 min

Aspect Ratio: 1.85 : 1

Continue reading here:

A.I. Artificial Intelligence (2001) – IMDb

Artificial intelligence – Wikipedia

Intelligence demonstrated by machines

Artificial intelligence (AI), sometimes called machine intelligence, is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans and other animals. In computer science AI research is defined as the study of “intelligent agents”: any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] Colloquially, the term “artificial intelligence” is applied when a machine mimics “cognitive” functions that humans associate with other human minds, such as “learning” and “problem solving”.

The scope of AI is disputed: as machines become increasingly capable, tasks considered as requiring “intelligence” are often removed from the definition, a phenomenon known as the AI effect, leading to the quip, “AI is whatever hasn’t been done yet.”[3] For instance, optical character recognition is frequently excluded from “artificial intelligence”, having become a routine technology.[4] Capabilities generally classified as AI as of 2017[update] include successfully understanding human speech, competing at the highest level in strategic game systems (such as chess and Go),[6] autonomous cars, intelligent routing in content delivery network and military simulations.

Artificial intelligence was founded as an academic discipline in 1956, and in the years since has experienced several waves of optimism,[7][8] followed by disappointment and the loss of funding (known as an “AI winter”),[9][10] followed by new approaches, success and renewed funding.[8][11] For most of its history, AI research has been divided into subfields that often fail to communicate with each other.[12] These sub-fields are based on technical considerations, such as particular goals (e.g. “robotics” or “machine learning”),[13] the use of particular tools (“logic” or artificial neural networks), or deep philosophical differences.[14][15][16] Subfields have also been based on social factors (particular institutions or the work of particular researchers).[12]

The traditional problems (or goals) of AI research include reasoning, knowledge representation, planning, learning, natural language processing, perception and the ability to move and manipulate objects.[13] General intelligence is among the field’s long-term goals.[17] Approaches include statistical methods, computational intelligence, and traditional symbolic AI. Many tools are used in AI, including versions of search and mathematical optimization, artificial neural networks, and methods based on statistics, probability and economics. The AI field draws upon computer science, mathematics, psychology, linguistics, philosophy and many others.

The field was founded on the claim that human intelligence “can be so precisely described that a machine can be made to simulate it”.[18] This raises philosophical arguments about the nature of the mind and the ethics of creating artificial beings endowed with human-like intelligence which are issues that have been explored by myth, fiction and philosophy since antiquity.[19] Some people also consider AI to be a danger to humanity if it progresses unabatedly.[20] Others believe that AI, unlike previous technological revolutions, will create a risk of mass unemployment.[21]

In the twenty-first century, AI techniques have experienced a resurgence following concurrent advances in computer power, large amounts of data, and theoretical understanding; and AI techniques have become an essential part of the technology industry, helping to solve many challenging problems in computer science.[22][11]

Thought-capable artificial beings appeared as storytelling devices in antiquity,[23] and have been common in fiction, as in Mary Shelley’s Frankenstein or Karel apek’s R.U.R. (Rossum’s Universal Robots).[24] These characters and their fates raised many of the same issues now discussed in the ethics of artificial intelligence.[19]

The study of mechanical or “formal” reasoning began with philosophers and mathematicians in antiquity. The study of mathematical logic led directly to Alan Turing’s theory of computation, which suggested that a machine, by shuffling symbols as simple as “0” and “1”, could simulate any conceivable act of mathematical deduction. This insight, that digital computers can simulate any process of formal reasoning, is known as the ChurchTuring thesis.[25] Along with concurrent discoveries in neurobiology, information theory and cybernetics, this led researchers to consider the possibility of building an electronic brain. Turing proposed that “if a human could not distinguish between responses from a machine and a human, the machine could be considered intelligent”.[26] The first work that is now generally recognized as AI was McCullouch and Pitts’ 1943 formal design for Turing-complete “artificial neurons”.

The field of AI research was born at a workshop at Dartmouth College in 1956.[28] Attendees Allen Newell (CMU), Herbert Simon (CMU), John McCarthy (MIT), Marvin Minsky (MIT) and Arthur Samuel (IBM) became the founders and leaders of AI research.[29] They and their students produced programs that the press described as “astonishing”: computers were learning checkers strategies (c. 1954)[31] (and by 1959 were reportedly playing better than the average human),[32] solving word problems in algebra, proving logical theorems (Logic Theorist, first run c. 1956) and speaking English.[33] By the middle of the 1960s, research in the U.S. was heavily funded by the Department of Defense[34] and laboratories had been established around the world.[35] AI’s founders were optimistic about the future: Herbert Simon predicted, “machines will be capable, within twenty years, of doing any work a man can do”. Marvin Minsky agreed, writing, “within a generation… the problem of creating ‘artificial intelligence’ will substantially be solved”.[7]

They failed to recognize the difficulty of some of the remaining tasks. Progress slowed and in 1974, in response to the criticism of Sir James Lighthill and ongoing pressure from the US Congress to fund more productive projects, both the U.S. and British governments cut off exploratory research in AI. The next few years would later be called an “AI winter”,[9] a period when obtaining funding for AI projects was difficult.

In the early 1980s, AI research was revived by the commercial success of expert systems,[37] a form of AI program that simulated the knowledge and analytical skills of human experts. By 1985 the market for AI had reached over a billion dollars. At the same time, Japan’s fifth generation computer project inspired the U.S and British governments to restore funding for academic research.[8] However, beginning with the collapse of the Lisp Machine market in 1987, AI once again fell into disrepute, and a second, longer-lasting hiatus began.[10]

In the late 1990s and early 21st century, AI began to be used for logistics, data mining, medical diagnosis and other areas.[22] The success was due to increasing computational power (see Moore’s law), greater emphasis on solving specific problems, new ties between AI and other fields (such as statistics, economics and mathematics), and a commitment by researchers to mathematical methods and scientific standards.[38] Deep Blue became the first computer chess-playing system to beat a reigning world chess champion, Garry Kasparov on 11 May 1997.

In 2011, a Jeopardy! quiz show exhibition match, IBM’s question answering system, Watson, defeated the two greatest Jeopardy! champions, Brad Rutter and Ken Jennings, by a significant margin. Faster computers, algorithmic improvements, and access to large amounts of data enabled advances in machine learning and perception; data-hungry deep learning methods started to dominate accuracy benchmarks around 2012.[41] The Kinect, which provides a 3D bodymotion interface for the Xbox 360 and the Xbox One use algorithms that emerged from lengthy AI research[42] as do intelligent personal assistants in smartphones.[43] In March 2016, AlphaGo won 4 out of 5 games of Go in a match with Go champion Lee Sedol, becoming the first computer Go-playing system to beat a professional Go player without handicaps.[6][44] In the 2017 Future of Go Summit, AlphaGo won a three-game match with Ke Jie,[45] who at the time continuously held the world No. 1 ranking for two years.[46][47] This marked the completion of a significant milestone in the development of Artificial Intelligence as Go is an extremely complex game, more so than Chess.

According to Bloomberg’s Jack Clark, 2015 was a landmark year for artificial intelligence, with the number of software projects that use AI within Google increased from a “sporadic usage” in 2012 to more than 2,700 projects. Clark also presents factual data indicating that error rates in image processing tasks have fallen significantly since 2011.[48] He attributes this to an increase in affordable neural networks, due to a rise in cloud computing infrastructure and to an increase in research tools and datasets.[11] Other cited examples include Microsoft’s development of a Skype system that can automatically translate from one language to another and Facebook’s system that can describe images to blind people.[48] In a 2017 survey, one in five companies reported they had “incorporated AI in some offerings or processes”.[49][50]

A typical AI perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] An AI’s intended goal function can be simple (“1 if the AI wins a game of Go, 0 otherwise”) or complex (“Do actions mathematically similar to the actions that got you rewards in the past”). Goals can be explicitly defined, or can be induced. If the AI is programmed for “reinforcement learning”, goals can be implicitly induced by rewarding some types of behavior and punishing others.[a] Alternatively, an evolutionary system can induce goals by using a “fitness function” to mutate and preferentially replicate high-scoring AI systems; this is similar to how animals evolved to innately desire certain goals such as finding food, or how dogs can be bred via artificial selection to possess desired traits. Some AI systems, such as nearest-neighbor, instead reason by analogy; these systems are not generally given goals, except to the degree that goals are somehow implicit in their training data. Such systems can still be benchmarked if the non-goal system is framed as a system whose “goal” is to successfully accomplish its narrow classification task.[53]

AI often revolves around the use of algorithms. An algorithm is a set of unambiguous instructions that a mechanical computer can execute.[b] A complex algorithm is often built on top of other, simpler, algorithms. A simple example of an algorithm is the following recipe for optimal play at tic-tac-toe:

Many AI algorithms are capable of learning from data; they can enhance themselves by learning new heuristics (strategies, or “rules of thumb”, that have worked well in the past), or can themselves write other algorithms. Some of the “learners” described below, including Bayesian networks, decision trees, and nearest-neighbor, could theoretically, if given infinite data, time, and memory, learn to approximate any function, including whatever combination of mathematical functions would best describe the entire world. These learners could therefore, in theory, derive all possible knowledge, by considering every possible hypothesis and matching it against the data. In practice, it is almost never possible to consider every possibility, because of the phenomenon of “combinatorial explosion”, where the amount of time needed to solve a problem grows exponentially. Much of AI research involves figuring out how to identify and avoid considering broad swaths of possibilities that are unlikely to be fruitful.[55] For example, when viewing a map and looking for the shortest driving route from Denver to New York in the East, one can in most cases skip looking at any path through San Francisco or other areas far to the West; thus, an AI wielding an pathfinding algorithm like A* can avoid the combinatorial explosion that would ensue if every possible route had to be ponderously considered in turn.[57]

The earliest (and easiest to understand) approach to AI was symbolism (such as formal logic): “If an otherwise healthy adult has a fever, then they may have influenza”. A second, more general, approach is Bayesian inference: “If the current patient has a fever, adjust the probability they have influenza in such-and-such way”. The third major approach, extremely popular in routine business AI applications, are analogizers such as SVM and nearest-neighbor: “After examining the records of known past patients whose temperature, symptoms, age, and other factors mostly match the current patient, X% of those patients turned out to have influenza”. A fourth approach is harder to intuitively understand, but is inspired by how the brain’s machinery works: the artificial neural network approach uses artificial “neurons” that can learn by comparing itself to the desired output and altering the strengths of the connections between its internal neurons to “reinforce” connections that seemed to be useful. These four main approaches can overlap with each other and with evolutionary systems; for example, neural nets can learn to make inferences, to generalize, and to make analogies. Some systems implicitly or explicitly use multiple of these approaches, alongside many other AI and non-AI algorithms;[58] the best approach is often different depending on the problem.[60]

Learning algorithms work on the basis that strategies, algorithms, and inferences that worked well in the past are likely to continue working well in the future. These inferences can be obvious, such as “since the sun rose every morning for the last 10,000 days, it will probably rise tomorrow morning as well”. They can be nuanced, such as “X% of families have geographically separate species with color variants, so there is an Y% chance that undiscovered black swans exist”. Learners also work on the basis of “Occam’s razor”: The simplest theory that explains the data is the likeliest. Therefore, to be successful, a learner must be designed such that it prefers simpler theories to complex theories, except in cases where the complex theory is proven substantially better. Settling on a bad, overly complex theory gerrymandered to fit all the past training data is known as overfitting. Many systems attempt to reduce overfitting by rewarding a theory in accordance with how well it fits the data, but penalizing the theory in accordance with how complex the theory is. Besides classic overfitting, learners can also disappoint by “learning the wrong lesson”. A toy example is that an image classifier trained only on pictures of brown horses and black cats might conclude that all brown patches are likely to be horses. A real-world example is that, unlike humans, current image classifiers don’t determine the spatial relationship between components of the picture; instead, they learn abstract patterns of pixels that humans are oblivious to, but that linearly correlate with images of certain types of real objects. Faintly superimposing such a pattern on a legitimate image results in an “adversarial” image that the system misclassifies.[c][63][64][65]

Compared with humans, existing AI lacks several features of human “commonsense reasoning”; most notably, humans have powerful mechanisms for reasoning about “nave physics” such as space, time, and physical interactions. This enables even young children to easily make inferences like “If I roll this pen off a table, it will fall on the floor”. Humans also have a powerful mechanism of “folk psychology” that helps them to interpret natural-language sentences such as “The city councilmen refused the demonstrators a permit because they advocated violence”. (A generic AI has difficulty inferring whether the councilmen or the demonstrators are the ones alleged to be advocating violence.)[68][69][70] This lack of “common knowledge” means that AI often makes different mistakes than humans make, in ways that can seem incomprehensible. For example, existing self-driving cars cannot reason about the location nor the intentions of pedestrians in the exact way that humans do, and instead must use non-human modes of reasoning to avoid accidents.[71][72][73]

The overall research goal of artificial intelligence is to create technology that allows computers and machines to function in an intelligent manner. The general problem of simulating (or creating) intelligence has been broken down into sub-problems. These consist of particular traits or capabilities that researchers expect an intelligent system to display. The traits described below have received the most attention.[13]

Early researchers developed algorithms that imitated step-by-step reasoning that humans use when they solve puzzles or make logical deductions.[74] By the late 1980s and 1990s, AI research had developed methods for dealing with uncertain or incomplete information, employing concepts from probability and economics.[75]

These algorithms proved to be insufficient for solving large reasoning problems, because they experienced a “combinatorial explosion”: they became exponentially slower as the problems grew larger.[55] In fact, even humans rarely use the step-by-step deduction that early AI research was able to model. They solve most of their problems using fast, intuitive judgements.[76]

Knowledge representation[77] and knowledge engineering[78] are central to classical AI research. Some “expert systems” attempt to gather together explicit knowledge possessed by experts in some narrow domain. In addition, some projects attempt to gather the “commonsense knowledge” known to the average person into a database containing extensive knowledge about the world. Among the things a comprehensive commonsense knowledge base would contain are: objects, properties, categories and relations between objects;[79] situations, events, states and time;[80] causes and effects;[81] knowledge about knowledge (what we know about what other people know);[82] and many other, less well researched domains. A representation of “what exists” is an ontology: the set of objects, relations, concepts, and properties formally described so that software agents can interpret them. The semantics of these are captured as description logic concepts, roles, and individuals, and typically implemented as classes, properties, and individuals in the Web Ontology Language.[83] The most general ontologies are called upper ontologies, which attempt to provide a foundation for all other knowledge[84] by acting as mediators between domain ontologies that cover specific knowledge about a particular knowledge domain (field of interest or area of concern). Such formal knowledge representations can be used in content-based indexing and retrieval,[85] scene interpretation,[86] clinical decision support,[87] knowledge discovery (mining “interesting” and actionable inferences from large databases),[88] and other areas.[89]

Among the most difficult problems in knowledge representation are:

Intelligent agents must be able to set goals and achieve them.[96] They need a way to visualize the futurea representation of the state of the world and be able to make predictions about how their actions will change itand be able to make choices that maximize the utility (or “value”) of available choices.[97]

In classical planning problems, the agent can assume that it is the only system acting in the world, allowing the agent to be certain of the consequences of its actions.[98] However, if the agent is not the only actor, then it requires that the agent can reason under uncertainty. This calls for an agent that can not only assess its environment and make predictions, but also evaluate its predictions and adapt based on its assessment.[99]

Multi-agent planning uses the cooperation and competition of many agents to achieve a given goal. Emergent behavior such as this is used by evolutionary algorithms and swarm intelligence.[100]

Machine learning, a fundamental concept of AI research since the field’s inception,[101] is the study of computer algorithms that improve automatically through experience.[102][103]

Unsupervised learning is the ability to find patterns in a stream of input. Supervised learning includes both classification and numerical regression. Classification is used to determine what category something belongs in, after seeing a number of examples of things from several categories. Regression is the attempt to produce a function that describes the relationship between inputs and outputs and predicts how the outputs should change as the inputs change.[103] Both classifiers and regression learners can be viewed as “function approximators” trying to learn an unknown (possibly implicit) function; for example, a spam classifier can be viewed as learning a function that maps from the text of an email to one of two categories, “spam” or “not spam”. Computational learning theory can assess learners by computational complexity, by sample complexity (how much data is required), or by other notions of optimization.[104] In reinforcement learning[105] the agent is rewarded for good responses and punished for bad ones. The agent uses this sequence of rewards and punishments to form a strategy for operating in its problem space.

Natural language processing[106] (NLP) gives machines the ability to read and understand human language. A sufficiently powerful natural language processing system would enable natural-language user interfaces and the acquisition of knowledge directly from human-written sources, such as newswire texts. Some straightforward applications of natural language processing include information retrieval, text mining, question answering[107] and machine translation.[108] Many current approaches use word co-occurrence frequencies to construct syntactic representations of text. “Keyword spotting” strategies for search are popular and scalable but dumb; a search query for “dog” might only match documents with the literal word “dog” and miss a document with the word “poodle”. “Lexical affinity” strategies use the occurrence of words such as “accident” to assess the sentiment of a document. Modern statistical NLP approaches can combine all these strategies as well as others, and often achieve acceptable accuracy at the page or paragraph level, but continue to lack the semantic understanding required to classify isolated sentences well. Besides the usual difficulties with encoding semantic commonsense knowledge, existing semantic NLP sometimes scales too poorly to be viable in business applications. Beyond semantic NLP, the ultimate goal of “narrative” NLP is to embody a full understanding of commonsense reasoning.[109]

Machine perception[110] is the ability to use input from sensors (such as cameras (visible spectrum or infrared), microphones, wireless signals, and active lidar, sonar, radar, and tactile sensors) to deduce aspects of the world. Applications include speech recognition,[111] facial recognition, and object recognition.[112] Computer vision is the ability to analyze visual input. Such input is usually ambiguous; a giant, fifty-meter-tall pedestrian far away may produce exactly the same pixels as a nearby normal-sized pedestrian, requiring the AI to judge the relative likelihood and reasonableness of different interpretations, for example by using its “object model” to assess that fifty-meter pedestrians do not exist.[113]

AI is heavily used in robotics.[114] Advanced robotic arms and other industrial robots, widely used in modern factories, can learn from experience how to move efficiently despite the presence of friction and gear slippage.[115] A modern mobile robot, when given a small, static, and visible environment, can easily determine its location and map its environment; however, dynamic environments, such as (in endoscopy) the interior of a patient’s breathing body, pose a greater challenge. Motion planning is the process of breaking down a movement task into “primitives” such as individual joint movements. Such movement often involves compliant motion, a process where movement requires maintaining physical contact with an object.[117][118] Moravec’s paradox generalizes that low-level sensorimotor skills that humans take for granted are, counterintuitively, difficult to program into a robot; the paradox is named after Hans Moravec, who stated in 1988 that “it is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility”.[119][120] This is attributed to the fact that, unlike checkers, physical dexterity has been a direct target of natural selection for millions of years.[121]

Moravec’s paradox can be extended to many forms of social intelligence.[123][124] Distributed multi-agent coordination of autonomous vehicles remains a difficult problem.[125] Affective computing is an interdisciplinary umbrella that comprises systems which recognize, interpret, process, or simulate human affects. Moderate successes related to affective computing include textual sentiment analysis and, more recently, multimodal affect analysis (see multimodal sentiment analysis), wherein AI classifies the affects displayed by a videotaped subject.[129]

In the long run, social skills and an understanding of human emotion and game theory would be valuable to a social agent. Being able to predict the actions of others by understanding their motives and emotional states would allow an agent to make better decisions. Some computer systems mimic human emotion and expressions to appear more sensitive to the emotional dynamics of human interaction, or to otherwise facilitate humancomputer interaction.[130] Similarly, some virtual assistants are programmed to speak conversationally or even to banter humorously; this tends to give naive users an unrealistic conception of how intelligent existing computer agents actually are.[131]

Historically, projects such as the Cyc knowledge base (1984) and the massive Japanese Fifth Generation Computer Systems initiative (19821992) attempted to cover the breadth of human cognition. These early projects failed to escape the limitations of non-quantitative symbolic logic models and, in retrospect, greatly underestimated the difficulty of cross-domain AI. Nowadays, the vast majority of current AI researchers work instead on tractable “narrow AI” applications (such as medical diagnosis or automobile navigation).[132] Many researchers predict that such “narrow AI” work in different individual domains will eventually be incorporated into a machine with artificial general intelligence (AGI), combining most of the narrow skills mentioned in this article and at some point even exceeding human ability in most or all these areas.[17][133] Many advances have general, cross-domain significance. One high-profile example is that DeepMind in the 2010s developed a “generalized artificial intelligence” that could learn many diverse Atari games on its own, and later developed a variant of the system which succeeds at sequential learning.[134][135][136] Besides transfer learning,[137] hypothetical AGI breakthroughs could include the development of reflective architectures that can engage in decision-theoretic metareasoning, and figuring out how to “slurp up” a comprehensive knowledge base from the entire unstructured Web. Some argue that some kind of (currently-undiscovered) conceptually straightforward, but mathematically difficult, “Master Algorithm” could lead to AGI. Finally, a few “emergent” approaches look to simulating human intelligence extremely closely, and believe that anthropomorphic features like an artificial brain or simulated child development may someday reach a critical point where general intelligence emerges.[139][140]

Many of the problems in this article may also require general intelligence, if machines are to solve the problems as well as people do. For example, even specific straightforward tasks, like machine translation, require that a machine read and write in both languages (NLP), follow the author’s argument (reason), know what is being talked about (knowledge), and faithfully reproduce the author’s original intent (social intelligence). A problem like machine translation is considered “AI-complete”, because all of these problems need to be solved simultaneously in order to reach human-level machine performance.

There is no established unifying theory or paradigm that guides AI research. Researchers disagree about many issues.[141] A few of the most long standing questions that have remained unanswered are these: should artificial intelligence simulate natural intelligence by studying psychology or neurobiology? Or is human biology as irrelevant to AI research as bird biology is to aeronautical engineering?[14]Can intelligent behavior be described using simple, elegant principles (such as logic or optimization)? Or does it necessarily require solving a large number of completely unrelated problems?[15]

In the 1940s and 1950s, a number of researchers explored the connection between neurobiology, information theory, and cybernetics. Some of them built machines that used electronic networks to exhibit rudimentary intelligence, such as W. Grey Walter’s turtles and the Johns Hopkins Beast. Many of these researchers gathered for meetings of the Teleological Society at Princeton University and the Ratio Club in England.[142] By 1960, this approach was largely abandoned, although elements of it would be revived in the 1980s.

When access to digital computers became possible in the middle 1950s, AI research began to explore the possibility that human intelligence could be reduced to symbol manipulation. The research was centered in three institutions: Carnegie Mellon University, Stanford and MIT, and as described below, each one developed its own style of research. John Haugeland named these symbolic approaches to AI “good old fashioned AI” or “GOFAI”.[143] During the 1960s, symbolic approaches had achieved great success at simulating high-level thinking in small demonstration programs. Approaches based on cybernetics or artificial neural networks were abandoned or pushed into the background.[144]Researchers in the 1960s and the 1970s were convinced that symbolic approaches would eventually succeed in creating a machine with artificial general intelligence and considered this the goal of their field.

Economist Herbert Simon and Allen Newell studied human problem-solving skills and attempted to formalize them, and their work laid the foundations of the field of artificial intelligence, as well as cognitive science, operations research and management science. Their research team used the results of psychological experiments to develop programs that simulated the techniques that people used to solve problems. This tradition, centered at Carnegie Mellon University would eventually culminate in the development of the Soar architecture in the middle 1980s.[145][146]

Unlike Simon and Newell, John McCarthy felt that machines did not need to simulate human thought, but should instead try to find the essence of abstract reasoning and problem-solving, regardless of whether people used the same algorithms.[14] His laboratory at Stanford (SAIL) focused on using formal logic to solve a wide variety of problems, including knowledge representation, planning and learning.[147] Logic was also the focus of the work at the University of Edinburgh and elsewhere in Europe which led to the development of the programming language Prolog and the science of logic programming.[148]

Researchers at MIT (such as Marvin Minsky and Seymour Papert)[149] found that solving difficult problems in vision and natural language processing required ad-hoc solutions they argued that there was no simple and general principle (like logic) that would capture all the aspects of intelligent behavior. Roger Schank described their “anti-logic” approaches as “scruffy” (as opposed to the “neat” paradigms at CMU and Stanford).[15] Commonsense knowledge bases (such as Doug Lenat’s Cyc) are an example of “scruffy” AI, since they must be built by hand, one complicated concept at a time.[150]

When computers with large memories became available around 1970, researchers from all three traditions began to build knowledge into AI applications.[151] This “knowledge revolution” led to the development and deployment of expert systems (introduced by Edward Feigenbaum), the first truly successful form of AI software.[37] The knowledge revolution was also driven by the realization that enormous amounts of knowledge would be required by many simple AI applications.

By the 1980s progress in symbolic AI seemed to stall and many believed that symbolic systems would never be able to imitate all the processes of human cognition, especially perception, robotics, learning and pattern recognition. A number of researchers began to look into “sub-symbolic” approaches to specific AI problems.[16] Sub-symbolic methods manage to approach intelligence without specific representations of knowledge.

This includes embodied, situated, behavior-based, and nouvelle AI. Researchers from the related field of robotics, such as Rodney Brooks, rejected symbolic AI and focused on the basic engineering problems that would allow robots to move and survive.[152] Their work revived the non-symbolic viewpoint of the early cybernetics researchers of the 1950s and reintroduced the use of control theory in AI. This coincided with the development of the embodied mind thesis in the related field of cognitive science: the idea that aspects of the body (such as movement, perception and visualization) are required for higher intelligence.

Within developmental robotics, developmental learning approaches are elaborated upon to allow robots to accumulate repertoires of novel skills through autonomous self-exploration, social interaction with human teachers, and the use of guidance mechanisms (active learning, maturation, motor synergies, etc.).[153][154]

Interest in neural networks and “connectionism” was revived by David Rumelhart and others in the middle of the 1980s.[157] Artificial neural networks are an example of soft computing — they are solutions to problems which cannot be solved with complete logical certainty, and where an approximate solution is often sufficient. Other soft computing approaches to AI include fuzzy systems, evolutionary computation and many statistical tools. The application of soft computing to AI is studied collectively by the emerging discipline of computational intelligence.[158]

Much of traditional GOFAI got bogged down on ad hoc patches to symbolic computation that worked on their own toy models but failed to generalize to real-world results. However, around the 1990s, AI researchers adopted sophisticated mathematical tools, such as hidden Markov models (HMM), information theory, and normative Bayesian decision theory to compare or to unify competing architectures. The shared mathematical language permitted a high level of collaboration with more established fields (like mathematics, economics or operations research).[d] Compared with GOFAI, new “statistical learning” techniques such as HMM and neural networks were gaining higher levels of accuracy in many practical domains such as data mining, without necessarily acquiring semantic understanding of the datasets. The increased successes with real-world data led to increasing emphasis on comparing different approaches against shared test data to see which approach performed best in a broader context than that provided by idiosyncratic toy models; AI research was becoming more scientific. Nowadays results of experiments are often rigorously measurable, and are sometimes (with difficulty) reproducible.[38][159] Different statistical learning techniques have different limitations; for example, basic HMM cannot model the infinite possible combinations of natural language. Critics note that the shift from GOFAI to statistical learning is often also a shift away from Explainable AI. In AGI research, some scholars caution against over-reliance on statistical learning, and argue that continuing research into GOFAI will still be necessary to attain general intelligence.

AI has developed a large number of tools to solve the most difficult problems in computer science. A few of the most general of these methods are discussed below.

Many problems in AI can be solved in theory by intelligently searching through many possible solutions:[168] Reasoning can be reduced to performing a search. For example, logical proof can be viewed as searching for a path that leads from premises to conclusions, where each step is the application of an inference rule.[169] Planning algorithms search through trees of goals and subgoals, attempting to find a path to a target goal, a process called means-ends analysis.[170] Robotics algorithms for moving limbs and grasping objects use local searches in configuration space.[115] Many learning algorithms use search algorithms based on optimization.

Simple exhaustive searches[171] are rarely sufficient for most real-world problems: the search space (the number of places to search) quickly grows to astronomical numbers. The result is a search that is too slow or never completes. The solution, for many problems, is to use “heuristics” or “rules of thumb” that prioritize choices in favor of those that are more likely to reach a goal and to do so in a shorter number of steps. In some search methodologies heuristics can also serve to entirely eliminate some choices that are unlikely to lead to a goal (called “pruning the search tree”). Heuristics supply the program with a “best guess” for the path on which the solution lies.[172] Heuristics limit the search for solutions into a smaller sample size.

A very different kind of search came to prominence in the 1990s, based on the mathematical theory of optimization. For many problems, it is possible to begin the search with some form of a guess and then refine the guess incrementally until no more refinements can be made. These algorithms can be visualized as blind hill climbing: we begin the search at a random point on the landscape, and then, by jumps or steps, we keep moving our guess uphill, until we reach the top. Other optimization algorithms are simulated annealing, beam search and random optimization.[173]

Evolutionary computation uses a form of optimization search. For example, they may begin with a population of organisms (the guesses) and then allow them to mutate and recombine, selecting only the fittest to survive each generation (refining the guesses). Classic evolutionary algorithms include genetic algorithms, gene expression programming, and genetic programming.[174] Alternatively, distributed search processes can coordinate via swarm intelligence algorithms. Two popular swarm algorithms used in search are particle swarm optimization (inspired by bird flocking) and ant colony optimization (inspired by ant trails).[175][176]

Logic[177] is used for knowledge representation and problem solving, but it can be applied to other problems as well. For example, the satplan algorithm uses logic for planning[178] and inductive logic programming is a method for learning.[179]

Several different forms of logic are used in AI research. Propositional logic[180] involves truth functions such as “or” and “not”. First-order logic[181] adds quantifiers and predicates, and can express facts about objects, their properties, and their relations with each other. Fuzzy set theory assigns a “degree of truth” (between 0 and 1) to vague statements such as “Alice is old” (or rich, or tall, or hungry) that are too linguistically imprecise to be completely true or false. Fuzzy logic is successfully used in control systems to allow experts to contribute vague rules such as “if you are close to the destination station and moving fast, increase the train’s brake pressure”; these vague rules can then be numerically refined within the system. Fuzzy logic fails to scale well in knowledge bases; many AI researchers question the validity of chaining fuzzy-logic inferences.[e][183][184]

Default logics, non-monotonic logics and circumscription[91] are forms of logic designed to help with default reasoning and the qualification problem. Several extensions of logic have been designed to handle specific domains of knowledge, such as: description logics;[79] situation calculus, event calculus and fluent calculus (for representing events and time);[80] causal calculus;[81] belief calculus;[185] and modal logics.[82]

Overall, qualitiative symbolic logic is brittle and scales poorly in the presence of noise or other uncertainty. Exceptions to rules are numerous, and it is difficult for logical systems to function in the presence of contradictory rules.[187]

Many problems in AI (in reasoning, planning, learning, perception, and robotics) require the agent to operate with incomplete or uncertain information. AI researchers have devised a number of powerful tools to solve these problems using methods from probability theory and economics.[188]

Bayesian networks[189] are a very general tool that can be used for a large number of problems: reasoning (using the Bayesian inference algorithm),[190] learning (using the expectation-maximization algorithm),[f][192] planning (using decision networks)[193] and perception (using dynamic Bayesian networks).[194] Probabilistic algorithms can also be used for filtering, prediction, smoothing and finding explanations for streams of data, helping perception systems to analyze processes that occur over time (e.g., hidden Markov models or Kalman filters).[194] Compared with symbolic logic, formal Bayesian inference is computationally expensive. For inference to be tractable, most observations must be conditionally independent of one another. Complicated graphs with diamonds or other “loops” (undirected cycles) can require a sophisticated method such as Markov Chain Monte Carlo, which spreads an ensemble of random walkers throughout the Bayesian network and attempts to converge to an assessment of the conditional probabilities. Bayesian networks are used on XBox Live to rate and match players; wins and losses are “evidence” of how good a player is. AdSense uses a Bayesian network with over 300 million edges to learn which ads to serve.

A key concept from the science of economics is “utility”: a measure of how valuable something is to an intelligent agent. Precise mathematical tools have been developed that analyze how an agent can make choices and plan, using decision theory, decision analysis,[196] and information value theory.[97] These tools include models such as Markov decision processes,[197] dynamic decision networks,[194] game theory and mechanism design.[198]

The simplest AI applications can be divided into two types: classifiers (“if shiny then diamond”) and controllers (“if shiny then pick up”). Controllers do, however, also classify conditions before inferring actions, and therefore classification forms a central part of many AI systems. Classifiers are functions that use pattern matching to determine a closest match. They can be tuned according to examples, making them very attractive for use in AI. These examples are known as observations or patterns. In supervised learning, each pattern belongs to a certain predefined class. A class can be seen as a decision that has to be made. All the observations combined with their class labels are known as a data set. When a new observation is received, that observation is classified based on previous experience.[199]

A classifier can be trained in various ways; there are many statistical and machine learning approaches. The decision tree[200] is perhaps the most widely used machine learning algorithm. Other widely used classifiers are the neural network,[202]k-nearest neighbor algorithm,[g][204]kernel methods such as the support vector machine (SVM),[h][206]Gaussian mixture model,[207] and the extremely popular naive Bayes classifier.[i][209] Classifier performance depends greatly on the characteristics of the data to be classified, such as the dataset size, the dimensionality, and the level of noise. Model-based classifiers perform well if the assumed model is an extremely good fit for the actual data. Otherwise, if no matching model is available, and if accuracy (rather than speed or scalability) is the sole concern, conventional wisdom is that discriminative classifiers (especially SVM) tend to be more accurate than model-based classifiers such as “naive Bayes” on most practical data sets.[210]

Neural networks, or neural nets, were inspired by the architecture of neurons in the human brain. A simple “neuron” N accepts input from multiple other neurons, each of which, when activated (or “fired”), cast a weighted “vote” for or against whether neuron N should itself activate. Learning requires an algorithm to adjust these weights based on the training data; one simple algorithm (dubbed “fire together, wire together”) is to increase the weight between two connected neurons when the activation of one triggers the successful activation of another. The net forms “concepts” that are distributed among a subnetwork of shared[j] neurons that tend to fire together; a concept meaning “leg” might be coupled with a subnetwork meaning “foot” that includes the sound for “foot”. Neurons have a continuous spectrum of activation; in addition, neurons can process inputs in a nonlinear way rather than weighing straightforward votes. Modern neural nets can learn both continuous functions and, surprisingly, digital logical operations. Neural networks’ early successes included predicting the stock market and (in 1995) a mostly self-driving car.[k] In the 2010s, advances in neural networks using deep learning thrust AI into widespread public consciousness and contributed to an enormous upshift in corporate AI spending; for example, AI-related M&A in 2017 was over 25 times as large as in 2015.[213][214]

The study of non-learning artificial neural networks[202] began in the decade before the field of AI research was founded, in the work of Walter Pitts and Warren McCullouch. Frank Rosenblatt invented the perceptron, a learning network with a single layer, similar to the old concept of linear regression. Early pioneers also include Alexey Grigorevich Ivakhnenko, Teuvo Kohonen, Stephen Grossberg, Kunihiko Fukushima, Christoph von der Malsburg, David Willshaw, Shun-Ichi Amari, Bernard Widrow, John Hopfield, Eduardo R. Caianiello, and others.

The main categories of networks are acyclic or feedforward neural networks (where the signal passes in only one direction) and recurrent neural networks (which allow feedback and short-term memories of previous input events). Among the most popular feedforward networks are perceptrons, multi-layer perceptrons and radial basis networks.[215] Neural networks can be applied to the problem of intelligent control (for robotics) or learning, using such techniques as Hebbian learning (“fire together, wire together”), GMDH or competitive learning.[216]

Today, neural networks are often trained by the backpropagation algorithm, which had been around since 1970 as the reverse mode of automatic differentiation published by Seppo Linnainmaa,[217][218] and was introduced to neural networks by Paul Werbos.[219][220][221]

Hierarchical temporal memory is an approach that models some of the structural and algorithmic properties of the neocortex.[222]

In short, most neural networks use some form of gradient descent on a hand-created neural topology. However, some research groups, such as Uber, argue that simple neuroevolution to mutate new neural network topologies and weights may be competitive with sophisticated gradient descent approaches. One advantage of neuroevolution is that it may be less prone to get caught in “dead ends”.[223]

Deep learning is any artificial neural network that can learn a long chain of causal links. For example, a feedforward network with six hidden layers can learn a seven-link causal chain (six hidden layers + output layer) and has a “credit assignment path” (CAP) depth of seven. Many deep learning systems need to be able to learn chains ten or more causal links in length.[224] Deep learning has transformed many important subfields of artificial intelligence, including computer vision, speech recognition, natural language processing and others.[225][226][224]

According to one overview,[227] the expression “Deep Learning” was introduced to the Machine Learning community by Rina Dechter in 1986[228] and gained traction afterIgor Aizenberg and colleagues introduced it to Artificial Neural Networks in 2000.[229] The first functional Deep Learning networks were published by Alexey Grigorevich Ivakhnenko and V. G. Lapa in 1965.[230][pageneeded] These networks are trained one layer at a time. Ivakhnenko’s 1971 paper[231] describes the learning of a deep feedforward multilayer perceptron with eight layers, already much deeper than many later networks. In 2006, a publication by Geoffrey Hinton and Ruslan Salakhutdinov introduced another way of pre-training many-layered feedforward neural networks (FNNs) one layer at a time, treating each layer in turn as an unsupervised restricted Boltzmann machine, then using supervised backpropagation for fine-tuning. Similar to shallow artificial neural networks, deep neural networks can model complex non-linear relationships. Over the last few years, advances in both machine learning algorithms and computer hardware have led to more efficient methods for training deep neural networks that contain many layers of non-linear hidden units and a very large output layer.[233]

Deep learning often uses convolutional neural networks (CNNs), whose origins can be traced back to the Neocognitron introduced by Kunihiko Fukushima in 1980.[234] In 1989, Yann LeCun and colleagues applied backpropagation to such an architecture. In the early 2000s, in an industrial application CNNs already processed an estimated 10% to 20% of all the checks written in the US.[235]Since 2011, fast implementations of CNNs on GPUs havewon many visual pattern recognition competitions.[224]

CNNs with 12 convolutional layers were used in conjunction with reinforcement learning by Deepmind’s “AlphaGo Lee”, the program that beat a top Go champion in 2016.[236]

Early on, deep learning was also applied to sequence learning with recurrent neural networks (RNNs)[237] which are in theory Turing complete[238] and can run arbitrary programs to process arbitrary sequences of inputs. The depth of an RNN is unlimited and depends on the length of its input sequence; thus, an RNN is an example of deep learning.[224] RNNs can be trained by gradient descent[239][240][241] but suffer from the vanishing gradient problem.[225][242] In 1992, it was shown that unsupervised pre-training of a stack of recurrent neural networks can speed up subsequent supervised learning of deep sequential problems.[243]

Numerous researchers now use variants of a deep learning recurrent NN called the long short-term memory (LSTM) network published by Hochreiter & Schmidhuber in 1997.[244] LSTM is often trained by Connectionist Temporal Classification (CTC).[245] At Google, Microsoft and Baidu this approach has revolutionised speech recognition.[246][247][248] For example, in 2015, Google’s speech recognition experienced a dramatic performance jump of 49% through CTC-trained LSTM, which is now available through Google Voice to billions of smartphone users.[249] Google also used LSTM to improve machine translation,[250] Language Modeling[251] and Multilingual Language Processing.[252] LSTM combined with CNNs also improved automatic image captioning[253] and a plethora of other applications.

AI, like electricity or the steam engine, is a general purpose technology. There is no consensus on how to characterize which tasks AI tends to excel at.[254] While projects such as AlphaZero have succeeded in generating their own knowledge from scratch, many other machine learning projects require large training datasets.[255][256] Researcher Andrew Ng has suggested, as a “highly imperfect rule of thumb”, that “almost anything a typical human can do with less than one second of mental thought, we can probably now or in the near future automate using AI.”[257] Moravec’s paradox suggests that AI lags humans at many tasks that the human brain has specifically evolved to perform well.[121]

Games provide a well-publicized benchmark for assessing rates of progress. AlphaGo around 2016 brought the era of classical board-game benchmarks to a close. Games of imperfect knowledge provide new challenges to AI in the area of game theory.[258][259] E-sports such as StarCraft continue to provide additional public benchmarks.[260][261] There are many competitions and prizes, such as the Imagenet Challenge, to promote research in artificial intelligence. The main areas of competition include general machine intelligence, conversational behavior, data-mining, robotic cars, and robot soccer as well as conventional games.[citation needed]

The “imitation game” (an interpretation of the 1950 Turing test that assesses whether a computer can imitate a human) is nowadays considered too exploitable to be a meaningful benchmark.[262] A derivative of the Turing test is the Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA). As the name implies, this helps to determine that a user is an actual person and not a computer posing as a human. In contrast to the standard Turing test, CAPTCHA is administered by a machine and targeted to a human as opposed to being administered by a human and targeted to a machine. A computer asks a user to complete a simple test then generates a grade for that test. Computers are unable to solve the problem, so correct solutions are deemed to be the result of a person taking the test. A common type of CAPTCHA is the test that requires the typing of distorted letters, numbers or symbols that appear in an image undecipherable by a computer.

Proposed “universal intelligence” tests aim to compare how well machines, humans, and even non-human animals perform on problem sets that are generic as possible. At an extreme, the test suite can contain every possible problem, weighted by Kolmogorov complexity; unfortunately, these problem sets tend to be dominated by impoverished pattern-matching exercises where a tuned AI can easily exceed human performance levels.[264][265]

AI is relevant to any intellectual task. Modern artificial intelligence techniques are pervasive and are too numerous to list here. Frequently, when a technique reaches mainstream use, it is no longer considered artificial intelligence; this phenomenon is described as the AI effect.

High-profile examples of AI include autonomous vehicles (such as drones and self-driving cars), medical diagnosis, creating art (such as poetry), proving mathematical theorems, playing games (such as Chess or Go), search engines (such as Google search), online assistants (such as Siri), image recognition in photographs, spam filtering, prediction of judicial decisions[268] and targeting online advertisements.[269][270]

With social media sites overtaking TV as a source for news for young people and news organisations increasingly reliant on social media platforms for generating distribution,[271] major publishers now use artificial intelligence (AI) technology to post stories more effectively and generate higher volumes of traffic.[272]

Artificial intelligence is breaking into the healthcare industry by assisting doctors. According to Bloomberg Technology, Microsoft has developed AI to help doctors find the right treatments for cancer.[273] There is a great amount of research and drugs developed relating to cancer. In detail, there are more than 800 medicines and vaccines to treat cancer. This negatively affects the doctors, because there are too many options to choose from, making it more difficult to choose the right drugs for the patients. Microsoft is working on a project to develop a machine called “Hanover”. Its goal is to memorize all the papers necessary to cancer and help predict which combinations of drugs will be most effective for each patient. One project that is being worked on at the moment is fighting myeloid leukemia, a fatal cancer where the treatment has not improved in decades. Another study was reported to have found that artificial intelligence was as good as trained doctors in identifying skin cancers.[274] Another study is using artificial intelligence to try and monitor multiple high-risk patients, and this is done by asking each patient numerous questions based on data acquired from live doctor to patient interactions.[275]

According to CNN, a recent study by surgeons at the Children’s National Medical Center in Washington successfully demonstrated surgery with an autonomous robot. The team supervised the robot while it performed soft-tissue surgery, stitching together a pig’s bowel during open surgery, and doing so better than a human surgeon, the team claimed.[276] IBM has created its own artificial intelligence computer, the IBM Watson, which has beaten human intelligence (at some levels). Watson not only won at the game show Jeopardy! against former champions,[277] but was declared a hero after successfully diagnosing a woman who was suffering from leukemia.[278]

Advancements in AI have contributed to the growth of the automotive industry through the creation and evolution of self-driving vehicles. As of 2016, there are over 30 companies utilizing AI into the creation of driverless cars. A few companies involved with AI include Tesla, Google, and Apple.[279]

Many components contribute to the functioning of self-driving cars. These vehicles incorporate systems such as braking, lane changing, collision prevention, navigation and mapping. Together, these systems, as well as high performance computers, are integrated into one complex vehicle.[280]

Recent developments in autonomous automobiles have made the innovation of self-driving trucks possible, though they are still in the testing phase. The UK government has passed legislation to begin testing of self-driving truck platoons in 2018.[281] Self-driving truck platoons are a fleet of self-driving trucks following the lead of one non-self-driving truck, so the truck platoons aren’t entirely autonomous yet. Meanwhile, the Daimler, a German automobile corporation, is testing the Freightliner Inspiration which is a semi-autonomous truck that will only be used on the highway.[282]

One main factor that influences the ability for a driver-less automobile to function is mapping. In general, the vehicle would be pre-programmed with a map of the area being driven. This map would include data on the approximations of street light and curb heights in order for the vehicle to be aware of its surroundings. However, Google has been working on an algorithm with the purpose of eliminating the need for pre-programmed maps and instead, creating a device that would be able to adjust to a variety of new surroundings.[283] Some self-driving cars are not equipped with steering wheels or brake pedals, so there has also been research focused on creating an algorithm that is capable of maintaining a safe environment for the passengers in the vehicle through awareness of speed and driving conditions.[284]

Another factor that is influencing the ability for a driver-less automobile is the safety of the passenger. To make a driver-less automobile, engineers must program it to handle high-risk situations. These situations could include a head-on collision with pedestrians. The car’s main goal should be to make a decision that would avoid hitting the pedestrians and saving the passengers in the car. But there is a possibility the car would need to make a decision that would put someone in danger. In other words, the car would need to decide to save the pedestrians or the passengers.[285] The programing of the car in these situations is crucial to a successful driver-less automobile.

Financial institutions have long used artificial neural network systems to detect charges or claims outside of the norm, flagging these for human investigation. The use of AI in banking can be traced back to 1987 when Security Pacific National Bank in US set-up a Fraud Prevention Task force to counter the unauthorised use of debit cards. Programs like Kasisto and Moneystream are using AI in financial services.

Banks use artificial intelligence systems today to organize operations, maintain book-keeping, invest in stocks, and manage properties. AI can react to changes overnight or when business is not taking place.[286] In August 2001, robots beat humans in a simulated financial trading competition.[287] AI has also reduced fraud and financial crimes by monitoring behavioral patterns of users for any abnormal changes or anomalies.[288]

The use of AI machines in the market in applications such as online trading and decision making has changed major economic theories.[289] For example, AI based buying and selling platforms have changed the law of supply and demand in that it is now possible to easily estimate individualized demand and supply curves and thus individualized pricing. Furthermore, AI machines reduce information asymmetry in the market and thus making markets more efficient while reducing the volume of trades. Furthermore, AI in the markets limits the consequences of behavior in the markets again making markets more efficient. Other theories where AI has had impact include in rational choice, rational expectations, game theory, Lewis turning point, portfolio optimization and counterfactual thinking.

In video games, artificial intelligence is routinely used to generate dynamic purposeful behavior in non-player characters (NPCs). In addition, well-understood AI techniques are routinely used for pathfinding. Some researchers consider NPC AI in games to be a “solved problem” for most production tasks. Games with more atypical AI include the AI director of Left 4 Dead (2008) and the neuroevolutionary training of platoons in Supreme Commander 2 (2010).[290][291]

Worldwide annual military spending on robotics rose from 5.1 billion USD in 2010 to 7.5 billion USD in 2015.[292][293] Military drones capable of autonomous action are widely considered a useful asset. In 2017, Vladimir Putin stated that “Whoever becomes the leader in (artificial intelligence) will become the ruler of the world”.[294][295] Many artificial intelligence researchers seek to distance themselves from military applications of AI.[296]

For financial statements audit, AI makes continuous audit possible. AI tools could analyze many sets of different information immediately. The potential benefit would be the overall audit risk will be reduced, the level of assurance will be increased and the time duration of audit will be reduced.[297]

A report by the Guardian newspaper in the UK in 2018 found that online gambling companies were using AI to predict the behavior of customers in order to target them with personalized promotions.[298] Developers of commercial AI platforms are also beginning to appeal more directly to casino operators, offering a range of existing and potential services to help them boost their profits and expand their customer base.[299]

Artificial Intelligence has inspired numerous creative applications including its usage to produce visual art. The exhibition “Thinking Machines: Art and Design in the Computer Age, 1959-1989” at MoMA [300] provides a good overview of the historical applications of AI for art, architecture, and design. Recent exhibitions showcasing the usage of AI to produce art include the Google-sponsored benefit and auction at the Gray Area Foundation in San Francisco, where artists experimented with the deepdream algorithm [301] and the exhibition “Unhuman: Art in the Age of AI,” which took place in Los Angeles and Frankfurt in the fall of 2017. [302][303] In the spring of 2018, the Association of Computing Machinery dedicated a special magazine issue to the subject of computers and art highlighting the role of machine learning in the arts.[304]

There are three philosophical questions related to AI:

Can a machine be intelligent? Can it “think”?

Follow this link:

Artificial intelligence – Wikipedia

Artificial Intelligence (3rd Edition): Winston: 9780201533774 …

This book explains how it is possible for computers to reason and perceive, thus introducing the field called artificial intelligence. From the book, you learn why the field is important, both as a branch of engineering and as a science.

If you are a computer scientist or an engineer, you will enjoy the book, because it provides a cornucopia of new ideas for representing knowledge, using knowledge, and building practical systems. If you are a psychologist, biologist, linguist, or philosopher, you will enjoy the book because it provides an exciting computational perspective on the mystery of intelligence.

This completely rewritten and updated edition of Artificial Intelligence reflects the revolutionary progress made since the previous edition was published.

Part I is about representing knowledge and about reasoning methods that make use of knowledge. The material covered includes the semantic-net family of representations, describe and match, generate and test, means-ends analysis, problem reduction, basic search, optimal search, adversarial search, rule chaining, the rete algorithm, frame inheritance, topological sorting, constraint propagation, logic, truth maintenance, planning, and cognitive modeling.

Part II is about learning, the sine qua non of intelligence. Some methods involve much reasoning; others just extract regularity from data. The material covered includes near-miss analysis, explanation-based learning, knowledge repair, case recording, version-space convergence, identification-tree construction, neural-net training, perceptron convergence, approximation-net construction, and simulated evolution.

Part III is about visual perception and language understanding. You learn not only about perception and language, but also about ideas that have been a major source of inspiration for people working in other subfields of artificial intelligence. The material covered includes object identification, stereo vision, shape from shading, a glimpse of modern linguistic theory, and transition-tree methods for building practical natural-language interfaces.

0201533774B04062001

See the rest here:

Artificial Intelligence (3rd Edition): Winston: 9780201533774 …

A.I. Artificial Intelligence (2001) – IMDb

Edit Storyline

In the not-so-far future the polar ice caps have melted and the resulting rise of the ocean waters has drowned all the coastal cities of the world. Withdrawn to the interior of the continents, the human race keeps advancing, reaching the point of creating realistic robots (called mechas) to serve them. One of the mecha-producing companies builds David, an artificial kid which is the first to have real feelings, especially a never-ending love for his “mother”, Monica. Monica is the woman who adopted him as a substitute for her real son, who remains in cryo-stasis, stricken by an incurable disease. David is living happily with Monica and her husband, but when their real son returns home after a cure is discovered, his life changes dramatically. Written byChris Makrozahopoulos

Budget:$100,000,000 (estimated)

Opening Weekend USA: $29,352,630,1 July 2001, Wide Release

Gross USA: $78,616,689, 23 September 2001

Cumulative Worldwide Gross: $235,927,000

Runtime: 146 min

Aspect Ratio: 1.85 : 1

Visit link:

A.I. Artificial Intelligence (2001) – IMDb

Artificial Intelligence research at Microsoft

At Microsoft, researchers in artificial intelligence are harnessing the explosion of digital data and computational power with advanced algorithms to enable collaborative and natural interactions between people and machines that extend the human ability to sense, learn and understand. The research infuses computers, materials and systems with the ability to reason, communicate and perform with humanlike skill and agility.

Microsofts deep investments in the field are advancing the state of the art in machine intelligence and perception, enabling computers that understand what they see, communicate in natural language, answer complex questions and interact with their environment. In addition, the companys researchers are thought leaders on the ethics and societal impacts of intelligent technologies. The research, tools and services that result from this investment are woven into existing and new products and, at the same time, made open and accessible to the broader community in a bid to accelerate innovation, democratize AI and solve the worlds most pressing challenges.

View original post here:

Artificial Intelligence research at Microsoft

What is Artificial Intelligence (AI)? – Definition from …

Artificial intelligence is a branch of computer science that aims to create intelligent machines. It has become an essential part of the technology industry.

Research associated with artificial intelligence is highly technical and specialized. The core problems of artificial intelligence include programming computers for certain traits such as:

Knowledge engineering is a core part of AI research. Machines can often act and react like humans only if they have abundant information relating to the world. Artificial intelligence must have access to objects, categories, properties and relations between all of them to implement knowledge engineering. Initiating common sense, reasoning and problem-solving power in machines is a difficult and tedious task.

Machine learning is also a core part of AI. Learning without any kind of supervision requires an ability to identify patterns in streams of inputs, whereas learning with adequate supervision involves classification and numerical regressions. Classification determines the category an object belongs to and regression deals with obtaining a set of numerical input or output examples, thereby discovering functions enabling the generation of suitable outputs from respective inputs. Mathematical analysis of machine learning algorithms and their performance is a well-defined branch of theoretical computer science often referred to as computational learning theory.

Machine perception deals with the capability to use sensory inputs to deduce the different aspects of the world, while computer vision is the power to analyze visual inputs with a few sub-problems such as facial, object and gesture recognition.

Robotics is also a major field related to AI. Robots require intelligence to handle tasks such as object manipulation and navigation, along with sub-problems of localization, motion planning and mapping.

Read the original post:

What is Artificial Intelligence (AI)? – Definition from …

What is AI (artificial intelligence)? – Definition from …

AI (artificial intelligence) is the simulation of human intelligence processes by machines, especially computer systems. These processes include learning (the acquisition of information and rules for using the information), reasoning (using the rules to reach approximate or definite conclusions) and self-correction. Particular applications of AI include expert systems, speech recognition and machine vision.

AI was coined by John McCarthy, an American computer scientist, in 1956 at The Dartmouth Conference where the discipline was born. Today, it is an umbrella term that encompasses everything from robotic process automation to actual robotics. It has gained prominence recently due, in part, to big data, or the increase in speed, size and variety of data businesses are now collecting. AI can perform tasks such as identifying patterns in the data more efficiently than humans, enabling businesses to gain more insight out of their data.

AI can be categorized in any number of ways, but here are two examples.

The first classifies AI systems as either weak AI or strong AI. Weak AI, also known as narrow AI, is an AI system that is designed and trained for a particular task. Virtual personal assistants, such as Apple’s Siri, are a form of weak AI.

Strong AI, also known as artificial general intelligence, is an AI system with generalized human cognitive abilities so that when presented with an unfamiliar task, it has enough intelligence to find a solution. The Turing Test, developed by mathematician Alan Turing in 1950, is a method used to determine if a computer can actually think like a human, although the method is controversial.

The second example is from Arend Hintze, an assistant professor of integrative biology and computer science and engineering at Michigan State University. He categorizes AI into four types, from the kind of AI systems that exist today to sentient systems, which do not yet exist. His categories are as follows:

Here is the original post:

What is AI (artificial intelligence)? – Definition from …

Benefits & Risks of Artificial Intelligence – Future of Life …

Many AI researchers roll their eyes when seeing this headline:Stephen Hawking warns that rise of robots may be disastrous for mankind. And as many havelost count of how many similar articles theyveseen.Typically, these articles are accompanied by an evil-looking robot carrying a weapon, and they suggest we should worry about robots rising up and killing us because theyve become conscious and/or evil.On a lighter note, such articles are actually rather impressive, because they succinctly summarize the scenario that AI researchers dontworry about. That scenario combines as many as three separate misconceptions: concern about consciousness, evil, androbots.

If you drive down the road, you have a subjective experience of colors, sounds, etc. But does a self-driving car have a subjective experience? Does it feel like anything at all to be a self-driving car?Although this mystery of consciousness is interesting in its own right, its irrelevant to AI risk. If you get struck by a driverless car, it makes no difference to you whether it subjectively feels conscious. In the same way, what will affect us humans is what superintelligent AIdoes, not how it subjectively feels.

The fear of machines turning evil is another red herring. The real worry isnt malevolence, but competence. A superintelligent AI is by definition very good at attaining its goals, whatever they may be, so we need to ensure that its goals are aligned with ours. Humans dont generally hate ants, but were more intelligent than they are so if we want to build a hydroelectric dam and theres an anthill there, too bad for the ants. The beneficial-AI movement wants to avoid placing humanity in the position of those ants.

The consciousness misconception is related to the myth that machines cant have goals.Machines can obviously have goals in the narrow sense of exhibiting goal-oriented behavior: the behavior of a heat-seeking missile is most economically explained as a goal to hit a target.If you feel threatened by a machine whose goals are misaligned with yours, then it is precisely its goals in this narrow sense that troubles you, not whether the machine is conscious and experiences a sense of purpose.If that heat-seeking missile were chasing you, you probably wouldnt exclaim: Im not worried, because machines cant have goals!

I sympathize with Rodney Brooks and other robotics pioneers who feel unfairly demonized by scaremongering tabloids,because some journalists seem obsessively fixated on robots and adorn many of their articles with evil-looking metal monsters with red shiny eyes. In fact, the main concern of the beneficial-AI movement isnt with robots but with intelligence itself: specifically, intelligence whose goals are misaligned with ours. To cause us trouble, such misaligned superhuman intelligence needs no robotic body, merely an internet connection this may enable outsmarting financial markets, out-inventing human researchers, out-manipulating human leaders, and developing weapons we cannot even understand. Even if building robots were physically impossible, a super-intelligent and super-wealthy AI could easily pay or manipulate many humans to unwittingly do its bidding.

The robot misconception is related to the myth that machines cant control humans. Intelligence enables control: humans control tigers not because we are stronger, but because we are smarter. This means that if we cede our position as smartest on our planet, its possible that we might also cede control.

View original post here:

Benefits & Risks of Artificial Intelligence – Future of Life …

Artificial intelligence – Wikipedia

Intelligence demonstrated by machines

Artificial intelligence (AI), sometimes called machine intelligence, is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans and other animals. In computer science AI research is defined as the study of “intelligent agents”: any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] Colloquially, the term “artificial intelligence” is applied when a machine mimics “cognitive” functions that humans associate with other human minds, such as “learning” and “problem solving”.

The scope of AI is disputed: as machines become increasingly capable, tasks considered as requiring “intelligence” are often removed from the definition, a phenomenon known as the AI effect, leading to the quip, “AI is whatever hasn’t been done yet.”[3] For instance, optical character recognition is frequently excluded from “artificial intelligence”, having become a routine technology.[4] Capabilities generally classified as AI as of 2017[update] include successfully understanding human speech, competing at the highest level in strategic game systems (such as chess and Go),[6] autonomous cars, intelligent routing in content delivery network and military simulations.

Artificial intelligence was founded as an academic discipline in 1956, and in the years since has experienced several waves of optimism,[7][8] followed by disappointment and the loss of funding (known as an “AI winter”),[9][10] followed by new approaches, success and renewed funding.[8][11] For most of its history, AI research has been divided into subfields that often fail to communicate with each other.[12] These sub-fields are based on technical considerations, such as particular goals (e.g. “robotics” or “machine learning”),[13] the use of particular tools (“logic” or artificial neural networks), or deep philosophical differences.[14][15][16] Subfields have also been based on social factors (particular institutions or the work of particular researchers).[12]

The traditional problems (or goals) of AI research include reasoning, knowledge representation, planning, learning, natural language processing, perception and the ability to move and manipulate objects.[13] General intelligence is among the field’s long-term goals.[17] Approaches include statistical methods, computational intelligence, and traditional symbolic AI. Many tools are used in AI, including versions of search and mathematical optimization, artificial neural networks, and methods based on statistics, probability and economics. The AI field draws upon computer science, mathematics, psychology, linguistics, philosophy and many others.

The field was founded on the claim that human intelligence “can be so precisely described that a machine can be made to simulate it”.[18] This raises philosophical arguments about the nature of the mind and the ethics of creating artificial beings endowed with human-like intelligence which are issues that have been explored by myth, fiction and philosophy since antiquity.[19] Some people also consider AI to be a danger to humanity if it progresses unabatedly.[20] Others believe that AI, unlike previous technological revolutions, will create a risk of mass unemployment.[21]

In the twenty-first century, AI techniques have experienced a resurgence following concurrent advances in computer power, large amounts of data, and theoretical understanding; and AI techniques have become an essential part of the technology industry, helping to solve many challenging problems in computer science.[22][11]

Thought-capable artificial beings appeared as storytelling devices in antiquity,[23] and have been common in fiction, as in Mary Shelley’s Frankenstein or Karel apek’s R.U.R. (Rossum’s Universal Robots).[24] These characters and their fates raised many of the same issues now discussed in the ethics of artificial intelligence.[19]

The study of mechanical or “formal” reasoning began with philosophers and mathematicians in antiquity. The study of mathematical logic led directly to Alan Turing’s theory of computation, which suggested that a machine, by shuffling symbols as simple as “0” and “1”, could simulate any conceivable act of mathematical deduction. This insight, that digital computers can simulate any process of formal reasoning, is known as the ChurchTuring thesis.[25] Along with concurrent discoveries in neurobiology, information theory and cybernetics, this led researchers to consider the possibility of building an electronic brain. Turing proposed that “if a human could not distinguish between responses from a machine and a human, the machine could be considered intelligent”.[26] The first work that is now generally recognized as AI was McCullouch and Pitts’ 1943 formal design for Turing-complete “artificial neurons”.

The field of AI research was born at a workshop at Dartmouth College in 1956.[28] Attendees Allen Newell (CMU), Herbert Simon (CMU), John McCarthy (MIT), Marvin Minsky (MIT) and Arthur Samuel (IBM) became the founders and leaders of AI research.[29] They and their students produced programs that the press described as “astonishing”: computers were learning checkers strategies (c. 1954)[31] (and by 1959 were reportedly playing better than the average human),[32] solving word problems in algebra, proving logical theorems (Logic Theorist, first run c. 1956) and speaking English.[33] By the middle of the 1960s, research in the U.S. was heavily funded by the Department of Defense[34] and laboratories had been established around the world.[35] AI’s founders were optimistic about the future: Herbert Simon predicted, “machines will be capable, within twenty years, of doing any work a man can do”. Marvin Minsky agreed, writing, “within a generation… the problem of creating ‘artificial intelligence’ will substantially be solved”.[7]

They failed to recognize the difficulty of some of the remaining tasks. Progress slowed and in 1974, in response to the criticism of Sir James Lighthill and ongoing pressure from the US Congress to fund more productive projects, both the U.S. and British governments cut off exploratory research in AI. The next few years would later be called an “AI winter”,[9] a period when obtaining funding for AI projects was difficult.

In the early 1980s, AI research was revived by the commercial success of expert systems,[37] a form of AI program that simulated the knowledge and analytical skills of human experts. By 1985 the market for AI had reached over a billion dollars. At the same time, Japan’s fifth generation computer project inspired the U.S and British governments to restore funding for academic research.[8] However, beginning with the collapse of the Lisp Machine market in 1987, AI once again fell into disrepute, and a second, longer-lasting hiatus began.[10]

In the late 1990s and early 21st century, AI began to be used for logistics, data mining, medical diagnosis and other areas.[22] The success was due to increasing computational power (see Moore’s law), greater emphasis on solving specific problems, new ties between AI and other fields (such as statistics, economics and mathematics), and a commitment by researchers to mathematical methods and scientific standards.[38] Deep Blue became the first computer chess-playing system to beat a reigning world chess champion, Garry Kasparov on 11 May 1997.

In 2011, a Jeopardy! quiz show exhibition match, IBM’s question answering system, Watson, defeated the two greatest Jeopardy! champions, Brad Rutter and Ken Jennings, by a significant margin. Faster computers, algorithmic improvements, and access to large amounts of data enabled advances in machine learning and perception; data-hungry deep learning methods started to dominate accuracy benchmarks around 2012.[41] The Kinect, which provides a 3D bodymotion interface for the Xbox 360 and the Xbox One use algorithms that emerged from lengthy AI research[42] as do intelligent personal assistants in smartphones.[43] In March 2016, AlphaGo won 4 out of 5 games of Go in a match with Go champion Lee Sedol, becoming the first computer Go-playing system to beat a professional Go player without handicaps.[6][44] In the 2017 Future of Go Summit, AlphaGo won a three-game match with Ke Jie,[45] who at the time continuously held the world No. 1 ranking for two years.[46][47] This marked the completion of a significant milestone in the development of Artificial Intelligence as Go is an extremely complex game, more so than Chess.

According to Bloomberg’s Jack Clark, 2015 was a landmark year for artificial intelligence, with the number of software projects that use AI within Google increased from a “sporadic usage” in 2012 to more than 2,700 projects. Clark also presents factual data indicating that error rates in image processing tasks have fallen significantly since 2011.[48] He attributes this to an increase in affordable neural networks, due to a rise in cloud computing infrastructure and to an increase in research tools and datasets.[11] Other cited examples include Microsoft’s development of a Skype system that can automatically translate from one language to another and Facebook’s system that can describe images to blind people.[48] In a 2017 survey, one in five companies reported they had “incorporated AI in some offerings or processes”.[49][50]

A typical AI perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] An AI’s intended goal function can be simple (“1 if the AI wins a game of Go, 0 otherwise”) or complex (“Do actions mathematically similar to the actions that got you rewards in the past”). Goals can be explicitly defined, or can be induced. If the AI is programmed for “reinforcement learning”, goals can be implicitly induced by rewarding some types of behavior and punishing others.[a] Alternatively, an evolutionary system can induce goals by using a “fitness function” to mutate and preferentially replicate high-scoring AI systems; this is similar to how animals evolved to innately desire certain goals such as finding food, or how dogs can be bred via artificial selection to possess desired traits. Some AI systems, such as nearest-neighbor, instead reason by analogy; these systems are not generally given goals, except to the degree that goals are somehow implicit in their training data. Such systems can still be benchmarked if the non-goal system is framed as a system whose “goal” is to successfully accomplish its narrow classification task.[53]

AI often revolves around the use of algorithms. An algorithm is a set of unambiguous instructions that a mechanical computer can execute.[b] A complex algorithm is often built on top of other, simpler, algorithms. A simple example of an algorithm is the following recipe for optimal play at tic-tac-toe:

Many AI algorithms are capable of learning from data; they can enhance themselves by learning new heuristics (strategies, or “rules of thumb”, that have worked well in the past), or can themselves write other algorithms. Some of the “learners” described below, including Bayesian networks, decision trees, and nearest-neighbor, could theoretically, if given infinite data, time, and memory, learn to approximate any function, including whatever combination of mathematical functions would best describe the entire world. These learners could therefore, in theory, derive all possible knowledge, by considering every possible hypothesis and matching it against the data. In practice, it is almost never possible to consider every possibility, because of the phenomenon of “combinatorial explosion”, where the amount of time needed to solve a problem grows exponentially. Much of AI research involves figuring out how to identify and avoid considering broad swaths of possibilities that are unlikely to be fruitful.[55] For example, when viewing a map and looking for the shortest driving route from Denver to New York in the East, one can in most cases skip looking at any path through San Francisco or other areas far to the West; thus, an AI wielding an pathfinding algorithm like A* can avoid the combinatorial explosion that would ensue if every possible route had to be ponderously considered in turn.[57]

The earliest (and easiest to understand) approach to AI was symbolism (such as formal logic): “If an otherwise healthy adult has a fever, then they may have influenza”. A second, more general, approach is Bayesian inference: “If the current patient has a fever, adjust the probability they have influenza in such-and-such way”. The third major approach, extremely popular in routine business AI applications, are analogizers such as SVM and nearest-neighbor: “After examining the records of known past patients whose temperature, symptoms, age, and other factors mostly match the current patient, X% of those patients turned out to have influenza”. A fourth approach is harder to intuitively understand, but is inspired by how the brain’s machinery works: the artificial neural network approach uses artificial “neurons” that can learn by comparing itself to the desired output and altering the strengths of the connections between its internal neurons to “reinforce” connections that seemed to be useful. These four main approaches can overlap with each other and with evolutionary systems; for example, neural nets can learn to make inferences, to generalize, and to make analogies. Some systems implicitly or explicitly use multiple of these approaches, alongside many other AI and non-AI algorithms;[58] the best approach is often different depending on the problem.[60]

Learning algorithms work on the basis that strategies, algorithms, and inferences that worked well in the past are likely to continue working well in the future. These inferences can be obvious, such as “since the sun rose every morning for the last 10,000 days, it will probably rise tomorrow morning as well”. They can be nuanced, such as “X% of families have geographically separate species with color variants, so there is an Y% chance that undiscovered black swans exist”. Learners also work on the basis of “Occam’s razor”: The simplest theory that explains the data is the likeliest. Therefore, to be successful, a learner must be designed such that it prefers simpler theories to complex theories, except in cases where the complex theory is proven substantially better. Settling on a bad, overly complex theory gerrymandered to fit all the past training data is known as overfitting. Many systems attempt to reduce overfitting by rewarding a theory in accordance with how well it fits the data, but penalizing the theory in accordance with how complex the theory is. Besides classic overfitting, learners can also disappoint by “learning the wrong lesson”. A toy example is that an image classifier trained only on pictures of brown horses and black cats might conclude that all brown patches are likely to be horses. A real-world example is that, unlike humans, current image classifiers don’t determine the spatial relationship between components of the picture; instead, they learn abstract patterns of pixels that humans are oblivious to, but that linearly correlate with images of certain types of real objects. Faintly superimposing such a pattern on a legitimate image results in an “adversarial” image that the system misclassifies.[c][63][64][65]

Compared with humans, existing AI lacks several features of human “commonsense reasoning”; most notably, humans have powerful mechanisms for reasoning about “nave physics” such as space, time, and physical interactions. This enables even young children to easily make inferences like “If I roll this pen off a table, it will fall on the floor”. Humans also have a powerful mechanism of “folk psychology” that helps them to interpret natural-language sentences such as “The city councilmen refused the demonstrators a permit because they advocated violence”. (A generic AI has difficulty inferring whether the councilmen or the demonstrators are the ones alleged to be advocating violence.)[68][69][70] This lack of “common knowledge” means that AI often makes different mistakes than humans make, in ways that can seem incomprehensible. For example, existing self-driving cars cannot reason about the location nor the intentions of pedestrians in the exact way that humans do, and instead must use non-human modes of reasoning to avoid accidents.[71][72][73]

The overall research goal of artificial intelligence is to create technology that allows computers and machines to function in an intelligent manner. The general problem of simulating (or creating) intelligence has been broken down into sub-problems. These consist of particular traits or capabilities that researchers expect an intelligent system to display. The traits described below have received the most attention.[13]

Early researchers developed algorithms that imitated step-by-step reasoning that humans use when they solve puzzles or make logical deductions.[74] By the late 1980s and 1990s, AI research had developed methods for dealing with uncertain or incomplete information, employing concepts from probability and economics.[75]

These algorithms proved to be insufficient for solving large reasoning problems, because they experienced a “combinatorial explosion”: they became exponentially slower as the problems grew larger.[55] In fact, even humans rarely use the step-by-step deduction that early AI research was able to model. They solve most of their problems using fast, intuitive judgements.[76]

Knowledge representation[77] and knowledge engineering[78] are central to classical AI research. Some “expert systems” attempt to gather together explicit knowledge possessed by experts in some narrow domain. In addition, some projects attempt to gather the “commonsense knowledge” known to the average person into a database containing extensive knowledge about the world. Among the things a comprehensive commonsense knowledge base would contain are: objects, properties, categories and relations between objects;[79] situations, events, states and time;[80] causes and effects;[81] knowledge about knowledge (what we know about what other people know);[82] and many other, less well researched domains. A representation of “what exists” is an ontology: the set of objects, relations, concepts, and properties formally described so that software agents can interpret them. The semantics of these are captured as description logic concepts, roles, and individuals, and typically implemented as classes, properties, and individuals in the Web Ontology Language.[83] The most general ontologies are called upper ontologies, which attempt to provide a foundation for all other knowledge[84] by acting as mediators between domain ontologies that cover specific knowledge about a particular knowledge domain (field of interest or area of concern). Such formal knowledge representations can be used in content-based indexing and retrieval,[85] scene interpretation,[86] clinical decision support,[87] knowledge discovery (mining “interesting” and actionable inferences from large databases),[88] and other areas.[89]

Among the most difficult problems in knowledge representation are:

Intelligent agents must be able to set goals and achieve them.[96] They need a way to visualize the futurea representation of the state of the world and be able to make predictions about how their actions will change itand be able to make choices that maximize the utility (or “value”) of available choices.[97]

In classical planning problems, the agent can assume that it is the only system acting in the world, allowing the agent to be certain of the consequences of its actions.[98] However, if the agent is not the only actor, then it requires that the agent can reason under uncertainty. This calls for an agent that can not only assess its environment and make predictions, but also evaluate its predictions and adapt based on its assessment.[99]

Multi-agent planning uses the cooperation and competition of many agents to achieve a given goal. Emergent behavior such as this is used by evolutionary algorithms and swarm intelligence.[100]

Machine learning, a fundamental concept of AI research since the field’s inception,[101] is the study of computer algorithms that improve automatically through experience.[102][103]

Unsupervised learning is the ability to find patterns in a stream of input. Supervised learning includes both classification and numerical regression. Classification is used to determine what category something belongs in, after seeing a number of examples of things from several categories. Regression is the attempt to produce a function that describes the relationship between inputs and outputs and predicts how the outputs should change as the inputs change.[103] Both classifiers and regression learners can be viewed as “function approximators” trying to learn an unknown (possibly implicit) function; for example, a spam classifier can be viewed as learning a function that maps from the text of an email to one of two categories, “spam” or “not spam”. Computational learning theory can assess learners by computational complexity, by sample complexity (how much data is required), or by other notions of optimization.[104] In reinforcement learning[105] the agent is rewarded for good responses and punished for bad ones. The agent uses this sequence of rewards and punishments to form a strategy for operating in its problem space.

Natural language processing[106] (NLP) gives machines the ability to read and understand human language. A sufficiently powerful natural language processing system would enable natural-language user interfaces and the acquisition of knowledge directly from human-written sources, such as newswire texts. Some straightforward applications of natural language processing include information retrieval, text mining, question answering[107] and machine translation.[108] Many current approaches use word co-occurrence frequencies to construct syntactic representations of text. “Keyword spotting” strategies for search are popular and scalable but dumb; a search query for “dog” might only match documents with the literal word “dog” and miss a document with the word “poodle”. “Lexical affinity” strategies use the occurrence of words such as “accident” to assess the sentiment of a document. Modern statistical NLP approaches can combine all these strategies as well as others, and often achieve acceptable accuracy at the page or paragraph level, but continue to lack the semantic understanding required to classify isolated sentences well. Besides the usual difficulties with encoding semantic commonsense knowledge, existing semantic NLP sometimes scales too poorly to be viable in business applications. Beyond semantic NLP, the ultimate goal of “narrative” NLP is to embody a full understanding of commonsense reasoning.[109]

Machine perception[110] is the ability to use input from sensors (such as cameras (visible spectrum or infrared), microphones, wireless signals, and active lidar, sonar, radar, and tactile sensors) to deduce aspects of the world. Applications include speech recognition,[111] facial recognition, and object recognition.[112] Computer vision is the ability to analyze visual input. Such input is usually ambiguous; a giant, fifty-meter-tall pedestrian far away may produce exactly the same pixels as a nearby normal-sized pedestrian, requiring the AI to judge the relative likelihood and reasonableness of different interpretations, for example by using its “object model” to assess that fifty-meter pedestrians do not exist.[113]

AI is heavily used in robotics.[114] Advanced robotic arms and other industrial robots, widely used in modern factories, can learn from experience how to move efficiently despite the presence of friction and gear slippage.[115] A modern mobile robot, when given a small, static, and visible environment, can easily determine its location and map its environment; however, dynamic environments, such as (in endoscopy) the interior of a patient’s breathing body, pose a greater challenge. Motion planning is the process of breaking down a movement task into “primitives” such as individual joint movements. Such movement often involves compliant motion, a process where movement requires maintaining physical contact with an object.[117][118] Moravec’s paradox generalizes that low-level sensorimotor skills that humans take for granted are, counterintuitively, difficult to program into a robot; the paradox is named after Hans Moravec, who stated in 1988 that “it is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility”.[119][120] This is attributed to the fact that, unlike checkers, physical dexterity has been a direct target of natural selection for millions of years.[121]

Moravec’s paradox can be extended to many forms of social intelligence.[123][124] Distributed multi-agent coordination of autonomous vehicles remains a difficult problem.[125] Affective computing is an interdisciplinary umbrella that comprises systems which recognize, interpret, process, or simulate human affects. Moderate successes related to affective computing include textual sentiment analysis and, more recently, multimodal affect analysis (see multimodal sentiment analysis), wherein AI classifies the affects displayed by a videotaped subject.[129]

In the long run, social skills and an understanding of human emotion and game theory would be valuable to a social agent. Being able to predict the actions of others by understanding their motives and emotional states would allow an agent to make better decisions. Some computer systems mimic human emotion and expressions to appear more sensitive to the emotional dynamics of human interaction, or to otherwise facilitate humancomputer interaction.[130] Similarly, some virtual assistants are programmed to speak conversationally or even to banter humorously; this tends to give naive users an unrealistic conception of how intelligent existing computer agents actually are.[131]

Historically, projects such as the Cyc knowledge base (1984) and the massive Japanese Fifth Generation Computer Systems initiative (19821992) attempted to cover the breadth of human cognition. These early projects failed to escape the limitations of non-quantitative symbolic logic models and, in retrospect, greatly underestimated the difficulty of cross-domain AI. Nowadays, the vast majority of current AI researchers work instead on tractable “narrow AI” applications (such as medical diagnosis or automobile navigation).[132] Many researchers predict that such “narrow AI” work in different individual domains will eventually be incorporated into a machine with artificial general intelligence (AGI), combining most of the narrow skills mentioned in this article and at some point even exceeding human ability in most or all these areas.[17][133] Many advances have general, cross-domain significance. One high-profile example is that DeepMind in the 2010s developed a “generalized artificial intelligence” that could learn many diverse Atari games on its own, and later developed a variant of the system which succeeds at sequential learning.[134][135][136] Besides transfer learning,[137] hypothetical AGI breakthroughs could include the development of reflective architectures that can engage in decision-theoretic metareasoning, and figuring out how to “slurp up” a comprehensive knowledge base from the entire unstructured Web. Some argue that some kind of (currently-undiscovered) conceptually straightforward, but mathematically difficult, “Master Algorithm” could lead to AGI. Finally, a few “emergent” approaches look to simulating human intelligence extremely closely, and believe that anthropomorphic features like an artificial brain or simulated child development may someday reach a critical point where general intelligence emerges.[139][140]

Many of the problems in this article may also require general intelligence, if machines are to solve the problems as well as people do. For example, even specific straightforward tasks, like machine translation, require that a machine read and write in both languages (NLP), follow the author’s argument (reason), know what is being talked about (knowledge), and faithfully reproduce the author’s original intent (social intelligence). A problem like machine translation is considered “AI-complete”, because all of these problems need to be solved simultaneously in order to reach human-level machine performance.

There is no established unifying theory or paradigm that guides AI research. Researchers disagree about many issues.[141] A few of the most long standing questions that have remained unanswered are these: should artificial intelligence simulate natural intelligence by studying psychology or neurobiology? Or is human biology as irrelevant to AI research as bird biology is to aeronautical engineering?[14]Can intelligent behavior be described using simple, elegant principles (such as logic or optimization)? Or does it necessarily require solving a large number of completely unrelated problems?[15]

In the 1940s and 1950s, a number of researchers explored the connection between neurobiology, information theory, and cybernetics. Some of them built machines that used electronic networks to exhibit rudimentary intelligence, such as W. Grey Walter’s turtles and the Johns Hopkins Beast. Many of these researchers gathered for meetings of the Teleological Society at Princeton University and the Ratio Club in England.[142] By 1960, this approach was largely abandoned, although elements of it would be revived in the 1980s.

When access to digital computers became possible in the middle 1950s, AI research began to explore the possibility that human intelligence could be reduced to symbol manipulation. The research was centered in three institutions: Carnegie Mellon University, Stanford and MIT, and as described below, each one developed its own style of research. John Haugeland named these symbolic approaches to AI “good old fashioned AI” or “GOFAI”.[143] During the 1960s, symbolic approaches had achieved great success at simulating high-level thinking in small demonstration programs. Approaches based on cybernetics or artificial neural networks were abandoned or pushed into the background.[144]Researchers in the 1960s and the 1970s were convinced that symbolic approaches would eventually succeed in creating a machine with artificial general intelligence and considered this the goal of their field.

Economist Herbert Simon and Allen Newell studied human problem-solving skills and attempted to formalize them, and their work laid the foundations of the field of artificial intelligence, as well as cognitive science, operations research and management science. Their research team used the results of psychological experiments to develop programs that simulated the techniques that people used to solve problems. This tradition, centered at Carnegie Mellon University would eventually culminate in the development of the Soar architecture in the middle 1980s.[145][146]

Unlike Simon and Newell, John McCarthy felt that machines did not need to simulate human thought, but should instead try to find the essence of abstract reasoning and problem-solving, regardless of whether people used the same algorithms.[14] His laboratory at Stanford (SAIL) focused on using formal logic to solve a wide variety of problems, including knowledge representation, planning and learning.[147] Logic was also the focus of the work at the University of Edinburgh and elsewhere in Europe which led to the development of the programming language Prolog and the science of logic programming.[148]

Researchers at MIT (such as Marvin Minsky and Seymour Papert)[149] found that solving difficult problems in vision and natural language processing required ad-hoc solutions they argued that there was no simple and general principle (like logic) that would capture all the aspects of intelligent behavior. Roger Schank described their “anti-logic” approaches as “scruffy” (as opposed to the “neat” paradigms at CMU and Stanford).[15] Commonsense knowledge bases (such as Doug Lenat’s Cyc) are an example of “scruffy” AI, since they must be built by hand, one complicated concept at a time.[150]

When computers with large memories became available around 1970, researchers from all three traditions began to build knowledge into AI applications.[151] This “knowledge revolution” led to the development and deployment of expert systems (introduced by Edward Feigenbaum), the first truly successful form of AI software.[37] The knowledge revolution was also driven by the realization that enormous amounts of knowledge would be required by many simple AI applications.

By the 1980s progress in symbolic AI seemed to stall and many believed that symbolic systems would never be able to imitate all the processes of human cognition, especially perception, robotics, learning and pattern recognition. A number of researchers began to look into “sub-symbolic” approaches to specific AI problems.[16] Sub-symbolic methods manage to approach intelligence without specific representations of knowledge.

This includes embodied, situated, behavior-based, and nouvelle AI. Researchers from the related field of robotics, such as Rodney Brooks, rejected symbolic AI and focused on the basic engineering problems that would allow robots to move and survive.[152] Their work revived the non-symbolic viewpoint of the early cybernetics researchers of the 1950s and reintroduced the use of control theory in AI. This coincided with the development of the embodied mind thesis in the related field of cognitive science: the idea that aspects of the body (such as movement, perception and visualization) are required for higher intelligence.

Within developmental robotics, developmental learning approaches are elaborated upon to allow robots to accumulate repertoires of novel skills through autonomous self-exploration, social interaction with human teachers, and the use of guidance mechanisms (active learning, maturation, motor synergies, etc.).[153][154]

Interest in neural networks and “connectionism” was revived by David Rumelhart and others in the middle of the 1980s.[157] Artificial neural networks are an example of soft computing — they are solutions to problems which cannot be solved with complete logical certainty, and where an approximate solution is often sufficient. Other soft computing approaches to AI include fuzzy systems, evolutionary computation and many statistical tools. The application of soft computing to AI is studied collectively by the emerging discipline of computational intelligence.[158]

Much of traditional GOFAI got bogged down on ad hoc patches to symbolic computation that worked on their own toy models but failed to generalize to real-world results. However, around the 1990s, AI researchers adopted sophisticated mathematical tools, such as hidden Markov models (HMM), information theory, and normative Bayesian decision theory to compare or to unify competing architectures. The shared mathematical language permitted a high level of collaboration with more established fields (like mathematics, economics or operations research).[d] Compared with GOFAI, new “statistical learning” techniques such as HMM and neural networks were gaining higher levels of accuracy in many practical domains such as data mining, without necessarily acquiring semantic understanding of the datasets. The increased successes with real-world data led to increasing emphasis on comparing different approaches against shared test data to see which approach performed best in a broader context than that provided by idiosyncratic toy models; AI research was becoming more scientific. Nowadays results of experiments are often rigorously measurable, and are sometimes (with difficulty) reproducible.[38][159] Different statistical learning techniques have different limitations; for example, basic HMM cannot model the infinite possible combinations of natural language. Critics note that the shift from GOFAI to statistical learning is often also a shift away from Explainable AI. In AGI research, some scholars caution against over-reliance on statistical learning, and argue that continuing research into GOFAI will still be necessary to attain general intelligence.

AI has developed a large number of tools to solve the most difficult problems in computer science. A few of the most general of these methods are discussed below.

Many problems in AI can be solved in theory by intelligently searching through many possible solutions:[168] Reasoning can be reduced to performing a search. For example, logical proof can be viewed as searching for a path that leads from premises to conclusions, where each step is the application of an inference rule.[169] Planning algorithms search through trees of goals and subgoals, attempting to find a path to a target goal, a process called means-ends analysis.[170] Robotics algorithms for moving limbs and grasping objects use local searches in configuration space.[115] Many learning algorithms use search algorithms based on optimization.

Simple exhaustive searches[171] are rarely sufficient for most real-world problems: the search space (the number of places to search) quickly grows to astronomical numbers. The result is a search that is too slow or never completes. The solution, for many problems, is to use “heuristics” or “rules of thumb” that prioritize choices in favor of those that are more likely to reach a goal and to do so in a shorter number of steps. In some search methodologies heuristics can also serve to entirely eliminate some choices that are unlikely to lead to a goal (called “pruning the search tree”). Heuristics supply the program with a “best guess” for the path on which the solution lies.[172] Heuristics limit the search for solutions into a smaller sample size.

A very different kind of search came to prominence in the 1990s, based on the mathematical theory of optimization. For many problems, it is possible to begin the search with some form of a guess and then refine the guess incrementally until no more refinements can be made. These algorithms can be visualized as blind hill climbing: we begin the search at a random point on the landscape, and then, by jumps or steps, we keep moving our guess uphill, until we reach the top. Other optimization algorithms are simulated annealing, beam search and random optimization.[173]

Evolutionary computation uses a form of optimization search. For example, they may begin with a population of organisms (the guesses) and then allow them to mutate and recombine, selecting only the fittest to survive each generation (refining the guesses). Classic evolutionary algorithms include genetic algorithms, gene expression programming, and genetic programming.[174] Alternatively, distributed search processes can coordinate via swarm intelligence algorithms. Two popular swarm algorithms used in search are particle swarm optimization (inspired by bird flocking) and ant colony optimization (inspired by ant trails).[175][176]

Logic[177] is used for knowledge representation and problem solving, but it can be applied to other problems as well. For example, the satplan algorithm uses logic for planning[178] and inductive logic programming is a method for learning.[179]

Several different forms of logic are used in AI research. Propositional logic[180] involves truth functions such as “or” and “not”. First-order logic[181] adds quantifiers and predicates, and can express facts about objects, their properties, and their relations with each other. Fuzzy set theory assigns a “degree of truth” (between 0 and 1) to vague statements such as “Alice is old” (or rich, or tall, or hungry) that are too linguistically imprecise to be completely true or false. Fuzzy logic is successfully used in control systems to allow experts to contribute vague rules such as “if you are close to the destination station and moving fast, increase the train’s brake pressure”; these vague rules can then be numerically refined within the system. Fuzzy logic fails to scale well in knowledge bases; many AI researchers question the validity of chaining fuzzy-logic inferences.[e][183][184]

Default logics, non-monotonic logics and circumscription[91] are forms of logic designed to help with default reasoning and the qualification problem. Several extensions of logic have been designed to handle specific domains of knowledge, such as: description logics;[79] situation calculus, event calculus and fluent calculus (for representing events and time);[80] causal calculus;[81] belief calculus;[185] and modal logics.[82]

Overall, qualitiative symbolic logic is brittle and scales poorly in the presence of noise or other uncertainty. Exceptions to rules are numerous, and it is difficult for logical systems to function in the presence of contradictory rules.[187]

Many problems in AI (in reasoning, planning, learning, perception, and robotics) require the agent to operate with incomplete or uncertain information. AI researchers have devised a number of powerful tools to solve these problems using methods from probability theory and economics.[188]

Bayesian networks[189] are a very general tool that can be used for a large number of problems: reasoning (using the Bayesian inference algorithm),[190] learning (using the expectation-maximization algorithm),[f][192] planning (using decision networks)[193] and perception (using dynamic Bayesian networks).[194] Probabilistic algorithms can also be used for filtering, prediction, smoothing and finding explanations for streams of data, helping perception systems to analyze processes that occur over time (e.g., hidden Markov models or Kalman filters).[194] Compared with symbolic logic, formal Bayesian inference is computationally expensive. For inference to be tractable, most observations must be conditionally independent of one another. Complicated graphs with diamonds or other “loops” (undirected cycles) can require a sophisticated method such as Markov Chain Monte Carlo, which spreads an ensemble of random walkers throughout the Bayesian network and attempts to converge to an assessment of the conditional probabilities. Bayesian networks are used on XBox Live to rate and match players; wins and losses are “evidence” of how good a player is. AdSense uses a Bayesian network with over 300 million edges to learn which ads to serve.

A key concept from the science of economics is “utility”: a measure of how valuable something is to an intelligent agent. Precise mathematical tools have been developed that analyze how an agent can make choices and plan, using decision theory, decision analysis,[196] and information value theory.[97] These tools include models such as Markov decision processes,[197] dynamic decision networks,[194] game theory and mechanism design.[198]

The simplest AI applications can be divided into two types: classifiers (“if shiny then diamond”) and controllers (“if shiny then pick up”). Controllers do, however, also classify conditions before inferring actions, and therefore classification forms a central part of many AI systems. Classifiers are functions that use pattern matching to determine a closest match. They can be tuned according to examples, making them very attractive for use in AI. These examples are known as observations or patterns. In supervised learning, each pattern belongs to a certain predefined class. A class can be seen as a decision that has to be made. All the observations combined with their class labels are known as a data set. When a new observation is received, that observation is classified based on previous experience.[199]

A classifier can be trained in various ways; there are many statistical and machine learning approaches. The decision tree[200] is perhaps the most widely used machine learning algorithm. Other widely used classifiers are the neural network,[202]k-nearest neighbor algorithm,[g][204]kernel methods such as the support vector machine (SVM),[h][206]Gaussian mixture model,[207] and the extremely popular naive Bayes classifier.[i][209] Classifier performance depends greatly on the characteristics of the data to be classified, such as the dataset size, the dimensionality, and the level of noise. Model-based classifiers perform well if the assumed model is an extremely good fit for the actual data. Otherwise, if no matching model is available, and if accuracy (rather than speed or scalability) is the sole concern, conventional wisdom is that discriminative classifiers (especially SVM) tend to be more accurate than model-based classifiers such as “naive Bayes” on most practical data sets.[210]

Neural networks, or neural nets, were inspired by the architecture of neurons in the human brain. A simple “neuron” N accepts input from multiple other neurons, each of which, when activated (or “fired”), cast a weighted “vote” for or against whether neuron N should itself activate. Learning requires an algorithm to adjust these weights based on the training data; one simple algorithm (dubbed “fire together, wire together”) is to increase the weight between two connected neurons when the activation of one triggers the successful activation of another. The net forms “concepts” that are distributed among a subnetwork of shared[j] neurons that tend to fire together; a concept meaning “leg” might be coupled with a subnetwork meaning “foot” that includes the sound for “foot”. Neurons have a continuous spectrum of activation; in addition, neurons can process inputs in a nonlinear way rather than weighing straightforward votes. Modern neural nets can learn both continuous functions and, surprisingly, digital logical operations. Neural networks’ early successes included predicting the stock market and (in 1995) a mostly self-driving car.[k] In the 2010s, advances in neural networks using deep learning thrust AI into widespread public consciousness and contributed to an enormous upshift in corporate AI spending; for example, AI-related M&A in 2017 was over 25 times as large as in 2015.[213][214]

The study of non-learning artificial neural networks[202] began in the decade before the field of AI research was founded, in the work of Walter Pitts and Warren McCullouch. Frank Rosenblatt invented the perceptron, a learning network with a single layer, similar to the old concept of linear regression. Early pioneers also include Alexey Grigorevich Ivakhnenko, Teuvo Kohonen, Stephen Grossberg, Kunihiko Fukushima, Christoph von der Malsburg, David Willshaw, Shun-Ichi Amari, Bernard Widrow, John Hopfield, Eduardo R. Caianiello, and others.

The main categories of networks are acyclic or feedforward neural networks (where the signal passes in only one direction) and recurrent neural networks (which allow feedback and short-term memories of previous input events). Among the most popular feedforward networks are perceptrons, multi-layer perceptrons and radial basis networks.[215] Neural networks can be applied to the problem of intelligent control (for robotics) or learning, using such techniques as Hebbian learning (“fire together, wire together”), GMDH or competitive learning.[216]

Today, neural networks are often trained by the backpropagation algorithm, which had been around since 1970 as the reverse mode of automatic differentiation published by Seppo Linnainmaa,[217][218] and was introduced to neural networks by Paul Werbos.[219][220][221]

Hierarchical temporal memory is an approach that models some of the structural and algorithmic properties of the neocortex.[222]

In short, most neural networks use some form of gradient descent on a hand-created neural topology. However, some research groups, such as Uber, argue that simple neuroevolution to mutate new neural network topologies and weights may be competitive with sophisticated gradient descent approaches. One advantage of neuroevolution is that it may be less prone to get caught in “dead ends”.[223]

Deep learning is any artificial neural network that can learn a long chain of causal links. For example, a feedforward network with six hidden layers can learn a seven-link causal chain (six hidden layers + output layer) and has a “credit assignment path” (CAP) depth of seven. Many deep learning systems need to be able to learn chains ten or more causal links in length.[224] Deep learning has transformed many important subfields of artificial intelligence, including computer vision, speech recognition, natural language processing and others.[225][226][224]

According to one overview,[227] the expression “Deep Learning” was introduced to the Machine Learning community by Rina Dechter in 1986[228] and gained traction afterIgor Aizenberg and colleagues introduced it to Artificial Neural Networks in 2000.[229] The first functional Deep Learning networks were published by Alexey Grigorevich Ivakhnenko and V. G. Lapa in 1965.[230][pageneeded] These networks are trained one layer at a time. Ivakhnenko’s 1971 paper[231] describes the learning of a deep feedforward multilayer perceptron with eight layers, already much deeper than many later networks. In 2006, a publication by Geoffrey Hinton and Ruslan Salakhutdinov introduced another way of pre-training many-layered feedforward neural networks (FNNs) one layer at a time, treating each layer in turn as an unsupervised restricted Boltzmann machine, then using supervised backpropagation for fine-tuning. Similar to shallow artificial neural networks, deep neural networks can model complex non-linear relationships. Over the last few years, advances in both machine learning algorithms and computer hardware have led to more efficient methods for training deep neural networks that contain many layers of non-linear hidden units and a very large output layer.[233]

Deep learning often uses convolutional neural networks (CNNs), whose origins can be traced back to the Neocognitron introduced by Kunihiko Fukushima in 1980.[234] In 1989, Yann LeCun and colleagues applied backpropagation to such an architecture. In the early 2000s, in an industrial application CNNs already processed an estimated 10% to 20% of all the checks written in the US.[235]Since 2011, fast implementations of CNNs on GPUs havewon many visual pattern recognition competitions.[224]

CNNs with 12 convolutional layers were used in conjunction with reinforcement learning by Deepmind’s “AlphaGo Lee”, the program that beat a top Go champion in 2016.[236]

Early on, deep learning was also applied to sequence learning with recurrent neural networks (RNNs)[237] which are in theory Turing complete[238] and can run arbitrary programs to process arbitrary sequences of inputs. The depth of an RNN is unlimited and depends on the length of its input sequence; thus, an RNN is an example of deep learning.[224] RNNs can be trained by gradient descent[239][240][241] but suffer from the vanishing gradient problem.[225][242] In 1992, it was shown that unsupervised pre-training of a stack of recurrent neural networks can speed up subsequent supervised learning of deep sequential problems.[243]

Numerous researchers now use variants of a deep learning recurrent NN called the long short-term memory (LSTM) network published by Hochreiter & Schmidhuber in 1997.[244] LSTM is often trained by Connectionist Temporal Classification (CTC).[245] At Google, Microsoft and Baidu this approach has revolutionised speech recognition.[246][247][248] For example, in 2015, Google’s speech recognition experienced a dramatic performance jump of 49% through CTC-trained LSTM, which is now available through Google Voice to billions of smartphone users.[249] Google also used LSTM to improve machine translation,[250] Language Modeling[251] and Multilingual Language Processing.[252] LSTM combined with CNNs also improved automatic image captioning[253] and a plethora of other applications.

AI, like electricity or the steam engine, is a general purpose technology. There is no consensus on how to characterize which tasks AI tends to excel at.[254] While projects such as AlphaZero have succeeded in generating their own knowledge from scratch, many other machine learning projects require large training datasets.[255][256] Researcher Andrew Ng has suggested, as a “highly imperfect rule of thumb”, that “almost anything a typical human can do with less than one second of mental thought, we can probably now or in the near future automate using AI.”[257] Moravec’s paradox suggests that AI lags humans at many tasks that the human brain has specifically evolved to perform well.[121]

Games provide a well-publicized benchmark for assessing rates of progress. AlphaGo around 2016 brought the era of classical board-game benchmarks to a close. Games of imperfect knowledge provide new challenges to AI in the area of game theory.[258][259] E-sports such as StarCraft continue to provide additional public benchmarks.[260][261] There are many competitions and prizes, such as the Imagenet Challenge, to promote research in artificial intelligence. The main areas of competition include general machine intelligence, conversational behavior, data-mining, robotic cars, and robot soccer as well as conventional games.[citation needed]

The “imitation game” (an interpretation of the 1950 Turing test that assesses whether a computer can imitate a human) is nowadays considered too exploitable to be a meaningful benchmark.[262] A derivative of the Turing test is the Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA). As the name implies, this helps to determine that a user is an actual person and not a computer posing as a human. In contrast to the standard Turing test, CAPTCHA is administered by a machine and targeted to a human as opposed to being administered by a human and targeted to a machine. A computer asks a user to complete a simple test then generates a grade for that test. Computers are unable to solve the problem, so correct solutions are deemed to be the result of a person taking the test. A common type of CAPTCHA is the test that requires the typing of distorted letters, numbers or symbols that appear in an image undecipherable by a computer.

Proposed “universal intelligence” tests aim to compare how well machines, humans, and even non-human animals perform on problem sets that are generic as possible. At an extreme, the test suite can contain every possible problem, weighted by Kolmogorov complexity; unfortunately, these problem sets tend to be dominated by impoverished pattern-matching exercises where a tuned AI can easily exceed human performance levels.[264][265]

AI is relevant to any intellectual task. Modern artificial intelligence techniques are pervasive and are too numerous to list here. Frequently, when a technique reaches mainstream use, it is no longer considered artificial intelligence; this phenomenon is described as the AI effect.

High-profile examples of AI include autonomous vehicles (such as drones and self-driving cars), medical diagnosis, creating art (such as poetry), proving mathematical theorems, playing games (such as Chess or Go), search engines (such as Google search), online assistants (such as Siri), image recognition in photographs, spam filtering, prediction of judicial decisions[268] and targeting online advertisements.[269][270]

With social media sites overtaking TV as a source for news for young people and news organisations increasingly reliant on social media platforms for generating distribution,[271] major publishers now use artificial intelligence (AI) technology to post stories more effectively and generate higher volumes of traffic.[272]

Artificial intelligence is breaking into the healthcare industry by assisting doctors. According to Bloomberg Technology, Microsoft has developed AI to help doctors find the right treatments for cancer.[273] There is a great amount of research and drugs developed relating to cancer. In detail, there are more than 800 medicines and vaccines to treat cancer. This negatively affects the doctors, because there are too many options to choose from, making it more difficult to choose the right drugs for the patients. Microsoft is working on a project to develop a machine called “Hanover”. Its goal is to memorize all the papers necessary to cancer and help predict which combinations of drugs will be most effective for each patient. One project that is being worked on at the moment is fighting myeloid leukemia, a fatal cancer where the treatment has not improved in decades. Another study was reported to have found that artificial intelligence was as good as trained doctors in identifying skin cancers.[274] Another study is using artificial intelligence to try and monitor multiple high-risk patients, and this is done by asking each patient numerous questions based on data acquired from live doctor to patient interactions.[275]

According to CNN, a recent study by surgeons at the Children’s National Medical Center in Washington successfully demonstrated surgery with an autonomous robot. The team supervised the robot while it performed soft-tissue surgery, stitching together a pig’s bowel during open surgery, and doing so better than a human surgeon, the team claimed.[276] IBM has created its own artificial intelligence computer, the IBM Watson, which has beaten human intelligence (at some levels). Watson not only won at the game show Jeopardy! against former champions,[277] but was declared a hero after successfully diagnosing a woman who was suffering from leukemia.[278]

Advancements in AI have contributed to the growth of the automotive industry through the creation and evolution of self-driving vehicles. As of 2016, there are over 30 companies utilizing AI into the creation of driverless cars. A few companies involved with AI include Tesla, Google, and Apple.[279]

Many components contribute to the functioning of self-driving cars. These vehicles incorporate systems such as braking, lane changing, collision prevention, navigation and mapping. Together, these systems, as well as high performance computers, are integrated into one complex vehicle.[280]

Recent developments in autonomous automobiles have made the innovation of self-driving trucks possible, though they are still in the testing phase. The UK government has passed legislation to begin testing of self-driving truck platoons in 2018.[281] Self-driving truck platoons are a fleet of self-driving trucks following the lead of one non-self-driving truck, so the truck platoons aren’t entirely autonomous yet. Meanwhile, the Daimler, a German automobile corporation, is testing the Freightliner Inspiration which is a semi-autonomous truck that will only be used on the highway.[282]

One main factor that influences the ability for a driver-less automobile to function is mapping. In general, the vehicle would be pre-programmed with a map of the area being driven. This map would include data on the approximations of street light and curb heights in order for the vehicle to be aware of its surroundings. However, Google has been working on an algorithm with the purpose of eliminating the need for pre-programmed maps and instead, creating a device that would be able to adjust to a variety of new surroundings.[283] Some self-driving cars are not equipped with steering wheels or brake pedals, so there has also been research focused on creating an algorithm that is capable of maintaining a safe environment for the passengers in the vehicle through awareness of speed and driving conditions.[284]

Another factor that is influencing the ability for a driver-less automobile is the safety of the passenger. To make a driver-less automobile, engineers must program it to handle high-risk situations. These situations could include a head-on collision with pedestrians. The car’s main goal should be to make a decision that would avoid hitting the pedestrians and saving the passengers in the car. But there is a possibility the car would need to make a decision that would put someone in danger. In other words, the car would need to decide to save the pedestrians or the passengers.[285] The programing of the car in these situations is crucial to a successful driver-less automobile.

Financial institutions have long used artificial neural network systems to detect charges or claims outside of the norm, flagging these for human investigation. The use of AI in banking can be traced back to 1987 when Security Pacific National Bank in US set-up a Fraud Prevention Task force to counter the unauthorised use of debit cards. Programs like Kasisto and Moneystream are using AI in financial services.

Banks use artificial intelligence systems today to organize operations, maintain book-keeping, invest in stocks, and manage properties. AI can react to changes overnight or when business is not taking place.[286] In August 2001, robots beat humans in a simulated financial trading competition.[287] AI has also reduced fraud and financial crimes by monitoring behavioral patterns of users for any abnormal changes or anomalies.[288]

The use of AI machines in the market in applications such as online trading and decision making has changed major economic theories.[289] For example, AI based buying and selling platforms have changed the law of supply and demand in that it is now possible to easily estimate individualized demand and supply curves and thus individualized pricing. Furthermore, AI machines reduce information asymmetry in the market and thus making markets more efficient while reducing the volume of trades. Furthermore, AI in the markets limits the consequences of behavior in the markets again making markets more efficient. Other theories where AI has had impact include in rational choice, rational expectations, game theory, Lewis turning point, portfolio optimization and counterfactual thinking.

In video games, artificial intelligence is routinely used to generate dynamic purposeful behavior in non-player characters (NPCs). In addition, well-understood AI techniques are routinely used for pathfinding. Some researchers consider NPC AI in games to be a “solved problem” for most production tasks. Games with more atypical AI include the AI director of Left 4 Dead (2008) and the neuroevolutionary training of platoons in Supreme Commander 2 (2010).[290][291]

Worldwide annual military spending on robotics rose from 5.1 billion USD in 2010 to 7.5 billion USD in 2015.[292][293] Military drones capable of autonomous action are widely considered a useful asset. In 2017, Vladimir Putin stated that “Whoever becomes the leader in (artificial intelligence) will become the ruler of the world”.[294][295] Many artificial intelligence researchers seek to distance themselves from military applications of AI.[296]

For financial statements audit, AI makes continuous audit possible. AI tools could analyze many sets of different information immediately. The potential benefit would be the overall audit risk will be reduced, the level of assurance will be increased and the time duration of audit will be reduced.[297]

A report by the Guardian newspaper in the UK in 2018 found that online gambling companies were using AI to predict the behavior of customers in order to target them with personalized promotions.[298] Developers of commercial AI platforms are also beginning to appeal more directly to casino operators, offering a range of existing and potential services to help them boost their profits and expand their customer base.[299]

Artificial Intelligence has inspired numerous creative applications including its usage to produce visual art. The exhibition “Thinking Machines: Art and Design in the Computer Age, 1959-1989” at MoMA [300] provides a good overview of the historical applications of AI for art, architecture, and design. Recent exhibitions showcasing the usage of AI to produce art include the Google-sponsored benefit and auction at the Gray Area Foundation in San Francisco, where artists experimented with the deepdream algorithm [301] and the exhibition “Unhuman: Art in the Age of AI,” which took place in Los Angeles and Frankfurt in the fall of 2017. [302][303] In the spring of 2018, the Association of Computing Machinery dedicated a special magazine issue to the subject of computers and art highlighting the role of machine learning in the arts.[304]

There are three philosophical questions related to AI:

Can a machine be intelligent? Can it “think”?

Link:

Artificial intelligence – Wikipedia

What is AI (artificial intelligence)? – Definition from …

AI (artificial intelligence) is the simulation of human intelligence processes by machines, especially computer systems. These processes include learning (the acquisition of information and rules for using the information), reasoning (using the rules to reach approximate or definite conclusions) and self-correction. Particular applications of AI include expert systems, speech recognition and machine vision.

AI was coined by John McCarthy, an American computer scientist, in 1956 at The Dartmouth Conference where the discipline was born. Today, it is an umbrella term that encompasses everything from robotic process automation to actual robotics. It has gained prominence recently due, in part, to big data, or the increase in speed, size and variety of data businesses are now collecting. AI can perform tasks such as identifying patterns in the data more efficiently than humans, enabling businesses to gain more insight out of their data.

AI can be categorized in any number of ways, but here are two examples.

The first classifies AI systems as either weak AI or strong AI. Weak AI, also known as narrow AI, is an AI system that is designed and trained for a particular task. Virtual personal assistants, such as Apple’s Siri, are a form of weak AI.

Strong AI, also known as artificial general intelligence, is an AI system with generalized human cognitive abilities so that when presented with an unfamiliar task, it has enough intelligence to find a solution. The Turing Test, developed by mathematician Alan Turing in 1950, is a method used to determine if a computer can actually think like a human, although the method is controversial.

The second example is from Arend Hintze, an assistant professor of integrative biology and computer science and engineering at Michigan State University. He categorizes AI into four types, from the kind of AI systems that exist today to sentient systems, which do not yet exist. His categories are as follows:

Go here to read the rest:

What is AI (artificial intelligence)? – Definition from …

What is Artificial Intelligence (AI)? – Definition from …

Artificial intelligence is a branch of computer science that aims to create intelligent machines. It has become an essential part of the technology industry.

Research associated with artificial intelligence is highly technical and specialized. The core problems of artificial intelligence include programming computers for certain traits such as:

Knowledge engineering is a core part of AI research. Machines can often act and react like humans only if they have abundant information relating to the world. Artificial intelligence must have access to objects, categories, properties and relations between all of them to implement knowledge engineering. Initiating common sense, reasoning and problem-solving power in machines is a difficult and tedious task.

Machine learning is also a core part of AI. Learning without any kind of supervision requires an ability to identify patterns in streams of inputs, whereas learning with adequate supervision involves classification and numerical regressions. Classification determines the category an object belongs to and regression deals with obtaining a set of numerical input or output examples, thereby discovering functions enabling the generation of suitable outputs from respective inputs. Mathematical analysis of machine learning algorithms and their performance is a well-defined branch of theoretical computer science often referred to as computational learning theory.

Machine perception deals with the capability to use sensory inputs to deduce the different aspects of the world, while computer vision is the power to analyze visual inputs with a few sub-problems such as facial, object and gesture recognition.

Robotics is also a major field related to AI. Robots require intelligence to handle tasks such as object manipulation and navigation, along with sub-problems of localization, motion planning and mapping.

Continued here:

What is Artificial Intelligence (AI)? – Definition from …

Artificial intelligence – Wikipedia

Intelligence demonstrated by machines

Artificial intelligence (AI), sometimes called machine intelligence, is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans and other animals. In computer science AI research is defined as the study of “intelligent agents”: any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] Colloquially, the term “artificial intelligence” is applied when a machine mimics “cognitive” functions that humans associate with other human minds, such as “learning” and “problem solving”.

The scope of AI is disputed: as machines become increasingly capable, tasks considered as requiring “intelligence” are often removed from the definition, a phenomenon known as the AI effect, leading to the quip, “AI is whatever hasn’t been done yet.”[3] For instance, optical character recognition is frequently excluded from “artificial intelligence”, having become a routine technology.[4] Capabilities generally classified as AI as of 2017[update] include successfully understanding human speech, competing at the highest level in strategic game systems (such as chess and Go[6]), autonomous cars, intelligent routing in content delivery network and military simulations.

Artificial intelligence was founded as an academic discipline in 1956, and in the years since has experienced several waves of optimism,[7][8] followed by disappointment and the loss of funding (known as an “AI winter”),[9][10] followed by new approaches, success and renewed funding.[8][11] For most of its history, AI research has been divided into subfields that often fail to communicate with each other.[12] These sub-fields are based on technical considerations, such as particular goals (e.g. “robotics” or “machine learning”),[13] the use of particular tools (“logic” or artificial neural networks), or deep philosophical differences.[14][15][16] Subfields have also been based on social factors (particular institutions or the work of particular researchers).[12]

The traditional problems (or goals) of AI research include reasoning, knowledge representation, planning, learning, natural language processing, perception and the ability to move and manipulate objects.[13] General intelligence is among the field’s long-term goals.[17] Approaches include statistical methods, computational intelligence, and traditional symbolic AI. Many tools are used in AI, including versions of search and mathematical optimization, artificial neural networks, and methods based on statistics, probability and economics. The AI field draws upon computer science, mathematics, psychology, linguistics, philosophy and many others.

The field was founded on the claim that human intelligence “can be so precisely described that a machine can be made to simulate it”.[18] This raises philosophical arguments about the nature of the mind and the ethics of creating artificial beings endowed with human-like intelligence which are issues that have been explored by myth, fiction and philosophy since antiquity.[19] Some people also consider AI to be a danger to humanity if it progresses unabatedly.[20] Others believe that AI, unlike previous technological revolutions, will create a risk of mass unemployment.[21]

In the twenty-first century, AI techniques have experienced a resurgence following concurrent advances in computer power, large amounts of data, and theoretical understanding; and AI techniques have become an essential part of the technology industry, helping to solve many challenging problems in computer science.[22][11]

Thought-capable artificial beings appeared as storytelling devices in antiquity,[23] and have been common in fiction, as in Mary Shelley’s Frankenstein or Karel apek’s R.U.R. (Rossum’s Universal Robots).[24] These characters and their fates raised many of the same issues now discussed in the ethics of artificial intelligence.[19]

The study of mechanical or “formal” reasoning began with philosophers and mathematicians in antiquity. The study of mathematical logic led directly to Alan Turing’s theory of computation, which suggested that a machine, by shuffling symbols as simple as “0” and “1”, could simulate any conceivable act of mathematical deduction. This insight, that digital computers can simulate any process of formal reasoning, is known as the ChurchTuring thesis.[25] Along with concurrent discoveries in neurobiology, information theory and cybernetics, this led researchers to consider the possibility of building an electronic brain. Turing proposed that “if a human could not distinguish between responses from a machine and a human, the machine could be considered intelligent”.[26] The first work that is now generally recognized as AI was McCullouch and Pitts’ 1943 formal design for Turing-complete “artificial neurons”.

The field of AI research was born at a workshop at Dartmouth College in 1956.[28] Attendees Allen Newell (CMU), Herbert Simon (CMU), John McCarthy (MIT), Marvin Minsky (MIT) and Arthur Samuel (IBM) became the founders and leaders of AI research.[29] They and their students produced programs that the press described as “astonishing”: computers were learning checkers strategies (c. 1954)[31] (and by 1959 were reportedly playing better than the average human),[32] solving word problems in algebra, proving logical theorems (Logic Theorist, first run c. 1956) and speaking English.[33] By the middle of the 1960s, research in the U.S. was heavily funded by the Department of Defense[34] and laboratories had been established around the world.[35] AI’s founders were optimistic about the future: Herbert Simon predicted, “machines will be capable, within twenty years, of doing any work a man can do”. Marvin Minsky agreed, writing, “within a generation… the problem of creating ‘artificial intelligence’ will substantially be solved”.[7]

They failed to recognize the difficulty of some of the remaining tasks. Progress slowed and in 1974, in response to the criticism of Sir James Lighthill and ongoing pressure from the US Congress to fund more productive projects, both the U.S. and British governments cut off exploratory research in AI. The next few years would later be called an “AI winter”,[9] a period when obtaining funding for AI projects was difficult.

In the early 1980s, AI research was revived by the commercial success of expert systems,[37] a form of AI program that simulated the knowledge and analytical skills of human experts. By 1985 the market for AI had reached over a billion dollars. At the same time, Japan’s fifth generation computer project inspired the U.S and British governments to restore funding for academic research.[8] However, beginning with the collapse of the Lisp Machine market in 1987, AI once again fell into disrepute, and a second, longer-lasting hiatus began.[10]

In the late 1990s and early 21st century, AI began to be used for logistics, data mining, medical diagnosis and other areas.[22] The success was due to increasing computational power (see Moore’s law), greater emphasis on solving specific problems, new ties between AI and other fields (such as statistics, economics and mathematics), and a commitment by researchers to mathematical methods and scientific standards.[38] Deep Blue became the first computer chess-playing system to beat a reigning world chess champion, Garry Kasparov on 11 May 1997.

In 2011, a Jeopardy! quiz show exhibition match, IBM’s question answering system, Watson, defeated the two greatest Jeopardy! champions, Brad Rutter and Ken Jennings, by a significant margin. Faster computers, algorithmic improvements, and access to large amounts of data enabled advances in machine learning and perception; data-hungry deep learning methods started to dominate accuracy benchmarks around 2012.[41] The Kinect, which provides a 3D bodymotion interface for the Xbox 360 and the Xbox One use algorithms that emerged from lengthy AI research[42] as do intelligent personal assistants in smartphones.[43] In March 2016, AlphaGo won 4 out of 5 games of Go in a match with Go champion Lee Sedol, becoming the first computer Go-playing system to beat a professional Go player without handicaps.[6][44] In the 2017 Future of Go Summit, AlphaGo won a three-game match with Ke Jie,[45] who at the time continuously held the world No. 1 ranking for two years.[46][47] This marked the completion of a significant milestone in the development of Artificial Intelligence as Go is an extremely complex game, more so than Chess.

According to Bloomberg’s Jack Clark, 2015 was a landmark year for artificial intelligence, with the number of software projects that use AI within Google increased from a “sporadic usage” in 2012 to more than 2,700 projects. Clark also presents factual data indicating that error rates in image processing tasks have fallen significantly since 2011.[48] He attributes this to an increase in affordable neural networks, due to a rise in cloud computing infrastructure and to an increase in research tools and datasets.[11] Other cited examples include Microsoft’s development of a Skype system that can automatically translate from one language to another and Facebook’s system that can describe images to blind people.[48] In a 2017 survey, one in five companies reported they had “incorporated AI in some offerings or processes”.[49][50]

A typical AI perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] An AI’s intended goal function can be simple (“1 if the AI wins a game of Go, 0 otherwise”) or complex (“Do actions mathematically similar to the actions that got you rewards in the past”). Goals can be explicitly defined, or can be induced. If the AI is programmed for “reinforcement learning”, goals can be implicitly induced by rewarding some types of behavior and punishing others.[a] Alternatively, an evolutionary system can induce goals by using a “fitness function” to mutate and preferentially replicate high-scoring AI systems; this is similar to how animals evolved to innately desire certain goals such as finding food, or how dogs can be bred via artificial selection to possess desired traits. Some AI systems, such as nearest-neighbor, instead reason by analogy; these systems are not generally given goals, except to the degree that goals are somehow implicit in their training data. Such systems can still be benchmarked if the non-goal system is framed as a system whose “goal” is to successfully accomplish its narrow classification task.[53]

AI often revolves around the use of algorithms. An algorithm is a set of unambiguous instructions that a mechanical computer can execute.[b] A complex algorithm is often built on top of other, simpler, algorithms. A simple example of an algorithm is the following recipe for optimal play at tic-tac-toe:

Many AI algorithms are capable of learning from data; they can enhance themselves by learning new heuristics (strategies, or “rules of thumb”, that have worked well in the past), or can themselves write other algorithms. Some of the “learners” described below, including Bayesian networks, decision trees, and nearest-neighbor, could theoretically, if given infinite data, time, and memory, learn to approximate any function, including whatever combination of mathematical functions would best describe the entire world. These learners could therefore, in theory, derive all possible knowledge, by considering every possible hypothesis and matching it against the data. In practice, it is almost never possible to consider every possibility, because of the phenomenon of “combinatorial explosion”, where the amount of time needed to solve a problem grows exponentially. Much of AI research involves figuring out how to identify and avoid considering broad swaths of possibilities that are unlikely to be fruitful.[55] For example, when viewing a map and looking for the shortest driving route from Denver to New York in the East, one can in most cases skip looking at any path through San Francisco or other areas far to the West; thus, an AI wielding an pathfinding algorithm like A* can avoid the combinatorial explosion that would ensue if every possible route had to be ponderously considered in turn.[57]

The earliest (and easiest to understand) approach to AI was symbolism (such as formal logic): “If an otherwise healthy adult has a fever, then they may have influenza”. A second, more general, approach is Bayesian inference: “If the current patient has a fever, adjust the probability they have influenza in such-and-such way”. The third major approach, extremely popular in routine business AI applications, are analogizers such as SVM and nearest-neighbor: “After examining the records of known past patients whose temperature, symptoms, age, and other factors mostly match the current patient, X% of those patients turned out to have influenza”. A fourth approach is harder to intuitively understand, but is inspired by how the brain’s machinery works: the artificial neural network approach uses artificial “neurons” that can learn by comparing itself to the desired output and altering the strengths of the connections between its internal neurons to “reinforce” connections that seemed to be useful. These four main approaches can overlap with each other and with evolutionary systems; for example, neural nets can learn to make inferences, to generalize, and to make analogies. Some systems implicitly or explicitly use multiple of these approaches, alongside many other AI and non-AI algorithms;[58] the best approach is often different depending on the problem.[60]

Learning algorithms work on the basis that strategies, algorithms, and inferences that worked well in the past are likely to continue working well in the future. These inferences can be obvious, such as “since the sun rose every morning for the last 10,000 days, it will probably rise tomorrow morning as well”. They can be nuanced, such as “X% of families have geographically separate species with color variants, so there is an Y% chance that undiscovered black swans exist”. Learners also work on the basis of “Occam’s razor”: The simplest theory that explains the data is the likeliest. Therefore, to be successful, a learner must be designed such that it prefers simpler theories to complex theories, except in cases where the complex theory is proven substantially better. Settling on a bad, overly complex theory gerrymandered to fit all the past training data is known as overfitting. Many systems attempt to reduce overfitting by rewarding a theory in accordance with how well it fits the data, but penalizing the theory in accordance with how complex the theory is. Besides classic overfitting, learners can also disappoint by “learning the wrong lesson”. A toy example is that an image classifier trained only on pictures of brown horses and black cats might conclude that all brown patches are likely to be horses. A real-world example is that, unlike humans, current image classifiers don’t determine the spatial relationship between components of the picture; instead, they learn abstract patterns of pixels that humans are oblivious to, but that linearly correlate with images of certain types of real objects. Faintly superimposing such a pattern on a legitimate image results in an “adversarial” image that the system misclassifies.[c][63][64][65]

Compared with humans, existing AI lacks several features of human “commonsense reasoning”; most notably, humans have powerful mechanisms for reasoning about “nave physics” such as space, time, and physical interactions. This enables even young children to easily make inferences like “If I roll this pen off a table, it will fall on the floor”. Humans also have a powerful mechanism of “folk psychology” that helps them to interpret natural-language sentences such as “The city councilmen refused the demonstrators a permit because they advocated violence”. (A generic AI has difficulty inferring whether the councilmen or the demonstrators are the ones alleged to be advocating violence.)[68][69][70] This lack of “common knowledge” means that AI often makes different mistakes than humans make, in ways that can seem incomprehensible. For example, existing self-driving cars cannot reason about the location nor the intentions of pedestrians in the exact way that humans do, and instead must use non-human modes of reasoning to avoid accidents.[71][72][73]

The overall research goal of artificial intelligence is to create technology that allows computers and machines to function in an intelligent manner. The general problem of simulating (or creating) intelligence has been broken down into sub-problems. These consist of particular traits or capabilities that researchers expect an intelligent system to display. The traits described below have received the most attention.[13]

Early researchers developed algorithms that imitated step-by-step reasoning that humans use when they solve puzzles or make logical deductions.[74] By the late 1980s and 1990s, AI research had developed methods for dealing with uncertain or incomplete information, employing concepts from probability and economics.[75]

These algorithms proved to be insufficient for solving large reasoning problems, because they experienced a “combinatorial explosion”: they became exponentially slower as the problems grew larger.[55] In fact, even humans rarely use the step-by-step deduction that early AI research was able to model. They solve most of their problems using fast, intuitive judgements.[76]

Knowledge representation[77] and knowledge engineering[78] are central to classical AI research. Some “expert systems” attempt to gather together explicit knowledge possessed by experts in some narrow domain. In addition, some projects attempt to gather the “commonsense knowledge” known to the average person into a database containing extensive knowledge about the world. Among the things a comprehensive commonsense knowledge base would contain are: objects, properties, categories and relations between objects;[79] situations, events, states and time;[80] causes and effects;[81] knowledge about knowledge (what we know about what other people know);[82] and many other, less well researched domains. A representation of “what exists” is an ontology: the set of objects, relations, concepts, and properties formally described so that software agents can interpret them. The semantics of these are captured as description logic concepts, roles, and individuals, and typically implemented as classes, properties, and individuals in the Web Ontology Language.[83] The most general ontologies are called upper ontologies, which attempt to provide a foundation for all other knowledge[84] by acting as mediators between domain ontologies that cover specific knowledge about a particular knowledge domain (field of interest or area of concern). Such formal knowledge representations can be used in content-based indexing and retrieval,[85] scene interpretation,[86] clinical decision support,[87] knowledge discovery (mining “interesting” and actionable inferences from large databases),[88] and other areas.[89]

Among the most difficult problems in knowledge representation are:

Intelligent agents must be able to set goals and achieve them.[96] They need a way to visualize the futurea representation of the state of the world and be able to make predictions about how their actions will change itand be able to make choices that maximize the utility (or “value”) of available choices.[97]

In classical planning problems, the agent can assume that it is the only system acting in the world, allowing the agent to be certain of the consequences of its actions.[98] However, if the agent is not the only actor, then it requires that the agent can reason under uncertainty. This calls for an agent that can not only assess its environment and make predictions, but also evaluate its predictions and adapt based on its assessment.[99]

Multi-agent planning uses the cooperation and competition of many agents to achieve a given goal. Emergent behavior such as this is used by evolutionary algorithms and swarm intelligence.[100]

Machine learning, a fundamental concept of AI research since the field’s inception,[101] is the study of computer algorithms that improve automatically through experience.[102][103]

Unsupervised learning is the ability to find patterns in a stream of input. Supervised learning includes both classification and numerical regression. Classification is used to determine what category something belongs in, after seeing a number of examples of things from several categories. Regression is the attempt to produce a function that describes the relationship between inputs and outputs and predicts how the outputs should change as the inputs change.[103] Both classifiers and regression learners can be viewed as “function approximators” trying to learn an unknown (possibly implicit) function; for example, a spam classifier can be viewed as learning a function that maps from the text of an email to one of two categories, “spam” or “not spam”. Computational learning theory can assess learners by computational complexity, by sample complexity (how much data is required), or by other notions of optimization.[104] In reinforcement learning[105] the agent is rewarded for good responses and punished for bad ones. The agent uses this sequence of rewards and punishments to form a strategy for operating in its problem space.

Natural language processing[106] (NLP) gives machines the ability to read and understand human language. A sufficiently powerful natural language processing system would enable natural-language user interfaces and the acquisition of knowledge directly from human-written sources, such as newswire texts. Some straightforward applications of natural language processing include information retrieval, text mining, question answering[107] and machine translation.[108] Many current approaches use word co-occurrence frequencies to construct syntactic representations of text. “Keyword spotting” strategies for search are popular and scalable but dumb; a search query for “dog” might only match documents with the literal word “dog” and miss a document with the word “poodle”. “Lexical affinity” strategies use the occurrence of words such as “accident” to assess the sentiment of a document. Modern statistical NLP approaches can combine all these strategies as well as others, and often achieve acceptable accuracy at the page or paragraph level, but continue to lack the semantic understanding required to classify isolated sentences well. Besides the usual difficulties with encoding semantic commonsense knowledge, existing semantic NLP sometimes scales too poorly to be viable in business applications. Beyond semantic NLP, the ultimate goal of “narrative” NLP is to embody a full understanding of commonsense reasoning.[109]

Machine perception[110] is the ability to use input from sensors (such as cameras (visible spectrum or infrared), microphones, wireless signals, and active lidar, sonar, radar, and tactile sensors) to deduce aspects of the world. Applications include speech recognition,[111] facial recognition, and object recognition.[112] Computer vision is the ability to analyze visual input. Such input is usually ambiguous; a giant, fifty-meter-tall pedestrian far away may produce exactly the same pixels as a nearby normal-sized pedestrian, requiring the AI to judge the relative likelihood and reasonableness of different interpretations, for example by using its “object model” to assess that fifty-meter pedestrians do not exist.[113]

AI is heavily used in robotics.[114] Advanced robotic arms and other industrial robots, widely used in modern factories, can learn from experience how to move efficiently despite the presence of friction and gear slippage.[115] A modern mobile robot, when given a small, static, and visible environment, can easily determine its location and map its environment; however, dynamic environments, such as (in endoscopy) the interior of a patient’s breathing body, pose a greater challenge. Motion planning is the process of breaking down a movement task into “primitives” such as individual joint movements. Such movement often involves compliant motion, a process where movement requires maintaining physical contact with an object.[117][118] Moravec’s paradox generalizes that low-level sensorimotor skills that humans take for granted are, counterintuitively, difficult to program into a robot; the paradox is named after Hans Moravec, who stated in 1988 that “it is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility”.[119][120] This is attributed to the fact that, unlike checkers, physical dexterity has been a direct target of natural selection for millions of years.[121]

Moravec’s paradox can be extended to many forms of social intelligence.[123][124] Distributed multi-agent coordination of autonomous vehicles remains a difficult problem.[125] Affective computing is an interdisciplinary umbrella that comprises systems which recognize, interpret, process, or simulate human affects. Moderate successes related to affective computing include textual sentiment analysis and, more recently, multimodal affect analysis (see multimodal sentiment analysis), wherein AI classifies the affects displayed by a videotaped subject.[129]

In the long run, social skills and an understanding of human emotion and game theory would be valuable to a social agent. Being able to predict the actions of others by understanding their motives and emotional states would allow an agent to make better decisions. Some computer systems mimic human emotion and expressions to appear more sensitive to the emotional dynamics of human interaction, or to otherwise facilitate humancomputer interaction.[130] Similarly, some virtual assistants are programmed to speak conversationally or even to banter humorously; this tends to give naive users an unrealistic conception of how intelligent existing computer agents actually are.[131]

Historically, projects such as the Cyc knowledge base (1984) and the massive Japanese Fifth Generation Computer Systems initiative (19821992) attempted to cover the breadth of human cognition. These early projects failed to escape the limitations of non-quantitative symbolic logic models and, in retrospect, greatly underestimated the difficulty of cross-domain AI. Nowadays, the vast majority of current AI researchers work instead on tractable “narrow AI” applications (such as medical diagnosis or automobile navigation).[132] Many researchers predict that such “narrow AI” work in different individual domains will eventually be incorporated into a machine with artificial general intelligence (AGI), combining most of the narrow skills mentioned in this article and at some point even exceeding human ability in most or all these areas.[17][133] Many advances have general, cross-domain significance. One high-profile example is that DeepMind in the 2010s developed a “generalized artificial intelligence” that could learn many diverse Atari games on its own, and later developed a variant of the system which succeeds at sequential learning.[134][135][136] Besides transfer learning,[137] hypothetical AGI breakthroughs could include the development of reflective architectures that can engage in decision-theoretic metareasoning, and figuring out how to “slurp up” a comprehensive knowledge base from the entire unstructured Web. Some argue that some kind of (currently-undiscovered) conceptually straightforward, but mathematically difficult, “Master Algorithm” could lead to AGI. Finally, a few “emergent” approaches look to simulating human intelligence extremely closely, and believe that anthropomorphic features like an artificial brain or simulated child development may someday reach a critical point where general intelligence emerges.[139][140]

Many of the problems in this article may also require general intelligence, if machines are to solve the problems as well as people do. For example, even specific straightforward tasks, like machine translation, require that a machine read and write in both languages (NLP), follow the author’s argument (reason), know what is being talked about (knowledge), and faithfully reproduce the author’s original intent (social intelligence). A problem like machine translation is considered “AI-complete”, because all of these problems need to be solved simultaneously in order to reach human-level machine performance.

There is no established unifying theory or paradigm that guides AI research. Researchers disagree about many issues.[141] A few of the most long standing questions that have remained unanswered are these: should artificial intelligence simulate natural intelligence by studying psychology or neurobiology? Or is human biology as irrelevant to AI research as bird biology is to aeronautical engineering?[14]Can intelligent behavior be described using simple, elegant principles (such as logic or optimization)? Or does it necessarily require solving a large number of completely unrelated problems?[15]

In the 1940s and 1950s, a number of researchers explored the connection between neurobiology, information theory, and cybernetics. Some of them built machines that used electronic networks to exhibit rudimentary intelligence, such as W. Grey Walter’s turtles and the Johns Hopkins Beast. Many of these researchers gathered for meetings of the Teleological Society at Princeton University and the Ratio Club in England.[142] By 1960, this approach was largely abandoned, although elements of it would be revived in the 1980s.

When access to digital computers became possible in the middle 1950s, AI research began to explore the possibility that human intelligence could be reduced to symbol manipulation. The research was centered in three institutions: Carnegie Mellon University, Stanford and MIT, and as described below, each one developed its own style of research. John Haugeland named these symbolic approaches to AI “good old fashioned AI” or “GOFAI”.[143] During the 1960s, symbolic approaches had achieved great success at simulating high-level thinking in small demonstration programs. Approaches based on cybernetics or artificial neural networks were abandoned or pushed into the background.[144]Researchers in the 1960s and the 1970s were convinced that symbolic approaches would eventually succeed in creating a machine with artificial general intelligence and considered this the goal of their field.

Economist Herbert Simon and Allen Newell studied human problem-solving skills and attempted to formalize them, and their work laid the foundations of the field of artificial intelligence, as well as cognitive science, operations research and management science. Their research team used the results of psychological experiments to develop programs that simulated the techniques that people used to solve problems. This tradition, centered at Carnegie Mellon University would eventually culminate in the development of the Soar architecture in the middle 1980s.[145][146]

Unlike Simon and Newell, John McCarthy felt that machines did not need to simulate human thought, but should instead try to find the essence of abstract reasoning and problem-solving, regardless of whether people used the same algorithms.[14] His laboratory at Stanford (SAIL) focused on using formal logic to solve a wide variety of problems, including knowledge representation, planning and learning.[147] Logic was also the focus of the work at the University of Edinburgh and elsewhere in Europe which led to the development of the programming language Prolog and the science of logic programming.[148]

Researchers at MIT (such as Marvin Minsky and Seymour Papert)[149] found that solving difficult problems in vision and natural language processing required ad-hoc solutions they argued that there was no simple and general principle (like logic) that would capture all the aspects of intelligent behavior. Roger Schank described their “anti-logic” approaches as “scruffy” (as opposed to the “neat” paradigms at CMU and Stanford).[15] Commonsense knowledge bases (such as Doug Lenat’s Cyc) are an example of “scruffy” AI, since they must be built by hand, one complicated concept at a time.[150]

When computers with large memories became available around 1970, researchers from all three traditions began to build knowledge into AI applications.[151] This “knowledge revolution” led to the development and deployment of expert systems (introduced by Edward Feigenbaum), the first truly successful form of AI software.[37] The knowledge revolution was also driven by the realization that enormous amounts of knowledge would be required by many simple AI applications.

By the 1980s progress in symbolic AI seemed to stall and many believed that symbolic systems would never be able to imitate all the processes of human cognition, especially perception, robotics, learning and pattern recognition. A number of researchers began to look into “sub-symbolic” approaches to specific AI problems.[16] Sub-symbolic methods manage to approach intelligence without specific representations of knowledge.

This includes embodied, situated, behavior-based, and nouvelle AI. Researchers from the related field of robotics, such as Rodney Brooks, rejected symbolic AI and focused on the basic engineering problems that would allow robots to move and survive.[152] Their work revived the non-symbolic viewpoint of the early cybernetics researchers of the 1950s and reintroduced the use of control theory in AI. This coincided with the development of the embodied mind thesis in the related field of cognitive science: the idea that aspects of the body (such as movement, perception and visualization) are required for higher intelligence.

Within developmental robotics, developmental learning approaches are elaborated upon to allow robots to accumulate repertoires of novel skills through autonomous self-exploration, social interaction with human teachers, and the use of guidance mechanisms (active learning, maturation, motor synergies, etc.).[153][154]

Interest in neural networks and “connectionism” was revived by David Rumelhart and others in the middle of the 1980s.[157] Artificial neural networks are an example of soft computing — they are solutions to problems which cannot be solved with complete logical certainty, and where an approximate solution is often sufficient. Other soft computing approaches to AI include fuzzy systems, evolutionary computation and many statistical tools. The application of soft computing to AI is studied collectively by the emerging discipline of computational intelligence.[158]

Much of traditional GOFAI got bogged down on ad hoc patches to symbolic computation that worked on their own toy models but failed to generalize to real-world results. However, around the 1990s, AI researchers adopted sophisticated mathematical tools, such as hidden Markov models (HMM), information theory, and normative Bayesian decision theory to compare or to unify competing architectures. The shared mathematical language permitted a high level of collaboration with more established fields (like mathematics, economics or operations research).[d] Compared with GOFAI, new “statistical learning” techniques such as HMM and neural networks were gaining higher levels of accuracy in many practical domains such as data mining, without necessarily acquiring semantic understanding of the datasets. The increased successes with real-world data led to increasing emphasis on comparing different approaches against shared test data to see which approach performed best in a broader context than that provided by idiosyncratic toy models; AI research was becoming more scientific. Nowadays results of experiments are often rigorously measurable, and are sometimes (with difficulty) reproducible.[38][159] Different statistical learning techniques have different limitations; for example, basic HMM cannot model the infinite possible combinations of natural language. Critics note that the shift from GOFAI to statistical learning is often also a shift away from Explainable AI. In AGI research, some scholars caution against over-reliance on statistical learning, and argue that continuing research into GOFAI will still be necessary to attain general intelligence.

AI has developed a large number of tools to solve the most difficult problems in computer science. A few of the most general of these methods are discussed below.

Many problems in AI can be solved in theory by intelligently searching through many possible solutions:[168] Reasoning can be reduced to performing a search. For example, logical proof can be viewed as searching for a path that leads from premises to conclusions, where each step is the application of an inference rule.[169] Planning algorithms search through trees of goals and subgoals, attempting to find a path to a target goal, a process called means-ends analysis.[170] Robotics algorithms for moving limbs and grasping objects use local searches in configuration space.[115] Many learning algorithms use search algorithms based on optimization.

Simple exhaustive searches[171] are rarely sufficient for most real-world problems: the search space (the number of places to search) quickly grows to astronomical numbers. The result is a search that is too slow or never completes. The solution, for many problems, is to use “heuristics” or “rules of thumb” that prioritize choices in favor of those that are more likely to reach a goal and to do so in a shorter number of steps. In some search methodologies heuristics can also serve to entirely eliminate some choices that are unlikely to lead to a goal (called “pruning the search tree”). Heuristics supply the program with a “best guess” for the path on which the solution lies.[172] Heuristics limit the search for solutions into a smaller sample size.

A very different kind of search came to prominence in the 1990s, based on the mathematical theory of optimization. For many problems, it is possible to begin the search with some form of a guess and then refine the guess incrementally until no more refinements can be made. These algorithms can be visualized as blind hill climbing: we begin the search at a random point on the landscape, and then, by jumps or steps, we keep moving our guess uphill, until we reach the top. Other optimization algorithms are simulated annealing, beam search and random optimization.[173]

Evolutionary computation uses a form of optimization search. For example, they may begin with a population of organisms (the guesses) and then allow them to mutate and recombine, selecting only the fittest to survive each generation (refining the guesses). Classic evolutionary algorithms include genetic algorithms, gene expression programming, and genetic programming.[174] Alternatively, distributed search processes can coordinate via swarm intelligence algorithms. Two popular swarm algorithms used in search are particle swarm optimization (inspired by bird flocking) and ant colony optimization (inspired by ant trails).[175][176]

Logic[177] is used for knowledge representation and problem solving, but it can be applied to other problems as well. For example, the satplan algorithm uses logic for planning[178] and inductive logic programming is a method for learning.[179]

Several different forms of logic are used in AI research. Propositional logic[180] involves truth functions such as “or” and “not”. First-order logic[181] adds quantifiers and predicates, and can express facts about objects, their properties, and their relations with each other. Fuzzy set theory assigns a “degree of truth” (between 0 and 1) to vague statements such as “Alice is old” (or rich, or tall, or hungry) that are too linguistically imprecise to be completely true or false. Fuzzy logic is successfully used in control systems to allow experts to contribute vague rules such as “if you are close to the destination station and moving fast, increase the train’s brake pressure”; these vague rules can then be numerically refined within the system. Fuzzy logic fails to scale well in knowledge bases; many AI researchers question the validity of chaining fuzzy-logic inferences.[e][183][184]

Default logics, non-monotonic logics and circumscription[91] are forms of logic designed to help with default reasoning and the qualification problem. Several extensions of logic have been designed to handle specific domains of knowledge, such as: description logics;[79] situation calculus, event calculus and fluent calculus (for representing events and time);[80] causal calculus;[81] belief calculus;[185] and modal logics.[82]

Overall, qualitiative symbolic logic is brittle and scales poorly in the presence of noise or other uncertainty. Exceptions to rules are numerous, and it is difficult for logical systems to function in the presence of contradictory rules.[187]

Many problems in AI (in reasoning, planning, learning, perception, and robotics) require the agent to operate with incomplete or uncertain information. AI researchers have devised a number of powerful tools to solve these problems using methods from probability theory and economics.[188]

Bayesian networks[189] are a very general tool that can be used for a large number of problems: reasoning (using the Bayesian inference algorithm),[190] learning (using the expectation-maximization algorithm),[f][192] planning (using decision networks)[193] and perception (using dynamic Bayesian networks).[194] Probabilistic algorithms can also be used for filtering, prediction, smoothing and finding explanations for streams of data, helping perception systems to analyze processes that occur over time (e.g., hidden Markov models or Kalman filters).[194] Compared with symbolic logic, formal Bayesian inference is computationally expensive. For inference to be tractable, most observations must be conditionally independent of one another. Complicated graphs with diamonds or other “loops” (undirected cycles) can require a sophisticated method such as Markov Chain Monte Carlo, which spreads an ensemble of random walkers throughout the Bayesian network and attempts to converge to an assessment of the conditional probabilities. Bayesian networks are used on XBox Live to rate and match players; wins and losses are “evidence” of how good a player is. AdSense uses a Bayesian network with over 300 million edges to learn which ads to serve.

A key concept from the science of economics is “utility”: a measure of how valuable something is to an intelligent agent. Precise mathematical tools have been developed that analyze how an agent can make choices and plan, using decision theory, decision analysis,[196] and information value theory.[97] These tools include models such as Markov decision processes,[197] dynamic decision networks,[194] game theory and mechanism design.[198]

The simplest AI applications can be divided into two types: classifiers (“if shiny then diamond”) and controllers (“if shiny then pick up”). Controllers do, however, also classify conditions before inferring actions, and therefore classification forms a central part of many AI systems. Classifiers are functions that use pattern matching to determine a closest match. They can be tuned according to examples, making them very attractive for use in AI. These examples are known as observations or patterns. In supervised learning, each pattern belongs to a certain predefined class. A class can be seen as a decision that has to be made. All the observations combined with their class labels are known as a data set. When a new observation is received, that observation is classified based on previous experience.[199]

A classifier can be trained in various ways; there are many statistical and machine learning approaches. The decision tree[200] is perhaps the most widely used machine learning algorithm. Other widely used classifiers are the neural network,[202]k-nearest neighbor algorithm,[g][204]kernel methods such as the support vector machine (SVM),[h][206]Gaussian mixture model,[207] the extremely popular naive Bayes classifier[i][209] and improved version of decision tree – decision stream.[210] Classifier performance depends greatly on the characteristics of the data to be classified, such as the dataset size, the dimensionality, and the level of noise. Model-based classifiers perform well if the assumed model is an extremely good fit for the actual data. Otherwise, if no matching model is available, and if accuracy (rather than speed or scalability) is the sole concern, conventional wisdom is that discriminative classifiers (especially SVM) tend to be more accurate than model-based classifiers such as “naive Bayes” on most practical data sets.[211]

Neural networks, or neural nets, were inspired by the architecture of neurons in the human brain. A simple “neuron” N accepts input from multiple other neurons, each of which, when activated (or “fired”), cast a weighted “vote” for or against whether neuron N should itself activate. Learning requires an algorithm to adjust these weights based on the training data; one simple algorithm (dubbed “fire together, wire together”) is to increase the weight between two connected neurons when the activation of one triggers the successful activation of another. The net forms “concepts” that are distributed among a subnetwork of shared[j] neurons that tend to fire together; a concept meaning “leg” might be coupled with a subnetwork meaning “foot” that includes the sound for “foot”. Neurons have a continuous spectrum of activation; in addition, neurons can process inputs in a nonlinear way rather than weighing straightforward votes. Modern neural nets can learn both continuous functions and, surprisingly, digital logical operations. Neural networks’ early successes included predicting the stock market and (in 1995) a mostly self-driving car.[k] In the 2010s, advances in neural networks using deep learning thrust AI into widespread public consciousness and contributed to an enormous upshift in corporate AI spending; for example, AI-related M&A in 2017 was over 25 times as large as in 2015.[214][215]

The study of non-learning artificial neural networks[202] began in the decade before the field of AI research was founded, in the work of Walter Pitts and Warren McCullouch. Frank Rosenblatt invented the perceptron, a learning network with a single layer, similar to the old concept of linear regression. Early pioneers also include Alexey Grigorevich Ivakhnenko, Teuvo Kohonen, Stephen Grossberg, Kunihiko Fukushima, Christoph von der Malsburg, David Willshaw, Shun-Ichi Amari, Bernard Widrow, John Hopfield, Eduardo R. Caianiello, and others.

The main categories of networks are acyclic or feedforward neural networks (where the signal passes in only one direction) and recurrent neural networks (which allow feedback and short-term memories of previous input events). Among the most popular feedforward networks are perceptrons, multi-layer perceptrons and radial basis networks.[216] Neural networks can be applied to the problem of intelligent control (for robotics) or learning, using such techniques as Hebbian learning (“fire together, wire together”), GMDH or competitive learning.[217]

Today, neural networks are often trained by the backpropagation algorithm, which had been around since 1970 as the reverse mode of automatic differentiation published by Seppo Linnainmaa,[218][219] and was introduced to neural networks by Paul Werbos.[220][221][222]

Hierarchical temporal memory is an approach that models some of the structural and algorithmic properties of the neocortex.[223]

In short, most neural networks use some form of gradient descent on a hand-created neural topology. However, some research groups, such as Uber, argue that simple neuroevolution to mutate new neural network topologies and weights may be competitive with sophisticated gradient descent approaches. One advantage of neuroevolution is that it may be less prone to get caught in “dead ends”.[224]

Deep learning is any artificial neural network that can learn a long chain of causal links. For example, a feedforward network with six hidden layers can learn a seven-link causal chain (six hidden layers + output layer) and has a “credit assignment path” (CAP) depth of seven. Many deep learning systems need to be able to learn chains ten or more causal links in length.[225] Deep learning has transformed many important subfields of artificial intelligence, including computer vision, speech recognition, natural language processing and others.[226][227][225]

According to one overview,[228] the expression “Deep Learning” was introduced to the Machine Learning community by Rina Dechter in 1986[229] and gained traction afterIgor Aizenberg and colleagues introduced it to Artificial Neural Networks in 2000.[230] The first functional Deep Learning networks were published by Alexey Grigorevich Ivakhnenko and V. G. Lapa in 1965.[231][pageneeded] These networks are trained one layer at a time. Ivakhnenko’s 1971 paper[232] describes the learning of a deep feedforward multilayer perceptron with eight layers, already much deeper than many later networks. In 2006, a publication by Geoffrey Hinton and Ruslan Salakhutdinov introduced another way of pre-training many-layered feedforward neural networks (FNNs) one layer at a time, treating each layer in turn as an unsupervised restricted Boltzmann machine, then using supervised backpropagation for fine-tuning. Similar to shallow artificial neural networks, deep neural networks can model complex non-linear relationships. Over the last few years, advances in both machine learning algorithms and computer hardware have led to more efficient methods for training deep neural networks that contain many layers of non-linear hidden units and a very large output layer.[234]

Deep learning often uses convolutional neural networks (CNNs), whose origins can be traced back to the Neocognitron introduced by Kunihiko Fukushima in 1980.[235] In 1989, Yann LeCun and colleagues applied backpropagation to such an architecture. In the early 2000s, in an industrial application CNNs already processed an estimated 10% to 20% of all the checks written in the US.[236]Since 2011, fast implementations of CNNs on GPUs havewon many visual pattern recognition competitions.[225]

CNNs with 12 convolutional layers were used in conjunction with reinforcement learning by Deepmind’s “AlphaGo Lee”, the program that beat a top Go champion in 2016.[237]

Early on, deep learning was also applied to sequence learning with recurrent neural networks (RNNs)[238] which are in theory Turing complete[239] and can run arbitrary programs to process arbitrary sequences of inputs. The depth of an RNN is unlimited and depends on the length of its input sequence; thus, an RNN is an example of deep learning.[225] RNNs can be trained by gradient descent[240][241][242] but suffer from the vanishing gradient problem.[226][243] In 1992, it was shown that unsupervised pre-training of a stack of recurrent neural networks can speed up subsequent supervised learning of deep sequential problems.[244]

Numerous researchers now use variants of a deep learning recurrent NN called the long short-term memory (LSTM) network published by Hochreiter & Schmidhuber in 1997.[245] LSTM is often trained by Connectionist Temporal Classification (CTC).[246] At Google, Microsoft and Baidu this approach has revolutionised speech recognition.[247][248][249] For example, in 2015, Google’s speech recognition experienced a dramatic performance jump of 49% through CTC-trained LSTM, which is now available through Google Voice to billions of smartphone users.[250] Google also used LSTM to improve machine translation,[251] Language Modeling[252] and Multilingual Language Processing.[253] LSTM combined with CNNs also improved automatic image captioning[254] and a plethora of other applications.

AI, like electricity or the steam engine, is a general purpose technology. There is no consensus on how to characterize which tasks AI tends to excel at.[255] While projects such as AlphaZero have succeeded in generating their own knowledge from scratch, many other machine learning projects require large training datasets.[256][257] Researcher Andrew Ng has suggested, as a “highly imperfect rule of thumb”, that “almost anything a typical human can do with less than one second of mental thought, we can probably now or in the near future automate using AI.”[258] Moravec’s paradox suggests that AI lags humans at many tasks that the human brain has specifically evolved to perform well.[121]

Games provide a well-publicized benchmark for assessing rates of progress. AlphaGo around 2016 brought the era of classical board-game benchmarks to an close. Games of imperfect knowledge provide new challenges to AI in the area of game theory.[259][260] E-sports such as StarCraft continue to provide additional public benchmarks.[261][262] There are many competitions and prizes, such as the Imagenet Challenge, to promote research in artificial intelligence. The main areas of competition include general machine intelligence, conversational behavior, data-mining, robotic cars, and robot soccer as well as conventional games.[citation needed]

The “imitation game” (an interpretation of the 1950 Turing test that assesses whether a computer can imitate a human) is nowadays considered too exploitable to be a meaningful benchmark.[263] A derivative of the Turing test is the Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA). As the name implies, this helps to determine that a user is an actual person and not a computer posing as a human. In contrast to the standard Turing test, CAPTCHA is administered by a machine and targeted to a human as opposed to being administered by a human and targeted to a machine. A computer asks a user to complete a simple test then generates a grade for that test. Computers are unable to solve the problem, so correct solutions are deemed to be the result of a person taking the test. A common type of CAPTCHA is the test that requires the typing of distorted letters, numbers or symbols that appear in an image undecipherable by a computer.

Proposed “universal intelligence” tests aim to compare how well machines, humans, and even non-human animals perform on problem sets that are generic as possible. At an extreme, the test suite can contain every possible problem, weighted by Kolmogorov complexity; unfortunately, these problem sets tend to be dominated by impoverished pattern-matching exercises where a tuned AI can easily exceed human performance levels.[265][266]

AI is relevant to any intellectual task. Modern artificial intelligence techniques are pervasive and are too numerous to list here. Frequently, when a technique reaches mainstream use, it is no longer considered artificial intelligence; this phenomenon is described as the AI effect.

High-profile examples of AI include autonomous vehicles (such as drones and self-driving cars), medical diagnosis, creating art (such as poetry), proving mathematical theorems, playing games (such as Chess or Go), search engines (such as Google search), online assistants (such as Siri), image recognition in photographs, spam filtering, prediction of judicial decisions[269] and targeting online advertisements.[270][271]

With social media sites overtaking TV as a source for news for young people and news organisations increasingly reliant on social media platforms for generating distribution,[272] major publishers now use artificial intelligence (AI) technology to post stories more effectively and generate higher volumes of traffic.[273]

Artificial intelligence is breaking into the healthcare industry by assisting doctors. According to Bloomberg Technology, Microsoft has developed AI to help doctors find the right treatments for cancer.[274] There is a great amount of research and drugs developed relating to cancer. In detail, there are more than 800 medicines and vaccines to treat cancer. This negatively affects the doctors, because there are too many options to choose from, making it more difficult to choose the right drugs for the patients. Microsoft is working on a project to develop a machine called “Hanover”. Its goal is to memorize all the papers necessary to cancer and help predict which combinations of drugs will be most effective for each patient. One project that is being worked on at the moment is fighting myeloid leukemia, a fatal cancer where the treatment has not improved in decades. Another study was reported to have found that artificial intelligence was as good as trained doctors in identifying skin cancers.[275] Another study is using artificial intelligence to try and monitor multiple high-risk patients, and this is done by asking each patient numerous questions based on data acquired from live doctor to patient interactions.[276]

According to CNN, a recent study by surgeons at the Children’s National Medical Center in Washington successfully demonstrated surgery with an autonomous robot. The team supervised the robot while it performed soft-tissue surgery, stitching together a pig’s bowel during open surgery, and doing so better than a human surgeon, the team claimed.[277] IBM has created its own artificial intelligence computer, the IBM Watson, which has beaten human intelligence (at some levels). Watson not only won at the game show Jeopardy! against former champions,[278] but was declared a hero after successfully diagnosing a woman who was suffering from leukemia.[279]

Advancements in AI have contributed to the growth of the automotive industry through the creation and evolution of self-driving vehicles. As of 2016, there are over 30 companies utilizing AI into the creation of driverless cars. A few companies involved with AI include Tesla, Google, and Apple.[280]

Many components contribute to the functioning of self-driving cars. These vehicles incorporate systems such as braking, lane changing, collision prevention, navigation and mapping. Together, these systems, as well as high performance computers, are integrated into one complex vehicle.[281]

Recent developments in autonomous automobiles have made the innovation of self-driving trucks possible, though they are still in the testing phase. The UK government has passed legislation to begin testing of self-driving truck platoons in 2018.[282] Self-driving truck platoons are a fleet of self-driving trucks following the lead of one non-self-driving truck, so the truck platoons aren’t entirely autonomous yet. Meanwhile, the Daimler, a German automobile corporation, is testing the Freightliner Inspiration which is a semi-autonomous truck that will only be used on the highway.[283]

One main factor that influences the ability for a driver-less automobile to function is mapping. In general, the vehicle would be pre-programmed with a map of the area being driven. This map would include data on the approximations of street light and curb heights in order for the vehicle to be aware of its surroundings. However, Google has been working on an algorithm with the purpose of eliminating the need for pre-programmed maps and instead, creating a device that would be able to adjust to a variety of new surroundings.[284] Some self-driving cars are not equipped with steering wheels or brake pedals, so there has also been research focused on creating an algorithm that is capable of maintaining a safe environment for the passengers in the vehicle through awareness of speed and driving conditions.[285]

Another factor that is influencing the ability for a driver-less automobile is the safety of the passenger. To make a driver-less automobile, engineers must program it to handle high-risk situations. These situations could include a head-on collision with pedestrians. The car’s main goal should be to make a decision that would avoid hitting the pedestrians and saving the passengers in the car. But there is a possibility the car would need to make a decision that would put someone in danger. In other words, the car would need to decide to save the pedestrians or the passengers.[286] The programing of the car in these situations is crucial to a successful driver-less automobile.

Financial institutions have long used artificial neural network systems to detect charges or claims outside of the norm, flagging these for human investigation. The use of AI in banking can be traced back to 1987 when Security Pacific National Bank in US set-up a Fraud Prevention Task force to counter the unauthorised use of debit cards. Programs like Kasisto and Moneystream are using AI in financial services.

Banks use artificial intelligence systems today to organize operations, maintain book-keeping, invest in stocks, and manage properties. AI can react to changes overnight or when business is not taking place.[287] In August 2001, robots beat humans in a simulated financial trading competition.[288] AI has also reduced fraud and financial crimes by monitoring behavioral patterns of users for any abnormal changes or anomalies.[289]

The use of AI machines in the market in applications such as online trading and decision making has changed major economic theories.[290] For example, AI based buying and selling platforms have changed the law of supply and demand in that it is now possible to easily estimate individualized demand and supply curves and thus individualized pricing. Furthermore, AI machines reduce information asymmetry in the market and thus making markets more efficient while reducing the volume of trades. Furthermore, AI in the markets limits the consequences of behavior in the markets again making markets more efficient. Other theories where AI has had impact include in rational choice, rational expectations, game theory, Lewis turning point, portfolio optimization and counterfactual thinking.

In video games, artificial intelligence is routinely used to generate dynamic purposeful behavior in non-player characters (NPCs). In addition, well-understood AI techniques are routinely used for pathfinding. Some researchers consider NPC AI in games to be a “solved problem” for most production tasks. Games with more atypical AI include the AI director of Left 4 Dead (2008) and the neuroevolutionary training of platoons in Supreme Commander 2 (2010).[291][292]

Worldwide annual military spending on robotics rose from 5.1 billion USD in 2010 to 7.5 billion USD in 2015.[293][294] Military drones capable of autonomous action are widely considered a useful asset. In 2017, Vladimir Putin stated that “Whoever becomes the leader in (artificial intelligence) will become the ruler of the world”.[295][296] Many artificial intelligence researchers seek to distance themselves from military applications of AI.[297]

For financial statements audit, AI makes continuous audit possible. AI tools could analyze many sets of different information immediately. The potential benefit would be the overall audit risk will be reduced, the level of assurance will be increased and the time duration of audit will be reduced.[298]

A report by the Guardian newspaper in the UK in 2018 found that online gambling companies were using AI to predict the behavior of customers in order to target them with personalized promotions.[299] Developers of commercial AI platforms are also beginning to appeal more directly to casino operators, offering a range of existing and potential services to help them boost their profits and expand their customer base.[300]

There are three philosophical questions related to AI:

Can a machine be intelligent? Can it “think”?

Widespread use of artificial intelligence could have unintended consequences that are dangerous or undesirable. Scientists from the Future of Life Institute, among others, described some short-term research goals to see how AI influences the economy, the laws and ethics that are involved with AI and how to minimize AI security risks. In the long-term, the scientists have proposed to continue optimizing function while minimizing possible security risks that come along with new technologies.[311]

The rest is here:

Artificial intelligence – Wikipedia

Online Artificial Intelligence Courses | Microsoft …

The Microsoft Professional Program (MPP) is a collection of courses that teach skills in several core technology tracks that help you excel in the industry’s newest job roles.

These courses are created and taught by experts and feature quizzes, hands-on labs, and engaging communities. For each track you complete, you earn a certificate of completion from Microsoft proving that you mastered those skills.

Excerpt from:

Online Artificial Intelligence Courses | Microsoft …

History of artificial intelligence – Wikipedia

The history of Artificial Intelligence (AI) began in antiquity, with myths, stories and rumors of artificial beings endowed with intelligence or consciousness by master craftsmen; as Pamela McCorduck writes, AI began with “an ancient wish to forge the gods.”

The seeds of modern AI were planted by classical philosophers who attempted to describe the process of human thinking as the mechanical manipulation of symbols. This work culminated in the invention of the programmable digital computer in the 1940s, a machine based on the abstract essence of mathematical reasoning. This device and the ideas behind it inspired a handful of scientists to begin seriously discussing the possibility of building an electronic brain.

The field of AI research was founded at a workshop held on the campus of Dartmouth College during the summer of 1956. Those who attended would become the leaders of AI research for decades. Many of them predicted that a machine as intelligent as a human being would exist in no more than a generation and they were given millions of dollars to make this vision come true.

Eventually it became obvious that they had grossly underestimated the difficulty of the project due to computer hardware limitations. In 1973, in response to the criticism of James Lighthill and ongoing pressure from congress, the U.S. and British Governments stopped funding undirected research into artificial intelligence, and the difficult years that followed would later be known as an “AI winter”. Seven years later, a visionary initiative by the Japanese Government inspired governments and industry to provide AI with billions of dollars, but by the late 80s the investors became disillusioned by the absence of the needed computer power (hardware) and withdrew funding again.

Investment and interest in AI boomed in the first decades of the 21st century, when machine learning was successfully applied to many problems in academia and industry due to the presence of powerful computer hardware. As in previous “AI summers”, some observers (such as Ray Kurzweil) predicted the imminent arrival of artificial general intelligence: a machine with intellectual capabilities that exceed the abilities of human beings.

The dream of artificial intelligence was first thought of in Indian philosophies like those of Charvaka, a famous philosophy tradition dating back to 1500 BCE and surviving documents around 600 BCE. McCorduck (2004) writes “artificial intelligence in one form or another is an idea that has pervaded intellectual history, a dream in urgent need of being realized,” expressed in humanity’s myths, legends, stories, speculation and clockwork automatons.

Mechanical men and artificial beings appear in Greek myths, such as the golden robots of Hephaestus and Pygmalion’s Galatea.[4]In the Middle Ages, there were rumors of secret mystical or alchemical means of placing mind into matter, such as Jbir ibn Hayyn’s Takwin, Paracelsus’ homunculus and Rabbi Judah Loew’s Golem.[5]By the 19th century, ideas about artificial men and thinking machines were developed in fiction, as in Mary Shelley’s Frankenstein or Karel apek’s R.U.R. (Rossum’s Universal Robots),and speculation, such as Samuel Butler’s “Darwin among the Machines.”AI has continued to be an important element of science fiction into the present.

Realistic humanoid automatons were built by craftsman from every civilization, including Yan Shi,[8]Hero of Alexandria,[9]Al-Jazari, Pierre Jaquet-Droz,and Wolfgang von Kempelen.[11]The oldest known automatons were the sacred statues of ancient Egypt and Greece. The faithful believed that craftsman had imbued these figures with very real minds, capable of wisdom and emotionHermes Trismegistus wrote that “by discovering the true nature of the gods, man has been able to reproduce it.”[12][13]

Artificial intelligence is based on the assumption that the process of human thought can be mechanized. The study of mechanicalor “formal”reasoning has a long history. Chinese, Indian and Greek philosophers all developed structured methods of formal deduction in the first millennium BCE. Their ideas were developed over the centuries by philosophers such as Aristotle (who gave a formal analysis of the syllogism), Euclid (whose Elements was a model of formal reasoning), al-Khwrizm (who developed algebra and gave his name to “algorithm”) and European scholastic philosophers such as William of Ockham and Duns Scotus.[14]

Majorcan philosopher Ramon Llull (12321315) developed several logical machines devoted to the production of knowledge by logical means;[15] Llull described his machines as mechanical entities that could combine basic and undeniable truths by simple logical operations, produced by the machine by mechanical meanings, in such ways as to produce all the possible knowledge.[16] Llull’s work had a great influence on Gottfried Leibniz, who redeveloped his ideas.[17]

In the 17th century, Leibniz, Thomas Hobbes and Ren Descartes explored the possibility that all rational thought could be made as systematic as algebra or geometry.[18]Hobbes famously wrote in Leviathan: “reason is nothing but reckoning”.[19]Leibniz envisioned a universal language of reasoning (his characteristica universalis) which would reduce argumentation to calculation, so that “there would be no more need of disputation between two philosophers than between two accountants. For it would suffice to take their pencils in hand, down to their slates, and to say each other (with a friend as witness, if they liked): Let us calculate.”[20]These philosophers had begun to articulate the physical symbol system hypothesis that would become the guiding faith of AI research.

In the 20th century, the study of mathematical logic provided the essential breakthrough that made artificial intelligence seem plausible. The foundations had been set by such works as Boole’s The Laws of Thought and Frege’s Begriffsschrift. Building on Frege’s system, Russell and Whitehead presented a formal treatment of the foundations of mathematics in their masterpiece, the Principia Mathematica in 1913. Inspired by Russell’s success, David Hilbert challenged mathematicians of the 1920s and 30s to answer this fundamental question: “can all of mathematical reasoning be formalized?”[14]His question was answered by Gdel’s incompleteness proof, Turing’s machine and Church’s Lambda calculus.[14][21]

Their answer was surprising in two ways. First, they proved that there were, in fact, limits to what mathematical logic could accomplish. But second (and more important for AI) their work suggested that, within these limits, any form of mathematical reasoning could be mechanized. The Church-Turing thesis implied that a mechanical device, shuffling symbols as simple as 0 and 1, could imitate any conceivable process of mathematical deduction. The key insight was the Turing machinea simple theoretical construct that captured the essence of abstract symbol manipulation. This invention would inspire a handful of scientists to begin discussing the possibility of thinking machines.[14][23]

Calculating machines were built in antiquity and improved throughout history by many mathematicians, including (once again) philosopher Gottfried Leibniz. In the early 19th century, Charles Babbage designed a programmable computer (the Analytical Engine), although it was never built. Ada Lovelace speculated that the machine “might compose elaborate and scientific pieces of music of any degree of complexity or extent”.[24] (She is often credited as the first programmer because of a set of notes she wrote that completely detail a method for calculating Bernoulli numbers with the Engine.)

The first modern computers were the massive code breaking machines of the Second World War (such as Z3, ENIAC and Colossus). The latter two of these machines were based on the theoretical foundation laid by Alan Turing[25] and developed by John von Neumann.[26]

In the 1940s and 50s, a handful of scientists from a variety of fields (mathematics, psychology, engineering, economics and political science) began to discuss the possibility of creating an artificial brain. The field of artificial intelligence research was founded as an academic discipline in 1956.

The earliest research into thinking machines was inspired by a confluence of ideas that became prevalent in the late 1930s, 1940s, and early 1950s. Recent research in neurology had shown that the brain was an electrical network of neurons that fired in all-or-nothing pulses. Norbert Wiener’s cybernetics described control and stability in electrical networks. Claude Shannon’s information theory described digital signals (i.e., all-or-nothing signals). Alan Turing’s theory of computation showed that any form of computation could be described digitally. The close relationship between these ideas suggested that it might be possible to construct an electronic brain.[27]

Examples of work in this vein includes robots such as W. Grey Walter’s turtles and the Johns Hopkins Beast. These machines did not use computers, digital electronics or symbolic reasoning; they were controlled entirely by analog circuitry.[28]

Walter Pitts and Warren McCulloch analyzed networks of idealized artificial neurons and showed how they might perform simple logical functions. They were the first to describe what later researchers would call a neural network.[29] One of the students inspired by Pitts and McCulloch was a young Marvin Minsky, then a 24-year-old graduate student. In 1951 (with Dean Edmonds) he built the first neural net machine, the SNARC.[30]Minsky was to become one of the most important leaders and innovators in AI for the next 50 years.

In 1950 Alan Turing published a landmark paper in which he speculated about the possibility of creating machines that think.[31]He noted that “thinking” is difficult to define and devised his famous Turing Test. If a machine could carry on a conversation (over a teleprinter) that was indistinguishable from a conversation with a human being, then it was reasonable to say that the machine was “thinking”. This simplified version of the problem allowed Turing to argue convincingly that a “thinking machine” was at least plausible and the paper answered all the most common objections to the proposition.[32] The Turing Test was the first serious proposal in the philosophy of artificial intelligence.

In 1951, using the Ferranti Mark 1 machine of the University of Manchester, Christopher Strachey wrote a checkers program and Dietrich Prinz wrote one for chess.[33] Arthur Samuel’s checkers program, developed in the middle 50s and early 60s, eventually achieved sufficient skill to challenge a respectable amateur.[34] Game AI would continue to be used as a measure of progress in AI throughout its history.

When access to digital computers became possible in the middle fifties, a few scientists instinctively recognized that a machine that could manipulate numbers could also manipulate symbols and that the manipulation of symbols could well be the essence of human thought. This was a new approach to creating thinking machines.[35]

In 1955, Allen Newell and (future Nobel Laureate) Herbert A. Simon created the “Logic Theorist” (with help from J. C. Shaw). The program would eventually prove 38 of the first 52 theorems in Russell and Whitehead’s Principia Mathematica, and find new and more elegant proofs for some.[36]Simon said that they had “solved the venerable mind/body problem, explaining how a system composed of matter can have the properties of mind.”[37](This was an early statement of the philosophical position John Searle would later call “Strong AI”: that machines can contain minds just as human bodies do.)[38]

The Dartmouth Conference of 1956[39]was organized by Marvin Minsky, John McCarthy and two senior scientists: Claude Shannon and Nathan Rochester of IBM. The proposal for the conference included this assertion: “every aspect of learning or any other feature of intelligence can be so precisely described that a machine can be made to simulate it”.[40]The participants included Ray Solomonoff, Oliver Selfridge, Trenchard More, Arthur Samuel, Allen Newell and Herbert A. Simon, all of whom would create important programs during the first decades of AI research.[41]At the conference Newell and Simon debuted the “Logic Theorist” and McCarthy persuaded the attendees to accept “Artificial Intelligence” as the name of the field.[42]The 1956 Dartmouth conference was the moment that AI gained its name, its mission, its first success and its major players, and is widely considered the birth of AI.[43]

The years after the Dartmouth conference were an era of discovery, of sprinting across new ground. The programs that were developed during this time were, to most people, simply “astonishing”:[44] computers were solving algebra word problems, proving theorems in geometry and learning to speak English. Few at the time would have believed that such “intelligent” behavior by machines was possible at all.[45] Researchers expressed an intense optimism in private and in print, predicting that a fully intelligent machine would be built in less than 20 years.[46] Government agencies like DARPA poured money into the new field.[47]

There were many successful programs and new directions in the late 50s and 1960s. Among the most influential were these:

Many early AI programs used the same basic algorithm. To achieve some goal (like winning a game or proving a theorem), they proceeded step by step towards it (by making a move or a deduction) as if searching through a maze, backtracking whenever they reached a dead end. This paradigm was called “reasoning as search”.[48]

The principal difficulty was that, for many problems, the number of possible paths through the “maze” was simply astronomical (a situation known as a “combinatorial explosion”). Researchers would reduce the search space by using heuristics or “rules of thumb” that would eliminate those paths that were unlikely to lead to a solution.[49]

Newell and Simon tried to capture a general version of this algorithm in a program called the “General Problem Solver”.[50] Other “searching” programs were able to accomplish impressive tasks like solving problems in geometry and algebra, such as Herbert Gelernter’s Geometry Theorem Prover (1958) and SAINT, written by Minsky’s student James Slagle (1961).[51] Other programs searched through goals and subgoals to plan actions, like the STRIPS system developed at Stanford to control the behavior of their robot Shakey.[52]

An important goal of AI research is to allow computers to communicate in natural languages like English. An early success was Daniel Bobrow’s program STUDENT, which could solve high school algebra word problems.[53]

A semantic net represents concepts (e.g. “house”,”door”) as nodes and relations among concepts (e.g. “has-a”) as links between the nodes. The first AI program to use a semantic net was written by Ross Quillian[54] and the most successful (and controversial) version was Roger Schank’s Conceptual dependency theory.[55]

Joseph Weizenbaum’s ELIZA could carry out conversations that were so realistic that users occasionally were fooled into thinking they were communicating with a human being and not a program. But in fact, ELIZA had no idea what she was talking about. She simply gave a canned response or repeated back what was said to her, rephrasing her response with a few grammar rules. ELIZA was the first chatterbot.[56]

In the late 60s, Marvin Minsky and Seymour Papert of the MIT AI Laboratory proposed that AI research should focus on artificially simple situations known as micro-worlds. They pointed out that in successful sciences like physics, basic principles were often best understood using simplified models like frictionless planes or perfectly rigid bodies. Much of the research focused on a “blocks world,” which consists of colored blocks of various shapes and sizes arrayed on a flat surface.[57]

This paradigm led to innovative work in machine vision by Gerald Sussman (who led the team), Adolfo Guzman, David Waltz (who invented “constraint propagation”), and especially Patrick Winston. At the same time, Minsky and Papert built a robot arm that could stack blocks, bringing the blocks world to life. The crowning achievement of the micro-world program was Terry Winograd’s SHRDLU. It could communicate in ordinary English sentences, plan operations and execute them.[58]

The first generation of AI researchers made these predictions about their work:

In June 1963, MIT received a $2.2 million grant from the newly created Advanced Research Projects Agency (later known as DARPA). The money was used to fund project MAC which subsumed the “AI Group” founded by Minsky and McCarthy five years earlier. DARPA continued to provide three million dollars a year until the 70s.[63]DARPA made similar grants to Newell and Simon’s program at CMU and to the Stanford AI Project (founded by John McCarthy in 1963).[64] Another important AI laboratory was established at Edinburgh University by Donald Michie in 1965.[65]These four institutions would continue to be the main centers of AI research (and funding) in academia for many years.[66]

The money was proffered with few strings attached: J. C. R. Licklider, then the director of ARPA, believed that his organization should “fund people, not projects!” and allowed researchers to pursue whatever directions might interest them.[67] This created a freewheeling atmosphere at MIT that gave birth to the hacker culture,[68] but this “hands off” approach would not last.

In Japan, Waseda University initiated the WABOT project in 1967, and in 1972 completed the WABOT-1, the world’s first full-scale intelligent humanoid robot,[69][70] or android. Its limb control system allowed it to walk with the lower limbs, and to grip and transport objects with hands, using tactile sensors. Its vision system allowed it to measure distances and directions to objects using external receptors, artificial eyes and ears. And its conversation system allowed it to communicate with a person in Japanese, with an artificial mouth.[71][72][73]

In the 1970s, AI was subject to critiques and financial setbacks. AI researchers had failed to appreciate the difficulty of the problems they faced. Their tremendous optimism had raised expectations impossibly high, and when the promised results failed to materialize, funding for AI disappeared.[74] At the same time, the field of connectionism (or neural nets) was shut down almost completely for 10 years by Marvin Minsky’s devastating criticism of perceptrons.[75]Despite the difficulties with public perception of AI in the late 70s, new ideas were explored in logic programming, commonsense reasoning and many other areas.[76]

In the early seventies, the capabilities of AI programs were limited. Even the most impressive could only handle trivial versions of the problems they were supposed to solve; all the programs were, in some sense, “toys”.[77] AI researchers had begun to run into several fundamental limits that could not be overcome in the 1970s. Although some of these limits would be conquered in later decades, others still stymie the field to this day.[78]

The agencies which funded AI research (such as the British government, DARPA and NRC) became frustrated with the lack of progress and eventually cut off almost all funding for undirected research into AI. The pattern began as early as 1966 when the ALPAC report appeared criticizing machine translation efforts. After spending 20 million dollars, the NRC ended all support.[86]In 1973, the Lighthill report on the state of AI research in England criticized the utter failure of AI to achieve its “grandiose objectives” and led to the dismantling of AI research in that country.[87](The report specifically mentioned the combinatorial explosion problem as a reason for AI’s failings.)[88]DARPA was deeply disappointed with researchers working on the Speech Understanding Research program at CMU and canceled an annual grant of three million dollars.[89]By 1974, funding for AI projects was hard to find.

Hans Moravec blamed the crisis on the unrealistic predictions of his colleagues. “Many researchers were caught up in a web of increasing exaggeration.”[90]However, there was another issue: since the passage of the Mansfield Amendment in 1969, DARPA had been under increasing pressure to fund “mission-oriented direct research, rather than basic undirected research”. Funding for the creative, freewheeling exploration that had gone on in the 60s would not come from DARPA. Instead, the money was directed at specific projects with clear objectives, such as autonomous tanks and battle management systems.[91]

Several philosophers had strong objections to the claims being made by AI researchers. One of the earliest was John Lucas, who argued that Gdel’s incompleteness theorem showed that a formal system (such as a computer program) could never see the truth of certain statements, while a human being could.[92] Hubert Dreyfus ridiculed the broken promises of the 1960s and critiqued the assumptions of AI, arguing that human reasoning actually involved very little “symbol processing” and a great deal of embodied, instinctive, unconscious “know how”.[93][94] John Searle’s Chinese Room argument, presented in 1980, attempted to show that a program could not be said to “understand” the symbols that it uses (a quality called “intentionality”). If the symbols have no meaning for the machine, Searle argued, then the machine can not be described as “thinking”.[95]

These critiques were not taken seriously by AI researchers, often because they seemed so far off the point. Problems like intractability and commonsense knowledge seemed much more immediate and serious. It was unclear what difference “know how” or “intentionality” made to an actual computer program. Minsky said of Dreyfus and Searle “they misunderstand, and should be ignored.”[96] Dreyfus, who taught at MIT, was given a cold shoulder: he later said that AI researchers “dared not be seen having lunch with me.”[97] Joseph Weizenbaum, the author of ELIZA, felt his colleagues’ treatment of Dreyfus was unprofessional and childish. Although he was an outspoken critic of Dreyfus’ positions, he “deliberately made it plain that theirs was not the way to treat a human being.”[98]

Weizenbaum began to have serious ethical doubts about AI when Kenneth Colby wrote a “computer program which can conduct psychotherapeutic dialogue” based on ELIZA.[99] Weizenbaum was disturbed that Colby saw a mindless program as a serious therapeutic tool. A feud began, and the situation was not helped when Colby did not credit Weizenbaum for his contribution to the program. In 1976, Weizenbaum published Computer Power and Human Reason which argued that the misuse of artificial intelligence has the potential to devalue human life.[100]

A perceptron was a form of neural network introduced in 1958 by Frank Rosenblatt, who had been a schoolmate of Marvin Minsky at the Bronx High School of Science. Like most AI researchers, he was optimistic about their power, predicting that “perceptron may eventually be able to learn, make decisions, and translate languages.” An active research program into the paradigm was carried out throughout the 1960s but came to a sudden halt with the publication of Minsky and Papert’s 1969 book Perceptrons. It suggested that there were severe limitations to what perceptrons could do and that Frank Rosenblatt’s predictions had been grossly exaggerated. The effect of the book was devastating: virtually no research at all was done in connectionism for 10 years. Eventually, a new generation of researchers would revive the field and thereafter it would become a vital and useful part of artificial intelligence. Rosenblatt would not live to see this, as he died in a boating accident shortly after the book was published.[75]

Logic was introduced into AI research as early as 1958, by John McCarthy in his Advice Taker proposal.[101]In 1963, J. Alan Robinson had discovered a simple method to implement deduction on computers, the resolution and unification algorithm. However, straightforward implementations, like those attempted by McCarthy and his students in the late 1960s, were especially intractable: the programs required astronomical numbers of steps to prove simple theorems.[102] A more fruitful approach to logic was developed in the 1970s by Robert Kowalski at the University of Edinburgh, and soon this led to the collaboration with French researchers Alain Colmerauer and Philippe Roussel who created the successful logic programming language Prolog.[103]Prolog uses a subset of logic (Horn clauses, closely related to “rules” and “production rules”) that permit tractable computation. Rules would continue to be influential, providing a foundation for Edward Feigenbaum’s expert systems and the continuing work by Allen Newell and Herbert A. Simon that would lead to Soar and their unified theories of cognition.[104]

Critics of the logical approach noted, as Dreyfus had, that human beings rarely used logic when they solved problems. Experiments by psychologists like Peter Wason, Eleanor Rosch, Amos Tversky, Daniel Kahneman and others provided proof.[105]McCarthy responded that what people do is irrelevant. He argued that what is really needed are machines that can solve problemsnot machines that think as people do.[106]

Among the critics of McCarthy’s approach were his colleagues across the country at MIT. Marvin Minsky, Seymour Papert and Roger Schank were trying to solve problems like “story understanding” and “object recognition” that required a machine to think like a person. In order to use ordinary concepts like “chair” or “restaurant” they had to make all the same illogical assumptions that people normally made. Unfortunately, imprecise concepts like these are hard to represent in logic. Gerald Sussman observed that “using precise language to describe essentially imprecise concepts doesn’t make them any more precise.”[107] Schank described their “anti-logic” approaches as “scruffy”, as opposed to the “neat” paradigms used by McCarthy, Kowalski, Feigenbaum, Newell and Simon.[108]

In 1975, in a seminal paper, Minsky noted that many of his fellow “scruffy” researchers were using the same kind of tool: a framework that captures all our common sense assumptions about something. For example, if we use the concept of a bird, there is a constellation of facts that immediately come to mind: we might assume that it flies, eats worms and so on. We know these facts are not always true and that deductions using these facts will not be “logical”, but these structured sets of assumptions are part of the context of everything we say and think. He called these structures “frames”. Schank used a version of frames he called “scripts” to successfully answer questions about short stories in English.[109] Many years later object-oriented programming would adopt the essential idea of “inheritance” from AI research on frames.

In the 1980s a form of AI program called “expert systems” was adopted by corporations around the world and knowledge became the focus of mainstream AI research. In those same years, the Japanese government aggressively funded AI with its fifth generation computer project. Another encouraging event in the early 1980s was the revival of connectionism in the work of John Hopfield and David Rumelhart. Once again, AI had achieved success.

An expert system is a program that answers questions or solves problems about a specific domain of knowledge, using logical rules that are derived from the knowledge of experts. The earliest examples were developed by Edward Feigenbaum and his students. Dendral, begun in 1965, identified compounds from spectrometer readings. MYCIN, developed in 1972, diagnosed infectious blood diseases. They demonstrated the feasibility of the approach.[110]

Expert systems restricted themselves to a small domain of specific knowledge (thus avoiding the commonsense knowledge problem) and their simple design made it relatively easy for programs to be built and then modified once they were in place. All in all, the programs proved to be useful: something that AI had not been able to achieve up to this point.[111]

In 1980, an expert system called XCON was completed at CMU for the Digital Equipment Corporation. It was an enormous success: it was saving the company 40 million dollars annually by 1986.[112] Corporations around the world began to develop and deploy expert systems and by 1985 they were spending over a billion dollars on AI, most of it to in-house AI departments. An industry grew up to support them, including hardware companies like Symbolics and Lisp Machines and software companies such as IntelliCorp and Aion.[113]

The power of expert systems came from the expert knowledge they contained. They were part of a new direction in AI research that had been gaining ground throughout the 70s. “AI researchers were beginning to suspectreluctantly, for it violated the scientific canon of parsimonythat intelligence might very well be based on the ability to use large amounts of diverse knowledge in different ways,”[114] writes Pamela McCorduck. “[T]he great lesson from the 1970s was that intelligent behavior depended very much on dealing with knowledge, sometimes quite detailed knowledge, of a domain where a given task lay”.[115] Knowledge based systems and knowledge engineering became a major focus of AI research in the 1980s.[116]

The 1980s also saw the birth of Cyc, the first attempt to attack the commonsense knowledge problem directly, by creating a massive database that would contain all the mundane facts that the average person knows. Douglas Lenat, who started and led the project, argued that there is no shortcut the only way for machines to know the meaning of human concepts is to teach them, one concept at a time, by hand. The project was not expected to be completed for many decades.[117]

Chess playing programs HiTech and Deep Thought defeated chess masters in 1989. Both were developed by Carnegie Mellon University; Deep Thought development paved the way for Deep Blue.[118]

In 1981, the Japanese Ministry of International Trade and Industry set aside $850 million for the Fifth generation computer project. Their objectives were to write programs and build machines that could carry on conversations, translate languages, interpret pictures, and reason like human beings.[119] Much to the chagrin of scruffies, they chose Prolog as the primary computer language for the project.[120]

Other countries responded with new programs of their own. The UK began the 350 million Alvey project. A consortium of American companies formed the Microelectronics and Computer Technology Corporation (or “MCC”) to fund large scale projects in AI and information technology.[121][122] DARPA responded as well, founding the Strategic Computing Initiative and tripling its investment in AI between 1984 and 1988.[123]

In 1982, physicist John Hopfield was able to prove that a form of neural network (now called a “Hopfield net”) could learn and process information in a completely new way. Around the same time, David Rumelhart popularized a new method for training neural networks called “backpropagation” (discovered years earlier by Paul Werbos). These two discoveries revived the field of connectionism which had been largely abandoned since 1970.[122][124]

The new field was unified and inspired by the appearance of Parallel Distributed Processing in 1986a two volume collection of papers edited by Rumelhart and psychologist James McClelland. Neural networks would become commercially successful in the 1990s, when they began to be used as the engines driving programs like optical character recognition and speech recognition.[122][125]

The business community’s fascination with AI rose and fell in the 1980s in the classic pattern of an economic bubble. The collapse was in the perception of AI by government agencies and investors the field continued to make advances despite the criticism. Rodney Brooks and Hans Moravec, researchers from the related field of robotics, argued for an entirely new approach to artificial intelligence.

The term “AI winter” was coined by researchers who had survived the funding cuts of 1974 when they became concerned that enthusiasm for expert systems had spiraled out of control and that disappointment would certainly follow.[126] Their fears were well founded: in the late 1980s and early 1990s, AI suffered a series of financial setbacks.

The first indication of a change in weather was the sudden collapse of the market for specialized AI hardware in 1987. Desktop computers from Apple and IBM had been steadily gaining speed and power and in 1987 they became more powerful than the more expensive Lisp machines made by Symbolics and others. There was no longer a good reason to buy them. An entire industry worth half a billion dollars was demolished overnight.[127]

Eventually the earliest successful expert systems, such as XCON, proved too expensive to maintain. They were difficult to update, they could not learn, they were “brittle” (i.e., they could make grotesque mistakes when given unusual inputs), and they fell prey to problems (such as the qualification problem) that had been identified years earlier. Expert systems proved useful, but only in a few special contexts.[128]

In the late 1980s, the Strategic Computing Initiative cut funding to AI “deeply and brutally.” New leadership at DARPA had decided that AI was not “the next wave” and directed funds towards projects that seemed more likely to produce immediate results.[129]

By 1991, the impressive list of goals penned in 1981 for Japan’s Fifth Generation Project had not been met. Indeed, some of them, like “carry on a casual conversation” had not been met by 2010.[130] As with other AI projects, expectations had run much higher than what was actually possible.[130]

In the late 1980s, several researchers advocated a completely new approach to artificial intelligence, based on robotics.[131] They believed that, to show real intelligence, a machine needs to have a body it needs to perceive, move, survive and deal with the world. They argued that these sensorimotor skills are essential to higher level skills like commonsense reasoning and that abstract reasoning was actually the least interesting or important human skill (see Moravec’s paradox). They advocated building intelligence “from the bottom up.”[132]

The approach revived ideas from cybernetics and control theory that had been unpopular since the sixties. Another precursor was David Marr, who had come to MIT in the late 1970s from a successful background in theoretical neuroscience to lead the group studying vision. He rejected all symbolic approaches (both McCarthy’s logic and Minsky’s frames), arguing that AI needed to understand the physical machinery of vision from the bottom up before any symbolic processing took place. (Marr’s work would be cut short by leukemia in 1980.)[133]

In a 1990 paper, “Elephants Don’t Play Chess,”[134] robotics researcher Rodney Brooks took direct aim at the physical symbol system hypothesis, arguing that symbols are not always necessary since “the world is its own best model. It is always exactly up to date. It always has every detail there is to be known. The trick is to sense it appropriately and often enough.”[135] In the 1980s and 1990s, many cognitive scientists also rejected the symbol processing model of the mind and argued that the body was essential for reasoning, a theory called the embodied mind thesis.[136]

The field of AI, now more than a half a century old, finally achieved some of its oldest goals. It began to be used successfully throughout the technology industry, although somewhat behind the scenes. Some of the success was due to increasing computer power and some was achieved by focusing on specific isolated problems and pursuing them with the highest standards of scientific accountability. Still, the reputation of AI, in the business world at least, was less than pristine. Inside the field there was little agreement on the reasons for AI’s failure to fulfill the dream of human level intelligence that had captured the imagination of the world in the 1960s. Together, all these factors helped to fragment AI into competing subfields focused on particular problems or approaches, sometimes even under new names that disguised the tarnished pedigree of “artificial intelligence”.[137] AI was both more cautious and more successful than it had ever been.

On 11 May 1997, Deep Blue became the first computer chess-playing system to beat a reigning world chess champion, Garry Kasparov.[138] The super computer was a specialized version of a framework produced by IBM, and was capable of processing twice as many moves per second as it had during the first match (which Deep Blue had lost), reportedly 200,000,000 moves per second. The event was broadcast live over the internet and received over 74 million hits.[139]

In 2005, a Stanford robot won the DARPA Grand Challenge by driving autonomously for 131 miles along an unrehearsed desert trail.[140] Two years later, a team from CMU won the DARPA Urban Challenge by autonomously navigating 55 miles in an Urban environment while adhering to traffic hazards and all traffic laws.[141] In February 2011, in a Jeopardy! quiz show exhibition match, IBM’s question answering system, Watson, defeated the two greatest Jeopardy! champions, Brad Rutter and Ken Jennings, by a significant margin.[142]

These successes were not due to some revolutionary new paradigm, but mostly on the tedious application of engineering skill and on the tremendous power of computers today.[143] In fact, Deep Blue’s computer was 10 million times faster than the Ferranti Mark 1 that Christopher Strachey taught to play chess in 1951.[144] This dramatic increase is measured by Moore’s law, which predicts that the speed and memory capacity of computers doubles every two years. The fundamental problem of “raw computer power” was slowly being overcome.

A new paradigm called “intelligent agents” became widely accepted during the 1990s.[145] Although earlier researchers had proposed modular “divide and conquer” approaches to AI,[146] the intelligent agent did not reach its modern form until Judea Pearl, Allen Newell, Leslie P. Kaelbling, and others brought concepts from decision theory and economics into the study of AI.[147] When the economist’s definition of a rational agent was married to computer science’s definition of an object or module, the intelligent agent paradigm was complete.

An intelligent agent is a system that perceives its environment and takes actions which maximize its chances of success. By this definition, simple programs that solve specific problems are “intelligent agents”, as are human beings and organizations of human beings, such as firms. The intelligent agent paradigm defines AI research as “the study of intelligent agents”. This is a generalization of some earlier definitions of AI: it goes beyond studying human intelligence; it studies all kinds of intelligence.[148]

The paradigm gave researchers license to study isolated problems and find solutions that were both verifiable and useful. It provided a common language to describe problems and share their solutions with each other, and with other fields that also used concepts of abstract agents, like economics and control theory. It was hoped that a complete agent architecture (like Newell’s SOAR) would one day allow researchers to build more versatile and intelligent systems out of interacting intelligent agents.[147][149]

AI researchers began to develop and use sophisticated mathematical tools more than they ever had in the past.[150] There was a widespread realization that many of the problems that AI needed to solve were already being worked on by researchers in fields like mathematics, economics or operations research. The shared mathematical language allowed both a higher level of collaboration with more established and successful fields and the achievement of results which were measurable and provable; AI had become a more rigorous “scientific” discipline. Russell & Norvig (2003) describe this as nothing less than a “revolution” and “the victory of the neats”.[151][152]

Judea Pearl’s highly influential 1988 book[153] brought probability and decision theory into AI. Among the many new tools in use were Bayesian networks, hidden Markov models, information theory, stochastic modeling and classical optimization. Precise mathematical descriptions were also developed for “computational intelligence” paradigms like neural networks and evolutionary algorithms.[151]

Algorithms originally developed by AI researchers began to appear as parts of larger systems. AI had solved a lot of very difficult problems[154]and their solutions proved to be useful throughout the technology industry,[155] such asdata mining,industrial robotics,logistics,[156]speech recognition,[157]banking software,[158]medical diagnosis[158]and Google’s search engine.[159]

The field of AI received little or no credit for these successes in the 1990s and early 2000s. Many of AI’s greatest innovations have been reduced to the status of just another item in the tool chest of computer science.[160] Nick Bostrom explains “A lot of cutting edge AI has filtered into general applications, often without being called AI because once something becomes useful enough and common enough it’s not labeled AI anymore.”[161]

Many researchers in AI in 1990s deliberately called their work by other names, such as informatics, knowledge-based systems, cognitive systems or computational intelligence. In part, this may be because they considered their field to be fundamentally different from AI, but also the new names help to procure funding. In the commercial world at least, the failed promises of the AI Winter continued to haunt AI research into the 2000s, as the New York Times reported in 2005: “Computer scientists and software engineers avoided the term artificial intelligence for fear of being viewed as wild-eyed dreamers.”[162][163][164]

In 1968, Arthur C. Clarke and Stanley Kubrick had imagined that by the year 2001, a machine would exist with an intelligence that matched or exceeded the capability of human beings. The character they created, HAL 9000, was based on a belief shared by many leading AI researchers that such a machine would exist by the year 2001.[165]

In 2001, AI founder Marvin Minsky asked “So the question is why didn’t we get HAL in 2001?”[166] Minsky believed that the answer is that the central problems, like commonsense reasoning, were being neglected, while most researchers pursued things like commercial applications of neural nets or genetic algorithms. John McCarthy, on the other hand, still blamed the qualification problem.[167] For Ray Kurzweil, the issue is computer power and, using Moore’s Law, he predicted that machines with human-level intelligence will appear by 2029.[168] Jeff Hawkins argued that neural net research ignores the essential properties of the human cortex, preferring simple models that have been successful at solving simple problems.[169] There were many other explanations and for each there was a corresponding research program underway.

In the first decades of the 21st century, access to large amounts of data (known as “big data”), faster computers and advanced machine learning techniques were successfully applied to many problems throughout the economy. In fact, McKinsey Global Institute estimated in their famous paper “Big data: The next frontier for innovation, competition, and productivity” that “by 2009, nearly all sectors in the US economy had at least an average of 200 terabytes of stored data”.

By 2016, the market for AI-related products, hardware, and software reached more than 8 billion dollars, and the New York Times reported that interest in AI had reached a “frenzy”.[170] The applications of big data began to reach into other fields as well, such as training models in ecology[171] and for various applications in economics.[172] Advances in deep learning (particularly deep convolutional neural networks and recurrent neural networks) drove progress and research in image and video processing, text analysis, and even speech recognition.[173]

Deep learning is a branch of machine learning that models high level abstractions in data by using a deep graph with many processing layers.[173] According to the Universal approximation theorem, deep-ness isn’t necessary for a neural network to be able to approximate arbitrary continuous functions. Even so, there are many problems that are common to shallow networks (such as overfitting) that deep networks help avoid.[174] As such, deep neural networks are able to realistically generate much more complex models as compared to their shallow counterparts.

However, deep learning has problems of its own. A common problem for recurrent neural networks is the vanishing gradient problem, which is where gradients passed between layers gradually shrink and literally disappear as they are rounded off to zero. There have been many methods developed to approach this problem, such as Long short-term memory units.

State-of-the-art deep neural network architectures can sometimes even rival human accuracy in fields like computer vision, specifically on things like the MNIST database, and traffic sign recognition.[175]

Language processing engines powered by smart search engines can easily beat humans at answering general trivia questions (such as IBM Watson), and recent developments in deep learning have produced astounding results in competing with humans, in things like Go and Doom (which, being a FPS, has sparked some controversy).[176][177][178][179]

Big data refers to a collection of data that cannot be captured, managed, and processed by conventional software tools within a certain time frame. It is a massive amount of decision-making, insight, and process optimization capabilities that require new processing models. In the Big Data Era written by Victor Meyer Schonberg and Kenneth Cooke, big data means that instead of random analysis (sample survey), all data is used for analysis. The 5V characteristics of big data (proposed by IBM): Volume, Velocity, Variety[180], Value[181], Veracity[182].The strategic significance of big data technology is not to master huge data information, but to specialize in these meaningful data. In other words, if big data is likened to an industry, the key to realizing profitability in this industry is to increase the Process capability of the data and realize the Value added of the data through Processing.

Artificial intelligence is a branch of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that responds in a manner similar to human intelligence. Research in this area includes robotics, speech recognition, image recognition, Natural language processing and expert systems. Since the birth of artificial intelligence, the theory and technology have become more and more mature, and the application fields have been expanding. It is conceivable that the technological products brought by artificial intelligence in the future will be the “container” of human wisdom. Artificial intelligence can simulate the information process of human consciousness and thinking. Artificial intelligence is not human intelligence, but it can be like human thinking, and it may exceed human intelligence.Artificial general intelligence is also referred to as “strong AI”,[183] “full AI”[184] or as the ability of a machine to perform “general intelligent action”.[3] Academic sources reserve “strong AI” to refer to machines capable of experiencing consciousness.

.

Continue reading here:

History of artificial intelligence – Wikipedia

What is Artificial Intelligence (AI)? – Definition from …

Artificial intelligence is a branch of computer science that aims to create intelligent machines. It has become an essential part of the technology industry.

Research associated with artificial intelligence is highly technical and specialized. The core problems of artificial intelligence include programming computers for certain traits such as:

Knowledge engineering is a core part of AI research. Machines can often act and react like humans only if they have abundant information relating to the world. Artificial intelligence must have access to objects, categories, properties and relations between all of them to implement knowledge engineering. Initiating common sense, reasoning and problem-solving power in machines is a difficult and tedious task.

Machine learning is also a core part of AI. Learning without any kind of supervision requires an ability to identify patterns in streams of inputs, whereas learning with adequate supervision involves classification and numerical regressions. Classification determines the category an object belongs to and regression deals with obtaining a set of numerical input or output examples, thereby discovering functions enabling the generation of suitable outputs from respective inputs. Mathematical analysis of machine learning algorithms and their performance is a well-defined branch of theoretical computer science often referred to as computational learning theory.

Machine perception deals with the capability to use sensory inputs to deduce the different aspects of the world, while computer vision is the power to analyze visual inputs with a few sub-problems such as facial, object and gesture recognition.

Robotics is also a major field related to AI. Robots require intelligence to handle tasks such as object manipulation and navigation, along with sub-problems of localization, motion planning and mapping.

Read this article:

What is Artificial Intelligence (AI)? – Definition from …

Libertarianism – Wikipedia

“Libertarians” redirects here. For political parties that may go by this name, see Libertarian Party.

Libertarianism (from Latin: libertas, meaning “freedom”) is a collection of political philosophies and movements that uphold liberty as a core principle.[1] Libertarians seek to maximize political freedom and autonomy, emphasizing freedom of choice, voluntary association, and individual judgment.[2][3][4] Libertarians share a skepticism of authority and state power, but they diverge on the scope of their opposition to existing political and economic systems. Various schools of libertarian thought offer a range of views regarding the legitimate functions of state and private power, often calling for the restriction or dissolution of coercive social institutions.[5]

Left-libertarian ideologies seek to abolish capitalism and private ownership of the means of production, or else to restrict their purview or effects, in favor of common or cooperative ownership and management, viewing private property as a barrier to freedom and liberty.[6][7][8][9] In contrast, modern right-libertarian ideologies, such as minarchism and anarcho-capitalism, instead advocate laissez-faire capitalism and strong private property rights,[10] such as in land, infrastructure, and natural resources.

The first recorded use of the term “libertarian” was in 1789, when William Belsham wrote about libertarianism in the context of metaphysics.[11]

“Libertarian” came to mean an advocate or defender of liberty, especially in the political and social spheres, as early as 1796, when the London Packet printed on 12 February: “Lately marched out of the Prison at Bristol, 450 of the French Libertarians”.[12] The word was again used in a political sense in 1802 in a short piece critiquing a poem by “the author of Gebir” and has since been used with this meaning.[13][14][15]

The use of the word “libertarian” to describe a new set of political positions has been traced to the French cognate, libertaire, coined in a letter French libertarian communist Joseph Djacque wrote to mutualist Pierre-Joseph Proudhon in 1857.[16][17][18] Djacque also used the term for his anarchist publication Le Libertaire: Journal du Mouvement Social, which was printed from 9 June 1858 to 4 February 1861 in New York City.[19][20] In the mid-1890s, Sbastien Faure began publishing a new Le Libertaire while France’s Third Republic enacted the lois sclrates (“villainous laws”), which banned anarchist publications in France. Libertarianism has frequently been used as a synonym for anarchism since this time.[21][22][23]

The term “libertarianism” was first used in the United States as a synonym for classic liberalism in May 1955 by writer Dean Russell, a colleague of Leonard Read and a classic liberal himself. He justified the choice of the word as follows: “Many of us call ourselves ‘liberals.’ And it is true that the word ‘liberal’ once described persons who respected the individual and feared the use of mass compulsions. But the leftists have now corrupted that once-proud term to identify themselves and their program of more government ownership of property and more controls over persons. As a result, those of us who believe in freedom must explain that when we call ourselves liberals, we mean liberals in the uncorrupted classical sense. At best, this is awkward and subject to misunderstanding. Here is a suggestion: Let those of us who love liberty trade-mark and reserve for our own use the good and honorable word ‘libertarian'”.[24]

Subsequently, a growing number of Americans with classical liberal beliefs in the United States began to describe themselves as “libertarian”. The person most responsible for popularizing the term “libertarian” was Murray Rothbard,[25] who started publishing libertarian works in the 1960s.

Libertarianism in the United States has been described as conservative on economic issues and liberal on personal freedom[26] (for common meanings of conservative and liberal in the United States) and it is also often associated with a foreign policy of non-interventionism.[27][28]

Although the word “libertarian” has been used to refer to socialists internationally, its meaning in the United States has deviated from its political origins.[29][30]

There is contention about whether left and right libertarianism “represent distinct ideologies as opposed to variations on a theme”.[31] All libertarians begin with a conception of personal autonomy from which they argue in favor of civil liberties and a reduction or elimination of the state.

Left-libertarianism encompasses those libertarian beliefs that claim the Earth’s natural resources belong to everyone in an egalitarian manner, either unowned or owned collectively. Contemporary left-libertarians such as Hillel Steiner, Peter Vallentyne, Philippe Van Parijs, Michael Otsuka and David Ellerman believe the appropriation of land must leave “enough and as good” for others or be taxed by society to compensate for the exclusionary effects of private property. Libertarian socialists (social and individualist anarchists, libertarian Marxists, council communists, Luxemburgists and DeLeonists) promote usufruct and socialist economic theories, including communism, collectivism, syndicalism and mutualism. They criticize the state for being the defender of private property and believe capitalism entails wage slavery.

Right-libertarianism[32] developed in the United States in the mid-20th century and is the most popular conception of libertarianism in that region.[33] It is commonly referred to as a continuation or radicalization of classical liberalism.[34][35] Right-libertarians, while often sharing left-libertarians’ advocacy for social freedom, also value the social institutions that enforce conditions of capitalism, while rejecting institutions that function in opposition to these on the grounds that such interventions represent unnecessary coercion of individuals and abrogation of their economic freedom.[36] Anarcho-capitalists[37][38] seek complete elimination of the state in favor of privately funded security services while minarchists defend “night-watchman states”, which maintain only those functions of government necessary to maintain conditions of capitalism and personal security.

Anarchism envisages freedom as a form of autonomy,[39] which Paul Goodman describes as “the ability to initiate a task and do it one’s own way, without orders from authorities who do not know the actual problem and the available means”.[40] All anarchists oppose political and legal authority, but collectivist strains also oppose the economic authority of private property.[41] These social anarchists emphasize mutual aid, whereas individualist anarchists extoll individual sovereignty.[42]

Some right-libertarians consider the non-aggression principle (NAP) to be a core part of their beliefs.[43][44]

Libertarians have been advocates and activists of civil liberties, including free love and free thought.[45][46] Advocates of free love viewed sexual freedom as a clear, direct expression of individual sovereignty and they particularly stressed women’s rights as most sexual laws discriminated against women: for example, marriage laws and anti-birth control measures.[47]

Free love appeared alongside anarcha-feminism and advocacy of LGBT rights. Anarcha-feminism developed as a synthesis of radical feminism and anarchism and views patriarchy as a fundamental manifestation of compulsory government. It was inspired by the late-19th-century writings of early feminist anarchists such as Lucy Parsons, Emma Goldman, Voltairine de Cleyre and Virginia Bolten. Anarcha-feminists, like other radical feminists, criticise and advocate the abolition of traditional conceptions of family, education and gender roles. Free Society (18951897 as The Firebrand, 18971904 as Free Society) was an anarchist newspaper in the United States that staunchly advocated free love and women’s rights, while criticizing “comstockery”, the censorship of sexual information.[48] In recent times, anarchism has also voiced opinions and taken action around certain sex-related subjects such as pornography,[49] BDSM[50] and the sex industry.[50]

Free thought is a philosophical viewpoint that holds opinions should be formed on the basis of science, logic and reason in contrast with authority, tradition or other dogmas.[51][52] In the United States, free thought was an anti-Christian, anti-clerical movement whose purpose was to make the individual politically and spiritually free to decide on religious matters. A number of contributors to Liberty were prominent figures in both free thought and anarchism. In 1901, Catalan anarchist and free-thinker Francesc Ferrer i Gurdia established “modern” or progressive schools in Barcelona in defiance of an educational system controlled by the Catholic Church.[53] Fiercely anti-clerical, Ferrer believed in “freedom in education”, i.e. education free from the authority of the church and state.[54] The schools’ stated goal was to “educate the working class in a rational, secular and non-coercive setting”. Later in the 20th century, Austrian Freudo-Marxist Wilhelm Reich became a consistent propagandist for sexual freedom going as far as opening free sex-counselling clinics in Vienna for working-class patients[55] as well as coining the phrase “sexual revolution” in one of his books from the 1940s.[56] During the early 1970s, the English anarchist and pacifist Alex Comfort achieved international celebrity for writing the sex manuals The Joy of Sex and More Joy of Sex.

Most left-libertarians are anarchists and believe the state inherently violates personal autonomy: “As Robert Paul Wolff has argued, since ‘the state is authority, the right to rule’, anarchism which rejects the State is the only political doctrine consistent with autonomy in which the individual alone is the judge of his moral constraints”.[41] Social anarchists believe the state defends private property, which they view as intrinsically harmful, while market-oriented left-libertarians argue that so-called free markets actually consist of economic privileges granted by the state. These latter libertarians advocate instead for freed markets, which are freed from these privileges.[57]

There is a debate amongst right-libertarians as to whether or not the state is legitimate: while anarcho-capitalists advocate its abolition, minarchists support minimal states, often referred to as night-watchman states. Libertarians take a skeptical view of government authority.[58][unreliable source?] Minarchists maintain that the state is necessary for the protection of individuals from aggression, theft, breach of contract and fraud. They believe the only legitimate governmental institutions are the military, police and courts, though some expand this list to include fire departments, prisons and the executive and legislative branches.[59] They justify the state on the grounds that it is the logical consequence of adhering to the non-aggression principle and argue that anarchism is immoral because it implies that the non-aggression principle is optional, that the enforcement of laws under anarchism is open to competition.[citation needed] Another common justification is that private defense agencies and court firms would tend to represent the interests of those who pay them enough.[60]

Anarcho-capitalists argue that the state violates the non-aggression principle (NAP) by its nature because governments use force against those who have not stolen or vandalized private property, assaulted anyone or committed fraud.[61][62] Linda & Morris Tannehill argue that no coercive monopoly of force can arise on a truly free market and that a government’s citizenry can not desert them in favor of a competent protection and defense agency.[63]

Left-libertarians believe that neither claiming nor mixing one’s labor with natural resources is enough to generate full private property rights[64][65] and maintain that natural resources ought to be held in an egalitarian manner, either unowned or owned collectively.[66]

Right-libertarians maintain that unowned natural resources “may be appropriated by the first person who discovers them, mixes his labor with them, or merely claims themwithout the consent of others, and with little or no payment to them”. They believe that natural resources are originally unowned and therefore private parties may appropriate them at will without the consent of, or owing to, others.[67]

Left-libertarians (social and individualist anarchists, libertarian Marxists and left-wing market anarchists) argue in favor of socialist theories such as communism, syndicalism and mutualism (anarchist economics). Daniel Gurin writes that “anarchism is really a synonym for socialism. The anarchist is primarily a socialist whose aim is to abolish the exploitation of man by man. Anarchism is only one of the streams of socialist thought, that stream whose main components are concern for liberty and haste to abolish the State”.[68]

Right-libertarians are economic liberals of either the Austrian School or Chicago school and support laissez-faire capitalism.[69]

Wage labour has long been compared by socialists and anarcho-syndicalists to slavery.[70][71][72][73] As a result, the term “wage slavery” is often utilised as a pejorative for wage labor.[74] Advocates of slavery looked upon the “comparative evils of Slave Society and of Free Society, of slavery to human Masters and slavery to Capital”[75] and proceeded to argue that wage slavery was actually worse than chattel slavery.[76] Slavery apologists like George Fitzhugh contended that workers only accepted wage labour with the passage of time, as they became “familiarized and inattentive to the infected social atmosphere they continually inhale[d]”.[75]

According to Noam Chomsky, analysis of the psychological implications of wage slavery goes back to the Enlightenment era. In his 1791 book On the Limits of State Action, classical liberal thinker Wilhelm von Humboldt explained how “whatever does not spring from a man’s free choice, or is only the result of instruction and guidance, does not enter into his very nature; he does not perform it with truly human energies, but merely with mechanical exactness” and so when the labourer works under external control “we may admire what he does, but we despise what he is”.[77] For Marxists, labour-as-commodity, which is how they regard wage labour,[78] provides an absolutely fundamental point of attack against capitalism.[79] “It can be persuasively argued”, noted philosopher John Nelson, “that the conception of the worker’s labour as a commodity confirms Marx’s stigmatization of the wage system of private capitalism as ‘wage-slavery;’ that is, as an instrument of the capitalist’s for reducing the worker’s condition to that of a slave, if not below it”.[80] That this objection is fundamental follows immediately from Marx’s conclusion that wage labour is the very foundation of capitalism: “Without a class dependent on wages, the moment individuals confront each other as free persons, there can be no production of surplus value; without the production of surplus-value there can be no capitalist production, and hence no capital and no capitalist!”.[81]

Left-libertarianism (or left-wing libertarianism) names several related, but distinct approaches to political and social theory which stresses both individual freedom and social equality. In its classical usage, left-libertarianism is a synonym for anti-authoritarian varieties of left-wing politics, i.e. libertarian socialism, which includes anarchism and libertarian Marxism, among others.[82][83] Left-libertarianism can also refer to political positions associated with academic philosophers Hillel Steiner, Philippe Van Parijs and Peter Vallentyne that combine self-ownership with an egalitarian approach to natural resouces.[84]

While maintaining full respect for personal property, left-libertarians are skeptical of or fully against private property, arguing that neither claiming nor mixing one’s labor with natural resources is enough to generate full private property rights[85][86] and maintain that natural resources (land, oil, gold and vegetation) should be held in an egalitarian manner, either unowned or owned collectively. Those left-libertarians who support private property do so under the condition that recompense is offered to the local community.[86] Many left-libertarian schools of thought are communist, advocating the eventual replacement of money with labor vouchers or decentralized planning.

On the other hand, left-wing market anarchism, which includes Pierre-Joseph Proudhon’s mutualism and Samuel Edward Konkin III’s agorism, appeals to left-wing concerns such as egalitarianism, gender and sexuality, class, immigration and environmentalism within the paradigm of a socialist free market.[82]

Right-libertarianism (or right-wing libertarianism) refers to libertarian political philosophies that advocate negative rights, natural law and a major reversal of the modern welfare state.[87] Right-libertarians strongly support private property rights and defend market distribution of natural resources and private property.[88] This position is contrasted with that of some versions of left-libertarianism, which maintain that natural resources belong to everyone in an egalitarian manner, either unowned or owned collectively.[89] Right-libertarianism includes anarcho-capitalism and laissez-faire, minarchist liberalism.[note 1]

Elements of libertarianism can be traced as far back as the ancient Chinese philosopher Lao-Tzu and the higher-law concepts of the Greeks and the Israelites.[90][91] In 17th-century England, libertarian ideas began to take modern form in the writings of the Levellers and John Locke. In the middle of that century, opponents of royal power began to be called Whigs, or sometimes simply “opposition” or “country” (as opposed to Court) writers.[92]

During the 18th century, classical liberal ideas flourished in Europe and North America.[93][94] Libertarians of various schools were influenced by classical liberal ideas.[95] For libertarian philosopher Roderick T. Long, both libertarian socialists and libertarian capitalists “share a commonor at least an overlapping intellectual ancestry… both claim the seventeenth century English Levellers and the eighteenth century French encyclopedists among their ideological forebears; and (also)… usually share an admiration for Thomas Jefferson[96][97][98] and Thomas Paine”.[99]

John Locke greatly influenced both libertarianism and the modern world in his writings published before and after the English Revolution of 1688, especially A Letter Concerning Toleration (1667), Two Treatises of Government (1689) and An Essay Concerning Human Understanding (1690). In the text of 1689, he established the basis of liberal political theory: that people’s rights existed before government; that the purpose of government is to protect personal and property rights; that people may dissolve governments that do not do so; and that representative government is the best form to protect rights.[100] The United States Declaration of Independence was inspired by Locke in its statement: “[T]o secure these rights, Governments are instituted among Men, deriving their just powers from the consent of the governed. That whenever any Form of Government becomes destructive of these ends, it is the Right of the People to alter or to abolish it”.[101] Nevertheless scholar Ellen Meiksins Wood says that “there are doctrines of individualism that are opposed to Lockean individualism… and non-Lockean individualism may encompass socialism”.[102]

According to Murray Rothbard, the libertarian creed emerged from the classical liberal challenges to an “absolute central State and a king ruling by divine right on top of an older, restrictive web of feudal land monopolies and urban guild controls and restrictions”, the mercantilism of a bureaucratic warfaring state allied with privileged merchants. The object of classical liberals was individual liberty in the economy, in personal freedoms and civil liberty, separation of state and religion, and peace as an alternative to imperial aggrandizement. He cites Locke’s contemporaries, the Levellers, who held similar views. Also influential were the English “Cato’s Letters” during the early 1700s, reprinted eagerly by American colonists who already were free of European aristocracy and feudal land monopolies.[101]

In January of 1776, only two years after coming to America from England, Thomas Paine published his pamphlet Common Sense calling for independence for the colonies.[103] Paine promoted classical liberal ideas in clear, concise language that allowed the general public to understand the debates among the political elites.[104] Common Sense was immensely popular in disseminating these ideas,[105] selling hundreds of thousands of copies.[106] Paine later would write the Rights of Man and The Age of Reason and participate in the French Revolution.[103] Paine’s theory of property showed a “libertarian concern” with the redistribution of resources.[107]

In 1793, William Godwin wrote a libertarian philosophical treatise, Enquiry Concerning Political Justice and its Influence on Morals and Happiness, which criticized ideas of human rights and of society by contract based on vague promises. He took classical liberalism to its logical anarchic conclusion by rejecting all political institutions, law, government and apparatus of coercion as well as all political protest and insurrection. Instead of institutionalized justice, Godwin proposed that people influence one another to moral goodness through informal reasoned persuasion, including in the associations they joined as this would facilitate happiness.[108][109]

Modern anarchism sprang from the secular or religious thought of the Enlightenment, particularly Jean-Jacques Rousseau’s arguments for the moral centrality of freedom.[110]

As part of the political turmoil of the 1790s in the wake of the French Revolution, William Godwin developed the first expression of modern anarchist thought.[111][112] According to Peter Kropotkin, Godwin was “the first to formulate the political and economical conceptions of anarchism, even though he did not give that name to the ideas developed in his work”,[113] while Godwin attached his anarchist ideas to an early Edmund Burke.[114]

Godwin is generally regarded as the founder of the school of thought known as philosophical anarchism. He argued in Political Justice (1793)[112][115] that government has an inherently malevolent influence on society and that it perpetuates dependency and ignorance. He thought that the spread of the use of reason to the masses would eventually cause government to wither away as an unnecessary force. Although he did not accord the state with moral legitimacy, he was against the use of revolutionary tactics for removing the government from power. Rather, Godwin advocated for its replacement through a process of peaceful evolution.[112][116]

His aversion to the imposition of a rules-based society led him to denounce, as a manifestation of the people’s “mental enslavement”, the foundations of law, property rights and even the institution of marriage. Godwin considered the basic foundations of society as constraining the natural development of individuals to use their powers of reasoning to arrive at a mutually beneficial method of social organization. In each case, government and its institutions are shown to constrain the development of our capacity to live wholly in accordance with the full and free exercise of private judgment.

In France, various anarchist currents were present during the Revolutionary period, with some revolutionaries using the term anarchiste in a positive light as early as September 1793.[117] The enrags opposed revolutionary government as a contradiction in terms. Denouncing the Jacobin dictatorship, Jean Varlet wrote in 1794 that “government and revolution are incompatible, unless the people wishes to set its constituted authorities in permanent insurrection against itself”.[118] In his “Manifesto of the Equals”, Sylvain Marchal looked forward to the disappearance, once and for all, of “the revolting distinction between rich and poor, of great and small, of masters and valets, of governors and governed”.[118]

Libertarian socialism, libertarian communism and libertarian Marxism are all phrases which activists with a variety of perspectives have applied to their views.[119] Anarchist communist philosopher Joseph Djacque was the first person to describe himself as a libertarian.[120] Unlike mutualist anarchist philosopher Pierre-Joseph Proudhon, he argued that “it is not the product of his or her labor that the worker has a right to, but to the satisfaction of his or her needs, whatever may be their nature”.[121][122] According to anarchist historian Max Nettlau, the first use of the term “libertarian communism” was in November 1880, when a French anarchist congress employed it to more clearly identify its doctrines.[123] The French anarchist journalist Sbastien Faure started the weekly paper Le Libertaire (The Libertarian) in 1895.[124]

Individualist anarchism refers to several traditions of thought within the anarchist movement that emphasize the individual and their will over any kinds of external determinants such as groups, society, traditions, and ideological systems.[125][126] An influential form of individualist anarchism called egoism[127] or egoist anarchism was expounded by one of the earliest and best-known proponents of individualist anarchism, the German Max Stirner.[128] Stirner’s The Ego and Its Own, published in 1844, is a founding text of the philosophy.[128] According to Stirner, the only limitation on the rights of the individual is their power to obtain what they desire,[129] without regard for God, state or morality.[130] Stirner advocated self-assertion and foresaw unions of egoists, non-systematic associations continually renewed by all parties’ support through an act of will,[131] which Stirner proposed as a form of organisation in place of the state.[132] Egoist anarchists argue that egoism will foster genuine and spontaneous union between individuals.[133] Egoism has inspired many interpretations of Stirner’s philosophy. It was re-discovered and promoted by German philosophical anarchist and LGBT activist John Henry Mackay. Josiah Warren is widely regarded as the first American anarchist,[134] and the four-page weekly paper he edited during 1833, The Peaceful Revolutionist, was the first anarchist periodical published.[135] For American anarchist historian Eunice Minette Schuster, “[i]t is apparent… that Proudhonian Anarchism was to be found in the United States at least as early as 1848 and that it was not conscious of its affinity to the Individualist Anarchism of Josiah Warren and Stephen Pearl Andrews… William B. Greene presented this Proudhonian Mutualism in its purest and most systematic form.”.[136] Later, Benjamin Tucker fused Stirner’s egoism with the economics of Warren and Proudhon in his eclectic influential publication Liberty. From these early influences, individualist anarchism in different countries attracted a small yet diverse following of bohemian artists and intellectuals,[137] free love and birth control advocates (anarchism and issues related to love and sex),[138][139] individualist naturists nudists (anarcho-naturism),[140][141][142] free thought and anti-clerical activists[143][144] as well as young anarchist outlaws in what became known as illegalism and individual reclamation[145][146] (European individualist anarchism and individualist anarchism in France). These authors and activists included Emile Armand, Han Ryner, Henri Zisly, Renzo Novatore, Miguel Gimenez Igualada, Adolf Brand and Lev Chernyi.

In 1873, the follower and translator of Proudhon, the Catalan Francesc Pi i Margall, became President of Spain with a program which wanted “to establish a decentralized, or “cantonalist,” political system on Proudhonian lines”,[147] who according to Rudolf Rocker had “political ideas…much in common with those of Richard Price, Joseph Priestly [sic], Thomas Paine, Jefferson, and other representatives of the Anglo-American liberalism of the first period. He wanted to limit the power of the state to a minimum and gradually replace it by a Socialist economic order”.[148] On the other hand, Fermn Salvochea was a mayor of the city of Cdiz and a president of the province of Cdiz. He was one of the main propagators of anarchist thought in that area in the late 19th century and is considered to be “perhaps the most beloved figure in the Spanish Anarchist movement of the 19th century”.[149][150] Ideologically, he was influenced by Bradlaugh, Owen and Paine, whose works he had studied during his stay in England and Kropotkin, whom he read later.[149] The revolutionary wave of 19171923 saw the active participation of anarchists in Russia and Europe. Russian anarchists participated alongside the Bolsheviks in both the February and October 1917 revolutions. However, Bolsheviks in central Russia quickly began to imprison or drive underground the libertarian anarchists. Many fled to the Ukraine.[151] There, in the Ukrainian Free Territory they fought in the Russian Civil War against the White movement, monarchists and other opponents of revolution and then against Bolsheviks as part of the Revolutionary Insurrectionary Army of Ukraine led by Nestor Makhno, who established an anarchist society in the region for a number of months. Expelled American anarchists Emma Goldman and Alexander Berkman protested Bolshevik policy before they left Russia.[152]

The victory of the Bolsheviks damaged anarchist movements internationally as workers and activists joined Communist parties. In France and the United States, for example, members of the major syndicalist movements of the CGT and IWW joined the Communist International.[153] In Paris, the Dielo Truda group of Russian anarchist exiles, which included Nestor Makhno, issued a 1926 manifesto, the Organizational Platform of the General Union of Anarchists (Draft), calling for new anarchist organizing structures.[154][155]

The Bavarian Soviet Republic of 19181919 had libertarian socialist characteristics.[156][157] In Italy, from 1918 to 1921 the anarcho-syndicalist trade union Unione Sindacale Italiana grew to 800,000 members.[158]

In the 1920s and 1930s, with the rise of fascism in Europe, anarchists began to fight fascists in Italy,[159] in France during the February 1934 riots[160] and in Spain where the CNT (Confederacin Nacional del Trabajo) boycott of elections led to a right-wing victory and its later participation in voting in 1936 helped bring the popular front back to power. This led to a ruling class attempted coup and the Spanish Civil War (19361939).[161] Gruppo Comunista Anarchico di Firenze held that the during early twentieth century, the terms libertarian communism and anarchist communism became synonymous within the international anarchist movement as a result of the close connection they had in Spain (anarchism in Spain) (with libertarian communism becoming the prevalent term).[162]

Murray Bookchin wrote that the Spanish libertarian movement of the mid-1930s was unique because its workers’ control and collectiveswhich came out of a three-generation “massive libertarian movement”divided the republican camp and challenged the Marxists. “Urban anarchists” created libertarian communist forms of organization which evolved into the CNT, a syndicalist union providing the infrastructure for a libertarian society. Also formed were local bodies to administer social and economic life on a decentralized libertarian basis. Much of the infrastructure was destroyed during the 1930s Spanish Civil War against authoritarian and fascist forces.[163] The Iberian Federation of Libertarian Youth[164] (FIJL, Spanish: Federacin Ibrica de Juventudes Libertarias), sometimes abbreviated as Libertarian Youth (Juventudes Libertarias), was a libertarian socialist[165] organisation created in 1932 in Madrid.[166] In February 1937, the FIJL organised a plenum of regional organisations (second congress of FIJL). In October 1938, from the 16th through the 30th in Barcelona the FIJL participated in a national plenum of the libertarian movement, also attended by members of the CNT and the Iberian Anarchist Federation (FAI).[167] The FIJL exists until today. When the republican forces lost the Spanish Civil War, the city of Madrid was turned over to the francoist forces in 1939 by the last non-francoist mayor of the city, the anarchist Melchor Rodrguez Garca.[168] During autumn of 1931, the “Manifesto of the 30” was published by militants of the anarchist trade union CNT and among those who signed it there was the CNT General Secretary (19221923) Joan Peiro, Angel Pestaa CNT (General Secretary in 1929) and Juan Lopez Sanchez. They were called treintismo and they were calling for “libertarian possibilism” which advocated achieving libertarian socialist ends with participation inside structures of contemporary parliamentary democracy.[169] In 1932, they establish the Syndicalist Party which participates in the 1936 spanish general elections and proceed to be a part of the leftist coalition of parties known as the Popular Front obtaining 2 congressmen (Pestaa and Benito Pabon). In 1938, Horacio Prieto, general secretary of the CNT, proposes that the Iberian Anarchist Federation transforms itself into a “Libertarian Socialist Party” and that it participates in the national elections.[170]

The Manifesto of Libertarian Communism was written in 1953 by Georges Fontenis for the Federation Communiste Libertaire of France. It is one of the key texts of the anarchist-communist current known as platformism.[171] In 1968, in Carrara, Italy the International of Anarchist Federations was founded during an international anarchist conference to advance libertarian solidarity. It wanted to form “a strong and organised workers movement, agreeing with the libertarian ideas”.[172][173] In the United States, the Libertarian League was founded in New York City in 1954 as a left-libertarian political organisation building on the Libertarian Book Club.[174][175] Members included Sam Dolgoff,[176] Russell Blackwell, Dave Van Ronk, Enrico Arrigoni[177] and Murray Bookchin.

In Australia, the Sydney Push was a predominantly left-wing intellectual subculture in Sydney from the late 1940s to the early 1970s which became associated with the label “Sydney libertarianism”. Well known associates of the Push include Jim Baker, John Flaus, Harry Hooton, Margaret Fink, Sasha Soldatow,[178] Lex Banning, Eva Cox, Richard Appleton, Paddy McGuinness, David Makinson, Germaine Greer, Clive James, Robert Hughes, Frank Moorhouse and Lillian Roxon. Amongst the key intellectual figures in Push debates were philosophers David J. Ivison, George Molnar, Roelof Smilde, Darcy Waters and Jim Baker, as recorded in Baker’s memoir Sydney Libertarians and the Push, published in the libertarian Broadsheet in 1975.[179] An understanding of libertarian values and social theory can be obtained from their publications, a few of which are available online.[180][181]

In 1969, French platformist anarcho-communist Daniel Gurin published an essay in 1969 called “Libertarian Marxism?” in which he dealt with the debate between Karl Marx and Mikhail Bakunin at the First International and afterwards suggested that “[l]ibertarian marxism rejects determinism and fatalism, giving the greater place to individual will, intuition, imagination, reflex speeds, and to the deep instincts of the masses, which are more far-seeing in hours of crisis than the reasonings of the ‘elites’; libertarian marxism thinks of the effects of surprise, provocation and boldness, refuses to be cluttered and paralysed by a heavy ‘scientific’ apparatus, doesn’t equivocate or bluff, and guards itself from adventurism as much as from fear of the unknown”.[182] Libertarian Marxist currents often draw from Marx and Engels’ later works, specifically the Grundrisse and The Civil War in France.[183] They emphasize the Marxist belief in the ability of the working class to forge its own destiny without the need for a revolutionary party or state.[184] Libertarian Marxism includes such currents as council communism, left communism, Socialisme ou Barbarie, Lettrism/Situationism and operaismo/autonomism and New Left.[185][unreliable source?] In the United States, from 1970 to 1981 there existed the publication Root & Branch[186] which had as a subtitle “A Libertarian Marxist Journal”.[187] In 1974, the Libertarian Communism journal was started in the United Kingdom by a group inside the Socialist Party of Great Britain.[188] In 1986, the anarcho-syndicalist Sam Dolgoff started and led the publication Libertarian Labor Review in the United States[189] which decided to rename itself as Anarcho-Syndicalist Review in order to avoid confusion with right-libertarian views.[190]

The indigenous anarchist tradition in the United States was largely individualist.[191] In 1825, Josiah Warren became aware of the social system of utopian socialist Robert Owen and began to talk with others in Cincinnati about founding a communist colony.[192] When this group failed to come to an agreement about the form and goals of their proposed community, Warren “sold his factory after only two years of operation, packed up his young family, and took his place as one of 900 or so Owenites who had decided to become part of the founding population of New Harmony, Indiana”.[193] Warren termed the phrase “cost the limit of price”[194] and “proposed a system to pay people with certificates indicating how many hours of work they did. They could exchange the notes at local time stores for goods that took the same amount of time to produce”.[195] He put his theories to the test by establishing an experimental labor-for-labor store called the Cincinnati Time Store where trade was facilitated by labor notes. The store proved successful and operated for three years, after which it was closed so that Warren could pursue establishing colonies based on mutualism, including Utopia and Modern Times. “After New Harmony failed, Warren shifted his ideological loyalties from socialism to anarchism (which was no great leap, given that Owen’s socialism had been predicated on Godwin’s anarchism)”.[196] Warren is widely regarded as the first American anarchist[195] and the four-page weekly paper The Peaceful Revolutionist he edited during 1833 was the first anarchist periodical published,[135] an enterprise for which he built his own printing press, cast his own type and made his own printing plates.[135]

Catalan historian Xavier Diez reports that the intentional communal experiments pioneered by Warren were influential in European individualist anarchists of the late 19th and early 20th centuries such as mile Armand and the intentional communities started by them.[197] Warren said that Stephen Pearl Andrews, individualist anarchist and close associate, wrote the most lucid and complete exposition of Warren’s own theories in The Science of Society, published in 1852.[198] Andrews was formerly associated with the Fourierist movement, but converted to radical individualism after becoming acquainted with the work of Warren. Like Warren, he held the principle of “individual sovereignty” as being of paramount importance. Contemporary American anarchist Hakim Bey reports:

Steven Pearl Andrews… was not a fourierist, but he lived through the brief craze for phalansteries in America and adopted a lot of fourierist principles and practices… a maker of worlds out of words. He syncretized abolitionism in the United States, free love, spiritual universalism, Warren, and Fourier into a grand utopian scheme he called the Universal Pantarchy… He was instrumental in founding several ‘intentional communities,’ including the ‘Brownstone Utopia’ on 14th St. in New York, and ‘Modern Times’ in Brentwood, Long Island. The latter became as famous as the best-known fourierist communes (Brook Farm in Massachusetts & the North American Phalanx in New Jersey)in fact, Modern Times became downright notorious (for ‘Free Love’) and finally foundered under a wave of scandalous publicity. Andrews (and Victoria Woodhull) were members of the infamous Section 12 of the 1st International, expelled by Marx for its anarchist, feminist, and spiritualist tendencies.[199]

For American anarchist historian Eunice Minette Schuster, “[it is apparent… that Proudhonian Anarchism was to be found in the United States at least as early as 1848 and that it was not conscious of its affinity to the Individualist Anarchism of Josiah Warren and Stephen Pearl Andrews. William B. Greene presented this Proudhonian Mutualism in its purest and most systematic form”.[200] William Batchelder Greene was a 19th-century mutualist individualist anarchist, Unitarian minister, soldier and promoter of free banking in the United States. Greene is best known for the works Mutual Banking, which proposed an interest-free banking system; and Transcendentalism, a critique of the New England philosophical school. After 1850, he became active in labor reform.[200] “He was elected vice-president of the New England Labor Reform League, the majority of the members holding to Proudhon’s scheme of mutual banking, and in 1869 president of the Massachusetts Labor Union”.[200] Greene then published Socialistic, Mutualistic, and Financial Fragments (1875).[200] He saw mutualism as the synthesis of “liberty and order”.[200] His “associationism… is checked by individualism… ‘Mind your own business,’ ‘Judge not that ye be not judged.’ Over matters which are purely personal, as for example, moral conduct, the individual is sovereign, as well as over that which he himself produces. For this reason he demands ‘mutuality’ in marriagethe equal right of a woman to her own personal freedom and property”.[200]

Poet, naturalist and transcendentalist Henry David Thoreau was an important early influence in individualist anarchist thought in the United States and Europe. He is best known for his book Walden, a reflection upon simple living in natural surroundings; and his essay Civil Disobedience (Resistance to Civil Government), an argument for individual resistance to civil government in moral opposition to an unjust state. In Walden, Thoreau advocates simple living and self-sufficiency among natural surroundings in resistance to the advancement of industrial civilization.[201] Civil Disobedience, first published in 1849, argues that people should not permit governments to overrule or atrophy their consciences and that people have a duty to avoid allowing such acquiescence to enable the government to make them the agents of injustice. These works influenced green anarchism, anarcho-primitivism and anarcho-pacifism,[202] as well as figures including Mohandas Gandhi, Martin Luther King, Jr., Martin Buber and Leo Tolstoy.[202] “Many have seen in Thoreau one of the precursors of ecologism and anarcho-primitivism represented today in John Zerzan. For George Woodcock this attitude can be also motivated by certain idea of resistance to progress and of rejection of the growing materialism which is the nature of American society in the mid-19th century”.[201] Zerzan included Thoreau’s “Excursions” in his edited compilation of anti-civilization writings, Against Civilization: Readings and Reflections.[203] Individualist anarchists such as Thoreau[204][205] do not speak of economics, but simply the right of disunion from the state and foresee the gradual elimination of the state through social evolution. Agorist author J. Neil Schulman cites Thoreau as a primary inspiration.[206]

Many economists since Adam Smith have argued thatunlike other taxesa land value tax would not cause economic inefficiency.[207] It would be a progressive tax[208]primarily paid by the wealthyand increase wages, reduce economic inequality, remove incentives to misuse real estate and reduce the vulnerability that economies face from credit and property bubbles.[209][210] Early proponents of this view include Thomas Paine, Herbert Spencer, and Hugo Grotius,[84] but the concept was widely popularized by the economist and social reformer Henry George.[211] George believed that people ought to own the fruits of their labor and the value of the improvements they make, thus he was opposed to income taxes, sales taxes, taxes on improvements and all other taxes on production, labor, trade or commerce. George was among the staunchest defenders of free markets and his book Protection or Free Trade was read into the U.S. Congressional Record.[212] Yet he did support direct management of natural monopolies as a last resort, such as right-of-way monopolies necessary for railroads. George advocated for elimination of intellectual property arrangements in favor of government sponsored prizes for inventors.[213][not in citation given] Early followers of George’s philosophy called themselves single taxers because they believed that the only legitimate, broad-based tax was land rent. The term Georgism was coined later, though some modern proponents prefer the term geoism instead,[214] leaving the meaning of “geo” (Earth in Greek) deliberately ambiguous. The terms “Earth Sharing”,[215] “geonomics”[216] and “geolibertarianism”[217] are used by some Georgists to represent a difference of emphasis, or real differences about how land rent should be spent, but all agree that land rent should be recovered from its private owners.

Individualist anarchism found in the United States an important space for discussion and development within the group known as the “Boston anarchists”.[218] Even among the 19th-century American individualists there was no monolithic doctrine and they disagreed amongst each other on various issues including intellectual property rights and possession versus property in land.[219][220][221] Some Boston anarchists, including Benjamin Tucker, identified as socialists, which in the 19th century was often used in the sense of a commitment to improving conditions of the working class (i.e. “the labor problem”).[222] Lysander Spooner, besides his individualist anarchist activism, was also an anti-slavery activist and member of the First International.[223] Tucker argued that the elimination of what he called “the four monopolies”the land monopoly, the money and banking monopoly, the monopoly powers conferred by patents and the quasi-monopolistic effects of tariffswould undermine the power of the wealthy and big business, making possible widespread property ownership and higher incomes for ordinary people, while minimizing the power of would-be bosses and achieving socialist goals without state action. Tucker’s anarchist periodical, Liberty, was published from August 1881 to April 1908. The publication, emblazoned with Proudhon’s quote that liberty is “Not the Daughter But the Mother of Order” was instrumental in developing and formalizing the individualist anarchist philosophy through publishing essays and serving as a forum for debate. Contributors included Benjamin Tucker, Lysander Spooner, Auberon Herbert, Dyer Lum, Joshua K. Ingalls, John Henry Mackay, Victor Yarros, Wordsworth Donisthorpe, James L. Walker, J. William Lloyd, Florence Finch Kelly, Voltairine de Cleyre, Steven T. Byington, John Beverley Robinson, Jo Labadie, Lillian Harman and Henry Appleton.[224] Later, Tucker and others abandoned their traditional support of natural rights and converted to an egoism modeled upon the philosophy of Max Stirner.[220] A number of natural rights proponents stopped contributing in protest and “[t]hereafter, Liberty championed egoism, although its general content did not change significantly”.[225] Several publications “were undoubtedly influenced by Liberty’s presentation of egoism. They included: I published by C.L. Swartz, edited by W.E. Gordak and J.W. Lloyd (all associates of Liberty); The Ego and The Egoist, both of which were edited by Edward H. Fulton. Among the egoist papers that Tucker followed were the German Der Eigene, edited by Adolf Brand, and The Eagle and The Serpent, issued from London. The latter, the most prominent English-language egoist journal, was published from 1898 to 1900 with the subtitle ‘A Journal of Egoistic Philosophy and Sociology'”.[225]

By around the start of the 20th century, the heyday of individualist anarchism had passed.[226] H. L. Mencken and Albert Jay Nock were the first prominent figures in the United States to describe themselves as libertarians;[227] they believed Franklin D. Roosevelt had co-opted the word “liberal” for his New Deal policies which they opposed and used “libertarian” to signify their allegiance to individualism.[citation needed] In 1914, Nock joined the staff of The Nation magazine, which at the time was supportive of liberal capitalism. A lifelong admirer of Henry George, Nock went on to become co-editor of The Freeman from 1920 to 1924, a publication initially conceived as a vehicle for the single tax movement, financed by the wealthy wife of the magazine’s other editor, Francis Neilson.[228] Critic H.L. Mencken wrote that “[h]is editorials during the three brief years of the Freeman set a mark that no other man of his trade has ever quite managed to reach. They were well-informed and sometimes even learned, but there was never the slightest trace of pedantry in them”.[229]

Executive Vice President of the Cato Institute, David Boaz, writes: “In 1943, at one of the lowest points for liberty and humanity in history, three remarkable women published books that could be said to have given birth to the modern libertarian movement”.[230] Isabel Paterson’s The God of the Machine, Rose Wilder Lane’s The Discovery of Freedom and Ayn Rand’s The Fountainhead each promoted individualism and capitalism. None of the three used the term libertarianism to describe their beliefs and Rand specifically rejected the label, criticizing the burgeoning American libertarian movement as the “hippies of the right”.[231] Rand’s own philosophy, Objectivism, is notedly similar to libertarianism and she accused libertarians of plagiarizing her ideas.[231] Rand stated:

All kinds of people today call themselves “libertarians,” especially something calling itself the New Right, which consists of hippies who are anarchists instead of leftist collectivists; but anarchists are collectivists. Capitalism is the one system that requires absolute objective law, yet libertarians combine capitalism and anarchism. That’s worse than anything the New Left has proposed. It’s a mockery of philosophy and ideology. They sling slogans and try to ride on two bandwagons. They want to be hippies, but don’t want to preach collectivism because those jobs are already taken. But anarchism is a logical outgrowth of the anti-intellectual side of collectivism. I could deal with a Marxist with a greater chance of reaching some kind of understanding, and with much greater respect. Anarchists are the scum of the intellectual world of the Left, which has given them up. So the Right picks up another leftist discard. That’s the libertarian movement.[232]

In 1946, Leonard E. Read founded the Foundation for Economic Education (FEE), an American nonprofit educational organization which promotes the principles of laissez-faire economics, private property, and limited government.[233] According to Gary North, former FEE director of seminars and a current Ludwig von Mises Institute scholar, FEE is the “granddaddy of all libertarian organizations”.[234] The initial officers of FEE were Leonard E. Read as President, Austrian School economist Henry Hazlitt as Vice-President and Chairman David Goodrich of B. F. Goodrich. Other trustees on the FEE board have included wealthy industrialist Jasper Crane of DuPont, H. W. Luhnow of William Volker & Co. and Robert Welch, founder of the John Birch Society.[236][237]

Austrian school economist Murray Rothbard was initially an enthusiastic partisan of the Old Right, particularly because of its general opposition to war and imperialism,[238] but long embraced a reading of American history that emphasized the role of elite privilege in shaping legal and political institutions. He was part of Ayn Rand’s circle for a brief period, but later harshly criticized Objectivism.[239] He praised Rand’s Atlas Shrugged and wrote that she “introduced me to the whole field of natural rights and natural law philosophy”, prompting him to learn “the glorious natural rights tradition”.[240](pp121, 132134) He soon broke with Rand over various differences, including his defense of anarchism. Rothbard was influenced by the work of the 19th-century American individualist anarchists[241] and sought to meld their advocacy of free markets and private defense with the principles of Austrian economics.[242] This new philosophy he called anarcho-capitalism.

Karl Hess, a speechwriter for Barry Goldwater and primary author of the Republican Party’s 1960 and 1964 platforms, became disillusioned with traditional politics following the 1964 presidential campaign in which Goldwater lost to Lyndon B. Johnson. He parted with the Republicans altogether after being rejected for employment with the party, and began work as a heavy-duty welder. Hess began reading American anarchists largely due to the recommendations of his friend Murray Rothbard and said that upon reading the works of communist anarchist Emma Goldman, he discovered that anarchists believed everything he had hoped the Republican Party would represent. For Hess, Goldman was the source for the best and most essential theories of Ayn Rand without any of the “crazy solipsism that Rand was so fond of”.[243] Hess and Rothbard founded the journal Left and Right: A Journal of Libertarian Thought, which was published from 1965 to 1968, with George Resch and Leonard P. Liggio. In 1969, they edited The Libertarian Forum 1969, which Hess left in 1971. Hess eventually put his focus on the small scale, stating that “Society is: people together making culture”. He deemed two of his cardinal social principles to be “opposition to central political authority” and “concern for people as individuals”. His rejection of standard American party politics was reflected in a lecture he gave during which he said: “The Democrats or liberals think that everybody is stupid and therefore they need somebody… to tell them how to behave themselves. The Republicans think everybody is lazy”.[244]

The Vietnam War split the uneasy alliance between growing numbers of American libertarians and conservatives who believed in limiting liberty to uphold moral virtues. Libertarians opposed to the war joined the draft resistance and peace movements, as well as organizations such as Students for a Democratic Society (SDS). In 1969 and 1970, Hess joined with others, including Murray Rothbard, Robert LeFevre, Dana Rohrabacher, Samuel Edward Konkin III and former SDS leader Carl Oglesby to speak at two “left-right” conferences which brought together activists from both the Old Right and the New Left in what was emerging as a nascent libertarian movement.[245] As part of his effort to unite right and left-libertarianism, Hess would join the SDS as well as the Industrial Workers of the World (IWW), of which he explained: “We used to have a labor movement in this country, until I.W.W. leaders were killed or imprisoned. You could tell labor unions had become captive when business and government began to praise them. They’re destroying the militant black leaders the same way now. If the slaughter continues, before long liberals will be asking, ‘What happened to the blacks? Why aren’t they militant anymore?'”.[246] Rothbard ultimately broke with the left, allying himself instead with the burgeoning paleoconservative movement.[247] He criticized the tendency of these left-libertarians to appeal to “‘free spirits,’ to people who don’t want to push other people around, and who don’t want to be pushed around themselves” in contrast to “the bulk of Americans,” who “might well be tight-assed conformists, who want to stamp out drugs in their vicinity, kick out people with strange dress habits, etc”.[248] This left-libertarian tradition has been carried to the present day by Samuel Edward Konkin III’s agorists, contemporary mutualists such as Kevin Carson and Roderick T. Long and other left-wing market anarchists.[249]

In 1971, a small group of Americans led by David Nolan formed the Libertarian Party,[250] which has run a presidential candidate every election year since 1972. Other libertarian organizations, such as the Center for Libertarian Studies and the Cato Institute, were also formed in the 1970s.[251] Philosopher John Hospers, a one-time member of Rand’s inner circle, proposed a non-initiation of force principle to unite both groups, but this statement later became a required “pledge” for candidates of the Libertarian Party and Hospers became its first presidential candidate in 1972.[citation needed] In the 1980s, Hess joined the Libertarian Party and served as editor of its newspaper from 1986 to 1990.

Modern libertarianism gained significant recognition in academia with the publication of Harvard University professor Robert Nozick’s Anarchy, State, and Utopia in 1974, for which he received a National Book Award in 1975.[252] In response to John Rawls’s A Theory of Justice, Nozick’s book supported a nightwatchman state on the grounds that it was an inevitable phenomenon which could arise without violating individual rights.[253]

In the early 1970s, Rothbard wrote that “[o]ne gratifying aspect of our rise to some prominence is that, for the first time in my memory, we, ‘our side,’ had captured a crucial word from the enemy… ‘Libertarians’… had long been simply a polite word for left-wing anarchists, that is for anti-private property anarchists, either of the communist or syndicalist variety. But now we had taken it over”.[254] Since the resurgence of neoliberalism in the 1970s, this modern American libertarianism has spread beyond North America via think tanks and political parties.[255][256]

A surge of popular interest in libertarian socialism occurred in western nations during the 1960s and 1970s.[257] Anarchism was influential in the Counterculture of the 1960s[258][259][260] and anarchists actively participated in the late sixties students and workers revolts.[261] In 1968, the International of Anarchist Federations was founded in Carrara, Italy during an international anarchist conference held there in 1968 by the three existing European federations of France, the Italian and the Iberian Anarchist Federation as well as the Bulgarian federation in French exile.[173][262] The uprisings of May 1968 also led to a small resurgence of interest in left communist ideas. Various small left communist groups emerged around the world, predominantly in the leading capitalist countries. A series of conferences of the communist left began in 1976, with the aim of promoting international and cross-tendency discussion, but these petered out in the 1980s without having increased the profile of the movement or its unity of ideas.[263] Left communist groups existing today include the International Communist Party, International Communist Current and the Internationalist Communist Tendency. The housing and employment crisis in most of Western Europe led to the formation of communes and squatter movements like that of Barcelona, Spain. In Denmark, squatters occupied a disused military base and declared the Freetown Christiania, an autonomous haven in central Copenhagen.

Around the turn of the 21st century, libertarian socialism grew in popularity and influence as part of the anti-war, anti-capitalist and anti-globalisation movements.[264] Anarchists became known for their involvement in protests against the meetings of the World Trade Organization (WTO), Group of Eight and the World Economic Forum. Some anarchist factions at these protests engaged in rioting, property destruction and violent confrontations with police. These actions were precipitated by ad hoc, leaderless, anonymous cadres known as black blocs and other organisational tactics pioneered in this time include security culture, affinity groups and the use of decentralised technologies such as the internet.[264] A significant event of this period was the confrontations at WTO conference in Seattle in 1999.[264] For English anarchist scholar Simon Critchley, “contemporary anarchism can be seen as a powerful critique of the pseudo-libertarianism of contemporary neo-liberalism…One might say that contemporary anarchism is about responsibility, whether sexual, ecological or socio-economic; it flows from an experience of conscience about the manifold ways in which the West ravages the rest; it is an ethical outrage at the yawning inequality, impoverishment and disenfranchisment that is so palpable locally and globally”.[265] This might also have been motivated by “the collapse of ‘really existing socialism’ and the capitulation to neo-liberalism of Western social democracy”.[266]

Libertarian socialists in the early 21st century have been involved in the alter-globalization movement, squatter movement; social centers; infoshops; anti-poverty groups such as Ontario Coalition Against Poverty and Food Not Bombs; tenants’ unions; housing cooperatives; intentional communities generally and egalitarian communities; anti-sexist organizing; grassroots media initiatives; digital media and computer activism; experiments in participatory economics; anti-racist and anti-fascist groups like Anti-Racist Action and Anti-Fascist Action; activist groups protecting the rights of immigrants and promoting the free movement of people, such as the No Border network; worker co-operatives, countercultural and artist groups; and the peace movement.

In the United States, polls (circa 2006) find that the views and voting habits of between 10 and 20 percent (and increasing) of voting age Americans may be classified as “fiscally conservative and socially liberal, or libertarian”.[267][268] This is based on pollsters and researchers defining libertarian views as fiscally conservative and socially liberal (based on the common United States meanings of the terms) and against government intervention in economic affairs and for expansion of personal freedoms.[267] Through 20 polls on this topic spanning 13 years, Gallup found that voters who are libertarian on the political spectrum ranged from 1723% of the United States electorate.[269] However, a 2014 Pew Poll found that 23% of Americans who identify as libertarians have no idea what the word means.[270]

2009 saw the rise of the Tea Party movement, an American political movement known for advocating a reduction in the United States national debt and federal budget deficit by reducing government spending and taxes, which had a significant libertarian component[271] despite having contrasts with libertarian values and views in some areas, such as nationalism, free trade, social issues and immigration.[272] A 2011 Reason-Rupe poll found that among those who self-identified as Tea Party supporters, 41 percent leaned libertarian and 59 percent socially conservative.[273] The movement, named after the Boston Tea Party, also contains conservative[274] and populist elements[275] and has sponsored multiple protests and supported various political candidates since 2009. Tea Party activities have declined since 2010 with the number of chapters across the country slipping from about 1,000 to 600.[276][277] Mostly, Tea Party organizations are said to have shifted away from national demonstrations to local issues.[276] Following the selection of Paul Ryan as Mitt Romney’s 2012 vice presidential running mate, The New York Times declared that Tea Party lawmakers are no longer a fringe of the conservative coalition, but now “indisputably at the core of the modern Republican Party”.[278]

In 2012, anti-war presidential candidates (Libertarian Republican Ron Paul and Libertarian Party candidate Gary Johnson) raised millions of dollars and garnered millions of votes despite opposition to their obtaining ballot access by Democrats and Republicans.[279] The 2012 Libertarian National Convention, which saw Gary Johnson and James P. Gray nominated as the 2012 presidential ticket for the Libertarian Party, resulted in the most successful result for a third-party presidential candidacy since 2000 and the best in the Libertarian Party’s history by vote number. Johnson received 1% of the popular vote, amounting to more than 1.2 million votes.[280][281] Johnson has expressed a desire to win at least 5 percent of the vote so that the Libertarian Party candidates could get equal ballot access and federal funding, thus subsequently ending the two-party system.[282][283][284]

Since the 1950s, many American libertarian organizations have adopted a free market stance, as well as supporting civil liberties and non-interventionist foreign policies. These include the Ludwig von Mises Institute, Francisco Marroqun University, the Foundation for Economic Education, Center for Libertarian Studies, the Cato Institute and Liberty International. The activist Free State Project, formed in 2001, works to bring 20,000 libertarians to New Hampshire to influence state policy.[285] Active student organizations include Students for Liberty and Young Americans for Liberty.

A number of countries have libertarian parties that run candidates for political office. In the United States, the Libertarian Party was formed in 1972 and is the third largest[286][287] American political party, with over 370,000 registered voters in the 35 states that allow registration as a Libertarian[288] and has hundreds of party candidates elected or appointed to public office.[289]

Current international anarchist federations which sometimes identify themselves as libertarian include the International of Anarchist Federations, the International Workers’ Association, and International Libertarian Solidarity. The largest organised anarchist movement today is in Spain, in the form of the Confederacin General del Trabajo (CGT) and the CNT. CGT membership was estimated to be around 100,000 for 2003.[290] Other active syndicalist movements include the Central Organisation of the Workers of Sweden and the Swedish Anarcho-syndicalist Youth Federation in Sweden; the Unione Sindacale Italiana in Italy; Workers Solidarity Alliance in the United States; and Solidarity Federation in the United Kingdom. The revolutionary industrial unionist Industrial Workers of the World claiming 2,000 paying members as well as the International Workers Association, an anarcho-syndicalist successor to the First International, also remain active. In the United States, there exists the Common Struggle Libertarian Communist Federation.

Criticism of libertarianism includes ethical, economic, environmental, pragmatic, and philosophical concerns.[291] It has also been argued that laissez-faire capitalism does not necessarily produce the best or most efficient outcome,[292] nor does its policy of deregulation prevent the abuse of natural resources. Furthermore, libertarianism has been criticized as utopian due to the lack of any such societies today.

Critics such as Corey Robin describe right-libertarianism as fundamentally a reactionary conservative ideology, united with more traditional conservative thought and goals by a desire to enforce hierarchical power and social relations:[293]

Conservatism, then, is not a commitment to limited government and libertyor a wariness of change, a belief in evolutionary reform, or a politics of virtue. These may be the byproducts of conservatism, one or more of its historically specific and ever-changing modes of expression. But they are not its animating purpose. Neither is conservatism a makeshift fusion of capitalists, Christians, and warriors, for that fusion is impelled by a more elemental forcethe opposition to the liberation of men and women from the fetters of their superiors, particularly in the private sphere. Such a view might seem miles away from the libertarian defense of the free market, with its celebration of the atomistic and autonomous individual. But it is not. When the libertarian looks out upon society, he does not see isolated individuals; he sees private, often hierarchical, groups, where a father governs his family and an owner his employees.

John Donahue argues that if political power were radically shifted to local authorities, parochial local interests would predominate at the expense of the whole and that this would exacerbate current problems with collective action.[294]

Michael Lind has observed that of the 195 countries in the world today, none have fully actualized a libertarian society:

If libertarianism was a good idea, wouldn’t at least one country have tried it? Wouldn’t there be at least one country, out of nearly two hundred, with minimal government, free trade, open borders, decriminalized drugs, no welfare state and no public education system?[295]

Lind has also criticised libertarianism, particularly the right-wing and free market variant of the ideology, as being incompatible with democracy and apologetic towards autocracy.[296]

See the article here:

Libertarianism – Wikipedia

Can Libertarianism Be a Governing Philosophy?

The discussion we are about to have naturally divides itself into two aspects:

First: Could libertarianism, if implemented, sustain a state apparatus and not devolve into autocracy or anarchy? By that I mean the lawless versions of autocracy and anarchy, not stable monarchy or emergent rule of law without a state. Second: even if the answer were Yesor, Yes, if . . . we would still need to know whether enough citizens desired a libertarian order that it could feasibly be voluntarily chosen. That is, I am ruling out involuntary imposition by force of libertarianism as a governing philosophy.

I will address both questions, but want to assert at the outset that the first is the more important and more fundamental one. If the answer to it is No, there is no point in moving on to the second question. If the answer is Yes, it may be possible to change peoples minds about accepting a libertarian order.

The Destinationalists

As I have argued elsewhere[1], there are two main paths to deriving libertarian principles, destinations and directions. The destinationist approach shares the method of most other ethical paradigms: the enunciation of timeless moral and ethical precepts that describe the ideal libertarian society.

What makes for a distinctly libertarian set of principles is two precepts:

The extreme forms of these principles, for destinationists, can be hard for outsiders to accept. One example is noted by Matt Zwolinski, who cites opinion data gathered from libertarians by Liberty magazine and presented in its periodic Liberty Poll. A survey question frequently included in the survey was:

Suppose that you are on a friends balcony on the 50th floor of a condominium complex. You trip, stumble and fall over the edge. You catch a flagpole on the next floor down. The owner opens his window and demands you stop trespassing.

Zwolinski writes that in 1988, 84 percent of respondents to the flagpole question

said they believed that in such circumstances they should enter the owners residence against the owners wishes. 2% (one respondent) said that they should let go and fall to their death, and 15% said they should hang on and wait for somebody to throw them a rope. In 1999, the numbers were 86%, 1%, and 13%. In 2008, they were 89.2%, 0.9%, and 9.9%.

The interesting thing is that, while the answers to the flagpole question were almost unchanged over time, with a slight upward drift in those who would aggress by trespassing, support for the non-aggression principle itself plummeted. Writes Zwolinski:

Respondents were asked to say whether they agreed or disagreed with [the non-aggression principle]. In 1988, a full 90% of respondents said that they agreed. By 1999, however, the percentage expressing agreement had dropped by almost half to 50%. And by 2008, it was down to 39.7%.

If we take support for the non-aggression principle as a Rorschach test, it does not appear that most people, maybe not even everyone who identifies as a libertarian, are fully convinced that the principle is an absolute categorical moral principle.

The Directionalists

Of course, it could be true that many who identify now as libertarians, and those who might be attracted to libertarianism in the future, are directionalists. A directional approach holds that any policy action that increases the liberty and welfare of individuals is an improvement, and should be supported by libertarians, even if the policy itself violates either the self-ownership principle or the non-aggression principle.

A useful example here might be school vouchers. Instead of being a monopoly provider of public school education, the state might specialize in funding but leave the provision of education at least partly to private sector actors. The destinationist would object (and correctly) that the policy still involves the initiation of violence in collecting taxes involuntarily imposed on at least individuals who would not pay without the threat of coercion. In contrast, the directionalist might support vouchers, since parents would at least be afforded more liberty in choosing schools for their children, and the system would be subject to more competition, thus holding providers responsible for the quality of education being delivered.

Here, then, is a slightly modified take on the central question: Would a hybrid version of libertarianism, one that advocated for the destination but accepted directional improvements, be a viable governing philosophy? Even with this amendment, allowing for directional improvements as part of the core governing philosophy, is libertarianismto use a trope of the momentsustainable? The reason this approach could be useful is that it correlates to one of the great divisions within the libertarian movement: the split between political anarchists, who believe that any coercive state apparatus is ultimately incompatible with liberty, and the minarchists, who believe that a limited government is desirable, even necessary, and that it is also possible.

Limiting Leviathan: Getting Power to Stay Where You Put It

For a state to be consistent with both the self-ownership principle and the non-aggression principle, there must be certain core rights to property, expression, and action that are inviolable. This inviolability extends even to situations where initiating force would greatly benefit most people, meaning that consequentialist considerations cannot outweigh the rights of individuals.

Where might such a state originate, and how could it be continually limited to only those functions for which it was originally justified? One common answer is a form of contractarianism. (Another is convention, which is beyond the scope in this essay. See Robert Sugden[2] and Gerard Gaus[3] for a review of some of the issues.) This is not to say that actual states are the results of explicitly contractual arrangements; rather, there is an as if element: rational citizens in a state of nature would have voluntarily consented to the limited coercion of a minarchist state, given the substantial and universal improvement in welfare that results from having a provider of public goods and a neutral enforcer of contracts. Without a state, claims the minarchist, these two functionspublic goods provision and contract enforcementare either impossible or so difficult as to make the move to create a coercive state universally welcome for all citizens.

Contractarianism is of course an enormous body of work in philosophy, ranging from Thomas Hobbes and Jean-Jacques Rousseau to David Gauthier and John Rawls. Our contractarians, the libertarian versions, start with James Buchanan and Jan Narveson. Buchanans contractarianism is stark: Rules start with us, and the justification for coercion is, but can only be, our consent to being coerced. It is not clear that Buchanan would accept the full justification of political authority by tacit contract, but Buchanan also claims that each group in society should start from where we are now, meaning that changes in the rules require something as close to unanimous consent as possible.[4]

Narvesons view is closer to the necessary evil claim for justifying government. We need a way to be secure from violence, and to be able to enter into binding agreements that are enforceable. He wrote in The Libertarian Idea (1988) that there is no alternative that can provide reasons to everyone for accepting it, no matter what their personal values or philosophy of life may be, and thus motivating this informal, yet society-wide institution. He goes on to say:

Without resort to obfuscating intuitions, of self-evident rights and the like, the contractarian view offers an intelligible account both of why it is rational to want a morality and of what, broadly speaking, the essentials of that morality must consist in: namely, those general rules that are universally advantageous to rational agents. We each need morality, first because we are vulnerable to the depredations of others, and second because we can all benefit from cooperation with others. So we need protection, in the form of the ability to rely on our fellows not to engage in activities harmful to us; and we need to be able to rely on those with whom we deal. We each need this regardless of what else we need or value.

The problem, or so the principled political anarchist would answer, is that Leviathan cannot be limited unless for some reason Leviathan wants to limit itself.

One of the most interesting proponent of this view is Anthony de Jasay, an independent philosopher of political economy. Jasay would not dispute the value of credible commitments for contracts. His quarrel comes when contractarians invoke a founding myth. When I think of the Social Contract (the capitals signify how important it is!), I am reminded of that scene from Monty Python where King Arthur is talking to the peasants:

King Arthur: I am your king.

Woman: Well, I didnt vote for you.

King Arthur: You dont vote for kings.

Woman: Well howd you become king then?

[holy music . . . ]

King Arthur: The Lady of the Lake, her arm clad in the purest shimmering samite held aloft Excalibur from the bosom of the water, signifying by divine providence that I, Arthur, was to carry Excalibur. That is why I am your king.

Dennis: [interrupting] Listen, strange women lyin in ponds distributin swords is no basis for a system of government. Supreme executive power derives from a mandate from the masses, not from some farcical aquatic ceremony.

According to Jasay, there are two distinct problems with contractarian justifications for the state. Each, separately and independently, is fatal for the project, in his view. Together they put paid to the notion that a libertarian could favor minarchism.

The first problem is the enforceable contracts justification. The second is the limiting Leviathan problem.

The usual statement of the first comes from Hobbes: Covenants, without the sword, are but words. That means that individuals cannot enter into binding agreements without some third party to enforce the agreement. Since entering into binding agreements is a central precondition for mutually beneficial exchange and broad-scale market cooperation, we need a powerful, neutral enforcer. So, we all agree on that; the enforcer collects the taxes that we all agreed on and, in exchange, enforces all our contracts for us. (See John Thrasher[5] for some caveats.)

Butwait. Jasay compares this to jumping over your own shadow. If contracts cannot be enforced save by coercion from a third party, how can the contract between citizens and the state be enforced? [I]t takes courage to affirm that rational people could unanimously wish to have a sovereign contract enforcer bound by no contract, wrote Jasay in his book Against Politics (1997). By courage he does not intend a compliment. Either those who make this claim are contradicting themselves (since we cant have contracts, well use a contract to solve the problem) or the argument is circular (cooperation requires enforceable contracts, but these require a norm of cooperation).

Jasay put the question this way in On Treating Like Cases Alike: Review of Politics by Principle Not Interest, his 1999 essay in the Independent Review:

If man can no more bind himself by contract than he can jump over his own shadow, how can he jump over his own shadow and bind himself in a social contract? He cannot be both incapable of collective action and capable of it when creating the coercive agency needed to enforce his commitment. One can, without resorting to a bootstrap theory, accept the idea of an exogenous coercive agent, a conqueror whose regime is better than anything the conquered people could organize for themselves. Consenting to such an accomplished fact, however, can hardly be represented as entering into a contract, complete with a contracts ethical implications of an act of free will. [Emphasis in original]

In sum, the former claimthat contracts cannot be enforcedcannot then be used to conjure enforceable contracts out of a shadow. The latter claimthat people will cooperate on their ownmeans that no state is necessary in the first place. The conclusion Jasay reaches is that states, if they exist, may well be able to compel people to obey. The usual argument goes like this:

The state exists and enjoys the monopoly of the use of force for some reason, probably a historical one, that we need not inquire into. What matters is that without the state, society could not function tolerably, if at all. Therefore all rational persons would choose to enter into a social contract to create it. Indeed, we should regard the state as if it were the result of our social contract, hence indisputably legitimate.[6]

Jasay concludes that this argument must be false. As Robert Nozick famously put it in Anarchy, State, and Utopia (1974), tacit consent isnt worth the paper its not written on. We cannot confect a claim that states deserve our obedience based on consent. For consent is what true political authority requires: not that our compliance can be compelled, but that the state deserves our compliance. Ordered anarchy with no formal state is therefore a better solution, in Jasays view, because consent is either not real or is not enough.

Of course, this is simply an extension of a long tradition in libertarian thought, dating at least to Lysander Spooner. As Spooner said:

If the majority, however large, of the people of a country, enter into a contract of government, called a constitution, by which they agree to aid, abet or accomplish any kind of injustice, or to destroy or invade the natural rights of any person or persons whatsoever, whether such persons be parties to the compact or not, this contract of government is unlawful and voidand for the same reason that a treaty between two nations for a similar purpose, or a contract of the same nature between two individuals, is unlawful and void. Such a contract of government has no moral sanction. It confers no rightful authority upon those appointed to administer it. It confers no legal or moral rights, and imposes no legal or moral obligation upon the people who are parties to it. The only duties, which any one can owe to it, or to the government established under color of its authority, are disobedience, resistance, destruction.[7]

Now for the other problem highlighted by Jasay, that of limiting Leviathan. Let us assume the best of state officials: that they genuinely intend to do good. We might make the standard Public Choice assumption that officials want to use power to benefit themselves, but let us put that aside; instead, officials genuinely want to improve the lives of their citizens.

This means a minarchist state is not sustainable. Officials, thinking of the society as a collective rather than as individuals with inviolable rights, will immediately discover opportunities to raise taxes, and create new programs and new powers that benefit those in need. In fact, it is precisely the failure of the Public Choice assumptions of narrow self-interest that ensure this outcome. It might be possible in theory to design a principal-agent system of bureaucratic contract that constrains selfish officials. But if state power attracts those who are willing to sacrifice the lives or welfare of some for the greater good, then minarchy is quickly breached and Leviathan swells without the possibility of constraint.

I hasten to add that it need not be true, for Jasays claim to go through, that the concept of the greater good have any empirical content. It is enough that a few people believe, and can brandish the greater good like a truncheon, smashing rules and laws designed to stop the expansion of state power. No one who wants to do good will pass up a chance to do good, even if it means changing the rules. This process is much like that described by F.A. Hayek in Why the Worst Get on Top (see Chapter 10 of The Road to Serfdom) or Bertrand de Jouvenels Power (1945).

So, again, we reach a contradiction: Either 1) minarchy is not possible, because it is overwhelmed by the desire to do good, or minarchy is not legitimate because it is based on a mythical tacit consent; or 2) no state, minarchist or otherwise, is necessary because people can limit their actions on their own. Citizens might conclude that such self-imposed limits on their own actions are morally required, and that reputation and competition can limit the extent of depredation and reward cooperation in settings with repeated interaction. Jasay would argue, then, that constitutions and parchment barriers are either unnecessary (if people are self-governing) or ineffective (if they are not). Leviathan either cannot exist or else it is illimitable.

But Thats Not Enough

What I have argued so far is that destinationist libertarianism that is fully faithful to the self-ownership principle and the non-aggression principle could not be an effective governing philosophy. The only exception to this claim would be if libertarianism were universally believed, and people all agreed to govern themselves in the absence of a coercive state apparatus of any kind. Of course, one could object that even then something like a state would emerge, because of the economies of scale in the provision of defense, leading to a dominant protection network as described by Nozick. Whether that structure of service-delivery is necessarily a state is an interesting question, but not central to our current inquiry.

My own view is that libertarianism is, and in fact should be, a philosophy of governing that is robust and useful. But then I am a thoroughgoing directionalist. The state and its deputized coercive instruments have expanded the scope and intensity of their activities far beyond what people need to achieve cooperative goals, and beyond what they want in terms of immanent intrusions into our private lives.

Given the constant push and pull of politics, and the desire of groups to create and maintain rents for themselves, the task of leaning into the prevailing winds of statism will never be done. But it is a coherent and useful governing philosophy. When someone asks how big the state should be, there arent many people who think the answer is zero. But thats not on the table, anyway. My answer is smaller than it is now. Any policy change that grants greater autonomy (but also responsibility) to individual citizens, or that lessens government control over private action, is desirable; and libertarians are crucial for providing compelling intellectual justifications for why this is so.

In short, I dont advocate abandoning destinationist debates. The positing of an ideal is an important device for recruitment and discussion. But at this point we have been going in the wrong direction, for decades. It should be possible to find allies and fellow travelers. They may want to get off the train long before we arrive at the end of the line, but for many miles our paths toward smaller government follow the same track.

[1] Michael Munger, Basic Income Is Not an Obligation, but It Might Be a Legitimate Choice, Basic Income Studies 6:2 (December 2011), 1-13.

[2] Robert Sugden, Can a Humean Be a Contractarian? in Perspectives in Moral Science, edited by Michael Baurmann and Bernd Lahno, Frankfurt School Verlag (2009), 1123.

[3] Gerald Gaus, Why the Conventionalist Needs the Social Contract (and Vice Versa), Rationality, Markets and Morals, Frankfurt School Verlag, 4 (2013), 7187.

[4] For more on the foundation of Buchanans thought, see my forthcoming essay in the Review of Austrian Economics, Thirty Years After the Nobel: James Buchanans Political Philosophy.

[5] John Thrasher, Uniqueness and Symmetry in Bargaining Theories of Justice, Philosophical Studies 167 (2014), 683699.

[6] Anthony de Jasay, Pious Lies: The Justification of States and Welfare States, Economic Affairs 24:2 (2004), 63-64.

[7] Lysander Spooner, The Unconstitutionality of Slavery (Boston: Bela Marsh, 1860), pp. 9-10.

Visit link:

Can Libertarianism Be a Governing Philosophy?

6 Reasons Why I Gave Up On Libertarianism Return Of Kings

These days, libertarianism tends to be quite discredited. It is now associated with the goofy candidature of Gary Johnson, having a rather narrow range of issueslegalize weed! less taxes!, cucking ones way to politics through sweeping all the embarrassing problems under the carpet, then surrendering to liberal virtue-signaling and endorsing anti-white diversity.

Now, everyone on the Alt-Right, manosphere und so wieser is laughing at those whose adhesion to a bunch of abstract premises leads to endorse globalist capital, and now that Trump officially heads the State, wed be better off if some private companies were nationalized than let to shadowy overlords.

To Americans, libertarianism has been a constant background presence. Its main icons, be them Ayn Rand, Murray Rothbard or Friedrich Hayek, were always read and discussed here and there, and never fell into oblivion although they barely had media attention. The academic and political standing of libertarianism may be marginal, it has always been granted small platforms and resurrected from time to time in the public landscape, one of the most conspicuous examples of it being the Tea Party demonstrations.

To a frog like yours trulyKek being now praised by thousands of well-meaning memers, I can embrace the frog moniker gladlylibertarianism does not have the same standing at all. In French universities, libertarian thinkers are barely discussed, even in classes that are supposed to tackle economics: for one hour spent talking about Hayek, Keynes easily enjoys ten, and the same goes on when comparing the attention given to, respectively, Adam Smith and Karl Marx.

On a wider perspective, a lot of the contemporary French identity is built on Jacobinism, i.e. on crushing underfoot organic regional sociability in the name of a bureaucratized and Masonic republic. The artificial construction of France is exactly the kind of endeavour libertarianism loathes. No matter why the public choices school, for example, is barely studied here: pompous leftist teachers and mediocre fonctionnaires are too busy gushing about themselves, sometimes hiding the emptiness of their life behind a ridiculous epic narrative that turns social achievements into heroic feats, to give a fair hearing to pertinent criticism.

When I found out about libertarianism, I was already sick of the dominant fifty shades of leftism political culture. The gloomy mediocrity of small bureaucrats, including most school teachers, combined with their petty political righteousness, always repelled me. Thus, the discovery oflaissez-faire advocates felt like stumbling on an entirely new scene of thoughtand my initial feeling was vindicated when I found about the naturalism often associated with it, something refreshing and intuitively more satisfying than the mainstream culture-obsessed, biology-denying view.

Libertarianism looked like it could solve everything. More entrepreneurship, more rights to those who actually create wealth and live through the good values of personal responsibility and work ethic, less parasitesbe they bureaucrats or immigrants, no more repressive speech laws. Coincidentally, a new translation of Ayn Rands Atlas Shrugged was published at this time: I devoured it, loving the sense of life, the heroism, the epic, the generally great and achieving ethos contained in it. Arent John Galt and Hank Rearden more appealing than any corrupt politician or beta bureaucrat that pretends to be altruistic while backstabbing his own colleagues and parasitizing the country?

Now, although I still support small-scale entrepreneurship wholeheartedly, I would never defend naked libertarianism, and here is why.

Part of the Rothschild family, where nepotism and consanguinity keep the money in

Unity makes strength, and trust is much easier to cultivate in a small group where everyone truly belongs than in an anonymous great society. Some ethnic groups, especially whites, tend to be instinctively individualistic, with a lot of people favouring personal liberty over belonging, while others, especially Jews, tend to favor extended family business and nepotism.

On a short-term basis, mobile individuals can do better than those who are bound to many social obligations. On the long run, however, extended families manage to create an environment of trust and concentrate capital. And whereas individuals may start cheating each other or scattering their wealth away, thanks to having no proper economic network, families and tribes will be able to invest heavily in some of their members and keep their wealth inside. This has been true for Jewish families, wherever their members work as moneylenders or diamond dealers, for Asians investing in new restaurants or any other business project of their own, and for North Africans taking over pubs and small shops in France.

The latter example is especially telling. White bartenders, butchers, grocers and the like have been chased off French suburbs by daily North African and black violence. No one helped them, everyone being afraid of getting harassed as well and busy with their own business. (Yep, just like what happened and still happens in Rotheram.) As a result, these isolated, unprotected shop-owners sold their outlet for a cheap price and fled. North Africans always covered each others violence and replied in groups against any hurdle, whereas whites lowered their heads and hoped not to be next on the list.

Atlas Shrugged was wrong. Loners get wrecked by groups. Packs of hyenas corner and eat the lone dog.

Libertarianism is not good for individuals on the long runit turns them into asocial weaklings, soon to be legally enslaved by global companies or beaten by groups, be they made of nepotistic family members or thugs.

How the middle classes end up after jobs have been sent overseas and wages lowered

People often believe, thanks to Leftist media and cuckservative posturing, that libertarians are big bosses. This is mostly, if not entirely, false. Most libertarians are middle class guys who want more opportunities, less taxation, and believe that libertarianism will help them to turn into successful entrepreneurs. They may be right in very specific circumstances: during the 2000s, small companies overturned the market of electronics, thus benefiting both to their independent founders and to society as a whole; but ultimately, they got bought by giants like Apple and Google, who are much better off when backed by a corrupt State than on a truly free market.

Libertarianism is a fake alternative, just as impossible to realize as communism: far from putting everyone at its place, it lets ample room to mafias, monopolies, unemployment caused by mechanization and global competition. If one wants the middle classes to survive, one must protect the employment and relative independence of its membersbankers and billionaires be damned.

Spontaneous order helped by a weak government. I hope they at least smoke weed.

A good feature of libertarianism is that it usually goes along with a positive stance on biology and human nature, in contrast with the everything is cultural and ought to be deconstructed left. However, this stance often leads to an exaggerated optimism about human nature. In a society of laissez-faire, the libertarians say, people flourish and the order appears spontaneously.

Well, this is plainly false. As all of the great religions say, after what Christians call the Fall, man is a sinner. If you let children flourish without moral standards and role models, they become spoiled, entitled, manipulative, emotionally fragile and deprived of self-control. If you let women flourish without suspicion, you let free rein to their propensities to hypergamy, hysteria, self-entitlement and everything we can witness in them today. If you let men do as they please, you let them become greedy, envious, and turning into bullies. As a Muslim proverb says, people must be flogged to enter into paradiseand as Aristotle put forth, virtues are trained dispositions, no matter the magnitude of innate talents and propensities.

Michelle The Man Obama and Lying Crooked at a Democrat meeting

When the laissez-faire rules, some will succeed on the market more than others, due to differences in investment, work, and natural abilities. Some will succeed enough to be able to buy someone elses business: this is the natural consequence of differences in wealth and of greed. When corrupt politicians enter the game, things become worse, as they will usually help some large business owners to shield their position against competitorsat the expense of most people, who then lose their independence and live off a wage.

At the end, what we get is a handful of very wealthy individuals who have managed to concentrate most capital and power levers into their hands and a big crowd of low-wage employees ready to cut each others throat for a small promotion, and females waiting in line to get notched by the one per cent while finding the other ninety-nine per cent boring.

Censorship by massive social pressure, monopoly over the institutions and crybullying is perfectly legal. What could go wrong?

On the surface, libertarianism looks good here, because it protects the individuals rights against left-hailing Statism and cuts off the welfare programs that have attracted dozens of millions of immigrants. Beneath, however, things are quite dire. Libertarianism enshrines the leftists right to free speech they abuse from, allows the pressure tactics used by radicals, and lets freethinking individuals getting singled out by SJWs as long as these do not resort to overt stealing or overt physical violence. As for the immigrants, libertarianism tends to oppose the very notion of non-private boundaries, thus letting the local cultures and identities defenseless against both greedy capitalists and subproletarian masses.

Supporting an ideology that allows the leftists to destroy society more or less legally equates to cucking, plain and simple. Desiring an ephemeral cohabitation with rabid ideological warriors is stupid. We should aim at a lasting victory, not at pretending to constrain them through useless means.

Am I the only one to find that Gary Johnson looks like a snail (Spongebob notwithstanding)?

In 2013, one of the rare French libertarians academic teachers, Jean-Louis Caccomo, was forced into a mental ward at the request of his university president. He then spent more than a year getting drugged. Mr. Caccomo had no real psychological problem: his confinement was part of a vicious strategy of pathologization and career-destruction that was already used by the Soviets. French libertarians could have wide denounced the abuse. Nonetheless, most of them freaked out, and almost no one dared to actually defend him publicly.

Why should rational egoists team up and risk their careers to defend one of themselves after all? They would rather posture at confidential social events, rail at organic solidarity and protectionism, or trolling the shit out of individuals of their own social milieu because Ive got the right to mock X, its my right to free speech! The few libertarian people I knew firsthand, the few events I have witnessed in that small milieu, were enough to give me serious doubts about libertarianism: how can a good political ideology breed such an unhealthy mindset?

Political ideologies are tools. They are not ends in themselves. All forms of government arent fit for any people or any era. Political actors must know at least the most important ones to get some inspiration, but ultimately, said actors win on the ground, not in philosophical debates.

Individualism, mindless consumerism, careerism, hedonism are part of the problem. Individual rights granted regardless of ones abilities, situation, and identity are a disaster. Time has come to overcome modernity, not stall in one of its false alternatives. The merchant caste must be regulated, though neither micromanaged or hampered by a parasitic bureaucracy nor denied its members right for small-scale independence. Individual rights must be conditional, boundaries must be restored, minority identities based on anti-white male resentment must be crushed so they cannot devour sociability from the inside again, and the pater familias must assert himself anew.

Long live the State and protectionism as long as they defend the backbone of society and healthy relationships between the sexes, and no quarter for those who think they have a right to wage grievance-mongering against us, no matter if they want to use the State or private companies. At the end, the socialism-libertarianism dichotomy is quite secondary.

Read Next: Sugar Baby Culture In The US Is Creating A Marketplace for Prostitution

Go here to see the original:

6 Reasons Why I Gave Up On Libertarianism Return Of Kings

Artificial intelligence – Wikipedia

Intelligence demonstrated by machines

Artificial intelligence (AI, also machine intelligence, MI) is intelligence demonstrated by machines, in contrast to the natural intelligence (NI) displayed by humans and animals. In computer science AI research is defined as the study of “intelligent agents”: any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] Colloquially, the term “artificial intelligence” is applied when a machine mimics “cognitive” functions that humans associate with other human minds, such as “learning” and “problem solving”.

The scope of AI is disputed: as machines become increasingly capable, tasks considered as requiring “intelligence” are often removed from the definition, a phenomenon known as the AI effect, leading to the quip, “AI is whatever hasn’t been done yet.”[3] For instance, optical character recognition is frequently excluded from “artificial intelligence”, having become a routine technology.[4] Capabilities generally classified as AI as of 2017[update] include successfully understanding human speech, competing at the highest level in strategic game systems (such as chess and Go[6]), autonomous cars, intelligent routing in content delivery network and military simulations.

Artificial intelligence was founded as an academic discipline in 1956, and in the years since has experienced several waves of optimism,[7][8] followed by disappointment and the loss of funding (known as an “AI winter”),[9][10] followed by new approaches, success and renewed funding.[8][11] For most of its history, AI research has been divided into subfields that often fail to communicate with each other.[12] These sub-fields are based on technical considerations, such as particular goals (e.g. “robotics” or “machine learning”),[13] the use of particular tools (“logic” or artificial neural networks), or deep philosophical differences.[14][15][16] Subfields have also been based on social factors (particular institutions or the work of particular researchers).[12]

The traditional problems (or goals) of AI research include reasoning, knowledge representation, planning, learning, natural language processing, perception and the ability to move and manipulate objects.[13] General intelligence is among the field’s long-term goals.[17] Approaches include statistical methods, computational intelligence, and traditional symbolic AI. Many tools are used in AI, including versions of search and mathematical optimization, artificial neural networks, and methods based on statistics, probability and economics. The AI field draws upon computer science, mathematics, psychology, linguistics, philosophy and many others.

The field was founded on the claim that human intelligence “can be so precisely described that a machine can be made to simulate it”.[18] This raises philosophical arguments about the nature of the mind and the ethics of creating artificial beings endowed with human-like intelligence which are issues that have been explored by myth, fiction and philosophy since antiquity.[19] Some people also consider AI to be a danger to humanity if it progresses unabatedly.[20] Others believe that AI, unlike previous technological revolutions, will create a risk of mass unemployment.[21]

In the twenty-first century, AI techniques have experienced a resurgence following concurrent advances in computer power, large amounts of data, and theoretical understanding; and AI techniques have become an essential part of the technology industry, helping to solve many challenging problems in computer science.[22][11]

Thought-capable artificial beings appeared as storytelling devices in antiquity,[23] and have been common in fiction, as in Mary Shelley’s Frankenstein or Karel apek’s R.U.R. (Rossum’s Universal Robots).[24] These characters and their fates raised many of the same issues now discussed in the ethics of artificial intelligence.[19]

The study of mechanical or “formal” reasoning began with philosophers and mathematicians in antiquity. The study of mathematical logic led directly to Alan Turing’s theory of computation, which suggested that a machine, by shuffling symbols as simple as “0” and “1”, could simulate any conceivable act of mathematical deduction. This insight, that digital computers can simulate any process of formal reasoning, is known as the ChurchTuring thesis.[25] Along with concurrent discoveries in neurobiology, information theory and cybernetics, this led researchers to consider the possibility of building an electronic brain. Turing proposed that “if a human could not distinguish between responses from a machine and a human, the machine could be considered intelligent”.[26] The first work that is now generally recognized as AI was McCullouch and Pitts’ 1943 formal design for Turing-complete “artificial neurons”.

The field of AI research was born at a workshop at Dartmouth College in 1956.[28] Attendees Allen Newell (CMU), Herbert Simon (CMU), John McCarthy (MIT), Marvin Minsky (MIT) and Arthur Samuel (IBM) became the founders and leaders of AI research.[29] They and their students produced programs that the press described as “astonishing”: computers were learning checkers strategies (c. 1954)[31] (and by 1959 were reportedly playing better than the average human),[32] solving word problems in algebra, proving logical theorems (Logic Theorist, first run c. 1956) and speaking English.[33] By the middle of the 1960s, research in the U.S. was heavily funded by the Department of Defense[34] and laboratories had been established around the world.[35] AI’s founders were optimistic about the future: Herbert Simon predicted, “machines will be capable, within twenty years, of doing any work a man can do”. Marvin Minsky agreed, writing, “within a generation… the problem of creating ‘artificial intelligence’ will substantially be solved”.[7]

They failed to recognize the difficulty of some of the remaining tasks. Progress slowed and in 1974, in response to the criticism of Sir James Lighthill and ongoing pressure from the US Congress to fund more productive projects, both the U.S. and British governments cut off exploratory research in AI. The next few years would later be called an “AI winter”,[9] a period when obtaining funding for AI projects was difficult.

In the early 1980s, AI research was revived by the commercial success of expert systems,[37] a form of AI program that simulated the knowledge and analytical skills of human experts. By 1985 the market for AI had reached over a billion dollars. At the same time, Japan’s fifth generation computer project inspired the U.S and British governments to restore funding for academic research.[8] However, beginning with the collapse of the Lisp Machine market in 1987, AI once again fell into disrepute, and a second, longer-lasting hiatus began.[10]

In the late 1990s and early 21st century, AI began to be used for logistics, data mining, medical diagnosis and other areas.[22] The success was due to increasing computational power (see Moore’s law), greater emphasis on solving specific problems, new ties between AI and other fields (such as statistics, economics and mathematics), and a commitment by researchers to mathematical methods and scientific standards.[38] Deep Blue became the first computer chess-playing system to beat a reigning world chess champion, Garry Kasparov on 11 May 1997.

In 2011, a Jeopardy! quiz show exhibition match, IBM’s question answering system, Watson, defeated the two greatest Jeopardy champions, Brad Rutter and Ken Jennings, by a significant margin. Faster computers, algorithmic improvements, and access to large amounts of data enabled advances in machine learning and perception; data-hungry deep learning methods started to dominate accuracy benchmarks around 2012.[41] The Kinect, which provides a 3D bodymotion interface for the Xbox 360 and the Xbox One use algorithms that emerged from lengthy AI research[42] as do intelligent personal assistants in smartphones.[43] In March 2016, AlphaGo won 4 out of 5 games of Go in a match with Go champion Lee Sedol, becoming the first computer Go-playing system to beat a professional Go player without handicaps.[6][44] In the 2017 Future of Go Summit, AlphaGo won a three-game match with Ke Jie,[45] who at the time continuously held the world No. 1 ranking for two years.[46][47] This marked the completion of a significant milestone in the development of Artificial Intelligence as Go is an extremely complex game, more so than Chess.

According to Bloomberg’s Jack Clark, 2015 was a landmark year for artificial intelligence, with the number of software projects that use AI within Google increased from a “sporadic usage” in 2012 to more than 2,700 projects. Clark also presents factual data indicating that error rates in image processing tasks have fallen significantly since 2011.[48] He attributes this to an increase in affordable neural networks, due to a rise in cloud computing infrastructure and to an increase in research tools and datasets.[11] Other cited examples include Microsoft’s development of a Skype system that can automatically translate from one language to another and Facebook’s system that can describe images to blind people.[48] In a 2017 survey, one in five companies reported they had “incorporated AI in some offerings or processes”.[49][50]

A typical AI perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] An AI’s intended goal function can be simple (“1 if the AI wins a game of Go, 0 otherwise”) or complex (“Do actions mathematically similar to the actions that got you rewards in the past”). Goals can be explicitly defined, or can be induced. If the AI is programmed for “reinforcement learning”, goals can be implicitly induced by rewarding some types of behavior and punishing others.[a] Alternatively, an evolutionary system can induce goals by using a “fitness function” to mutate and preferentially replicate high-scoring AI systems; this is similar to how animals evolved to innately desire certain goals such as finding food, or how dogs can be bred via artificial selection to possess desired traits. Some AI systems, such as nearest-neighbor, instead reason by analogy; these systems are not generally given goals, except to the degree that goals are somehow implicit in their training data. Such systems can still be benchmarked if the non-goal system is framed as a system whose “goal” is to successfully accomplish its narrow classification task.[53]

AI often revolves around the use of algorithms. An algorithm is a set of unambiguous instructions that a mechanical computer can execute.[b] A complex algorithm is often built on top of other, simpler, algorithms. A simple example of an algorithm is the following recipe for optimal play at tic-tac-toe:

Many AI algorithms are capable of learning from data; they can enhance themselves by learning new heuristics (strategies, or “rules of thumb”, that have worked well in the past), or can themselves write other algorithms. Some of the “learners” described below, including Bayesian networks, decision trees, and nearest-neighbor, could theoretically, if given infinite data, time, and memory, learn to approximate any function, including whatever combination of mathematical functions would best describe the entire world. These learners could therefore, in theory, derive all possible knowledge, by considering every possible hypothesis and matching it against the data. In practice, it is almost never possible to consider every possibility, because of the phenomenon of “combinatorial explosion”, where the amount of time needed to solve a problem grows exponentially. Much of AI research involves figuring out how to identify and avoid considering broad swaths of possibilities that are unlikely to be fruitful.[55] For example, when viewing a map and looking for the shortest driving route from Denver to New York in the East, one can in most cases skip looking at any path through San Francisco or other areas far to the West; thus, an AI wielding an pathfinding algorithm like A* can avoid the combinatorial explosion that would ensue if every possible route had to be ponderously considered in turn.[57]

The earliest (and easiest to understand) approach to AI was symbolism (such as formal logic): “If an otherwise healthy adult has a fever, then they may have influenza”. A second, more general, approach is Bayesian inference: “If the current patient has a fever, adjust the probability they have influenza in such-and-such way”. The third major approach, extremely popular in routine business AI applications, are analogizers such as SVM and nearest-neighbor: “After examining the records of known past patients whose temperature, symptoms, age, and other factors mostly match the current patient, X% of those patients turned out to have influenza”. A fourth approach is harder to intuitively understand, but is inspired by how the brain’s machinery works: the artificial neural network approach uses artificial “neurons” that can learn by comparing itself to the desired output and altering the strengths of the connections between its internal neurons to “reinforce” connections that seemed to be useful. These four main approaches can overlap with each other and with evolutionary systems; for example, neural nets can learn to make inferences, to generalize, and to make analogies. Some systems implicitly or explicitly use multiple of these approaches, alongside many other AI and non-AI algorithms;[58] the best approach is often different depending on the problem.[60]

Learning algorithms work on the basis that strategies, algorithms, and inferences that worked well in the past are likely to continue working well in the future. These inferences can be obvious, such as “since the sun rose every morning for the last 10,000 days, it will probably rise tomorrow morning as well”. They can be nuanced, such as “X% of families have geographically separate species with color variants, so there is an Y% chance that undiscovered black swans exist”. Learners also work on the basis of “Occam’s razor”: The simplest theory that explains the data is the likeliest. Therefore, to be successful, a learner must be designed such that it prefers simpler theories to complex theories, except in cases where the complex theory is proven substantially better. Settling on a bad, overly complex theory gerrymandered to fit all the past training data is known as overfitting. Many systems attempt to reduce overfitting by rewarding a theory in accordance with how well it fits the data, but penalizing the theory in accordance with how complex the theory is. Besides classic overfitting, learners can also disappoint by “learning the wrong lesson”. A toy example is that an image classifier trained only on pictures of brown horses and black cats might conclude that all brown patches are likely to be horses. A real-world example is that, unlike humans, current image classifiers don’t determine the spatial relationship between components of the picture; instead, they learn abstract patterns of pixels that humans are oblivious to, but that linearly correlate with images of certain types of real objects. Faintly superimposing such a pattern on a legitimate image results in an “adversarial” image that the system misclassifies.[c][63][64][65]

Compared with humans, existing AI lacks several features of human “commonsense reasoning”; most notably, humans have powerful mechanisms for reasoning about “nave physics” such as space, time, and physical interactions. This enables even young children to easily make inferences like “If I roll this pen off a table, it will fall on the floor”. Humans also have a powerful mechanism of “folk psychology” that helps them to interpret natural-language sentences such as “The city councilmen refused the demonstrators a permit because they advocated violence”. (A generic AI has difficulty inferring whether the councilmen or the demonstrators are the ones alleged to be advocating violence.)[68][69][70] This lack of “common knowledge” means that AI often makes different mistakes than humans make, in ways that can seem incomprehensible. For example, existing self-driving cars cannot reason about the location nor the intentions of pedestrians in the exact way that humans do, and instead must use non-human modes of reasoning to avoid accidents.[71][72][73]

The overall research goal of artificial intelligence is to create technology that allows computers and machines to function in an intelligent manner. The general problem of simulating (or creating) intelligence has been broken down into sub-problems. These consist of particular traits or capabilities that researchers expect an intelligent system to display. The traits described below have received the most attention.[13]

Early researchers developed algorithms that imitated step-by-step reasoning that humans use when they solve puzzles or make logical deductions.[74] By the late 1980s and 1990s, AI research had developed methods for dealing with uncertain or incomplete information, employing concepts from probability and economics.[75]

These algorithms proved to be insufficient for solving large reasoning problems, because they experienced a “combinatorial explosion”: they became exponentially slower as the problems grew larger.[55] In fact, even humans rarely use the step-by-step deduction that early AI research was able to model. They solve most of their problems using fast, intuitive judgements.[76]

Knowledge representation[77] and knowledge engineering[78] are central to classical AI research. Some “expert systems” attempt to gather together explicit knowledge possessed by experts in some narrow domain. In addition, some projects attempt to gather the “commonsense knowledge” known to the average person into a database containing extensive knowledge about the world. Among the things a comprehensive commonsense knowledge base would contain are: objects, properties, categories and relations between objects;[79] situations, events, states and time;[80] causes and effects;[81] knowledge about knowledge (what we know about what other people know);[82] and many other, less well researched domains. A representation of “what exists” is an ontology: the set of objects, relations, concepts, and properties formally described so that software agents can interpret them. The semantics of these are captured as description logic concepts, roles, and individuals, and typically implemented as classes, properties, and individuals in the Web Ontology Language.[83] The most general ontologies are called upper ontologies, which attempt to provide a foundation for all other knowledge[84] by acting as mediators between domain ontologies that cover specific knowledge about a particular knowledge domain (field of interest or area of concern). Such formal knowledge representations can be used in content-based indexing and retrieval,[85] scene interpretation,[86] clinical decision support,[87] knowledge discovery (mining “interesting” and actionable inferences from large databases),[88] and other areas.[89]

Among the most difficult problems in knowledge representation are:

Intelligent agents must be able to set goals and achieve them.[96] They need a way to visualize the futurea representation of the state of the world and be able to make predictions about how their actions will change itand be able to make choices that maximize the utility (or “value”) of available choices.[97]

In classical planning problems, the agent can assume that it is the only system acting in the world, allowing the agent to be certain of the consequences of its actions.[98] However, if the agent is not the only actor, then it requires that the agent can reason under uncertainty. This calls for an agent that can not only assess its environment and make predictions, but also evaluate its predictions and adapt based on its assessment.[99]

Multi-agent planning uses the cooperation and competition of many agents to achieve a given goal. Emergent behavior such as this is used by evolutionary algorithms and swarm intelligence.[100]

Machine learning, a fundamental concept of AI research since the field’s inception,[101] is the study of computer algorithms that improve automatically through experience.[102][103]

Unsupervised learning is the ability to find patterns in a stream of input. Supervised learning includes both classification and numerical regression. Classification is used to determine what category something belongs in, after seeing a number of examples of things from several categories. Regression is the attempt to produce a function that describes the relationship between inputs and outputs and predicts how the outputs should change as the inputs change.[103] Both classifiers and regression learners can be viewed as “function approximators” trying to learn an unknown (possibly implicit) function; for example, a spam classifier can be viewed as learning a function that maps from the text of an email to one of two categories, “spam” or “not spam”. Computational learning theory can assess learners by computational complexity, by sample complexity (how much data is required), or by other notions of optimization.[104] In reinforcement learning[105] the agent is rewarded for good responses and punished for bad ones. The agent uses this sequence of rewards and punishments to form a strategy for operating in its problem space.

Natural language processing[106] (NLP) gives machines the ability to read and understand human language. A sufficiently powerful natural language processing system would enable natural-language user interfaces and the acquisition of knowledge directly from human-written sources, such as newswire texts. Some straightforward applications of natural language processing include information retrieval, text mining, question answering[107] and machine translation.[108] Many current approaches use word co-occurrence frequencies to construct syntactic representations of text. “Keyword spotting” strategies for search are popular and scalable but dumb; a search query for “dog” might only match documents with the literal word “dog” and miss a document with the word “poodle”. “Lexical affinity” strategies use the occurrence of words such as “accident” to assess the sentiment of a document. Modern statistical NLP approaches can combine all these strategies as well as others, and often achieve acceptable accuracy at the page or paragraph level, but continue to lack the semantic understanding required to classify isolated sentences well. Besides the usual difficulties with encoding semantic commonsense knowledge, existing semantic NLP sometimes scales too poorly to be viable in business applications. Beyond semantic NLP, the ultimate goal of “narrative” NLP is to embody a full understanding of commonsense reasoning.[109]

Machine perception[110] is the ability to use input from sensors (such as cameras (visible spectrum or infrared), microphones, wireless signals, and active lidar, sonar, radar, and tactile sensors) to deduce aspects of the world. Applications include speech recognition,[111] facial recognition, and object recognition.[112] Computer vision is the ability to analyze visual input. Such input is usually ambiguous; a giant, fifty-meter-tall pedestrian far away may produce exactly the same pixels as a nearby normal-sized pedestrian, requiring the AI to judge the relative likelihood and reasonableness of different interpretations, for example by using its “object model” to assess that fifty-meter pedestrians do not exist.[113]

AI is heavily used in robotics.[114] Advanced robotic arms and other industrial robots, widely used in modern factories, can learn from experience how to move efficiently despite the presence of friction and gear slippage.[115] A modern mobile robot, when given a small, static, and visible environment, can easily determine its location and map its environment; however, dynamic environments, such as (in endoscopy) the interior of a patient’s breathing body, pose a greater challenge. Motion planning is the process of breaking down a movement task into “primitives” such as individual joint movements. Such movement often involves compliant motion, a process where movement requires maintaining physical contact with an object.[117][118] Moravec’s paradox generalizes that low-level sensorimotor skills that humans take for granted are, counterintuitively, difficult to program into a robot; the paradox is named after Hans Moravec, who stated in 1988 that “it is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility”.[119][120] This is attributed to the fact that, unlike checkers, physical dexterity has been a direct target of natural selection for millions of years.[121]

Moravec’s paradox can be extended to many forms of social intelligence.[123][124] Distributed multi-agent coordination of autonomous vehicles remains a difficult problem.[125] Affective computing is an interdisciplinary umbrella that comprises systems which recognize, interpret, process, or simulate human affects. Moderate successes related to affective computing include textual sentiment analysis and, more recently, multimodal affect analysis (see multimodal sentiment analysis), wherein AI classifies the affects displayed by a videotaped subject.[129]

In the long run, social skills and an understanding of human emotion and game theory would be valuable to a social agent. Being able to predict the actions of others by understanding their motives and emotional states would allow an agent to make better decisions. Some computer systems mimic human emotion and expressions to appear more sensitive to the emotional dynamics of human interaction, or to otherwise facilitate humancomputer interaction.[130] Similarly, some virtual assistants are programmed to speak conversationally or even to banter humorously; this tends to give naive users an unrealistic conception of how intelligent existing computer agents actually are.[131]

Historically, projects such as the Cyc knowledge base (1984) and the massive Japanese Fifth Generation Computer Systems initiative (19821992) attempted to cover the breadth of human cognition. These early projects failed to escape the limitations of non-quantitative symbolic logic models and, in retrospect, greatly underestimated the difficulty of cross-domain AI. Nowadays, the vast majority of current AI researchers work instead on tractable “narrow AI” applications (such as medical diagnosis or automobile navigation).[132] Many researchers predict that such “narrow AI” work in different individual domains will eventually be incorporated into a machine with artificial general intelligence (AGI), combining most of the narrow skills mentioned in this article and at some point even exceeding human ability in most or all these areas.[17][133] Many advances have general, cross-domain significance. One high-profile example is that DeepMind in the 2010s developed a “generalized artificial intelligence” that could learn many diverse Atari games on its own, and later developed a variant of the system which succeeds at sequential learning.[134][135][136] Besides transfer learning,[137] hypothetical AGI breakthroughs could include the development of reflective architectures that can engage in decision-theoretic metareasoning, and figuring out how to “slurp up” a comprehensive knowledge base from the entire unstructured Web. Some argue that some kind of (currently-undiscovered) conceptually straightforward, but mathematically difficult, “Master Algorithm” could lead to AGI. Finally, a few “emergent” approaches look to simulating human intelligence extremely closely, and believe that anthropomorphic features like an artificial brain or simulated child development may someday reach a critical point where general intelligence emerges.[139][140]

Many of the problems in this article may also require general intelligence, if machines are to solve the problems as well as people do. For example, even specific straightforward tasks, like machine translation, require that a machine read and write in both languages (NLP), follow the author’s argument (reason), know what is being talked about (knowledge), and faithfully reproduce the author’s original intent (social intelligence). A problem like machine translation is considered “AI-complete”, because all of these problems need to be solved simultaneously in order to reach human-level machine performance.

There is no established unifying theory or paradigm that guides AI research. Researchers disagree about many issues.[141] A few of the most long standing questions that have remained unanswered are these: should artificial intelligence simulate natural intelligence by studying psychology or neurobiology? Or is human biology as irrelevant to AI research as bird biology is to aeronautical engineering?[14] Can intelligent behavior be described using simple, elegant principles (such as logic or optimization)? Or does it necessarily require solving a large number of completely unrelated problems?[15]

In the 1940s and 1950s, a number of researchers explored the connection between neurobiology, information theory, and cybernetics. Some of them built machines that used electronic networks to exhibit rudimentary intelligence, such as W. Grey Walter’s turtles and the Johns Hopkins Beast. Many of these researchers gathered for meetings of the Teleological Society at Princeton University and the Ratio Club in England.[142] By 1960, this approach was largely abandoned, although elements of it would be revived in the 1980s.

When access to digital computers became possible in the middle 1950s, AI research began to explore the possibility that human intelligence could be reduced to symbol manipulation. The research was centered in three institutions: Carnegie Mellon University, Stanford and MIT, and as described below, each one developed its own style of research. John Haugeland named these symbolic approaches to AI “good old fashioned AI” or “GOFAI”.[143] During the 1960s, symbolic approaches had achieved great success at simulating high-level thinking in small demonstration programs. Approaches based on cybernetics or artificial neural networks were abandoned or pushed into the background.[144] Researchers in the 1960s and the 1970s were convinced that symbolic approaches would eventually succeed in creating a machine with artificial general intelligence and considered this the goal of their field.

Economist Herbert Simon and Allen Newell studied human problem-solving skills and attempted to formalize them, and their work laid the foundations of the field of artificial intelligence, as well as cognitive science, operations research and management science. Their research team used the results of psychological experiments to develop programs that simulated the techniques that people used to solve problems. This tradition, centered at Carnegie Mellon University would eventually culminate in the development of the Soar architecture in the middle 1980s.[145][146]

Unlike Simon and Newell, John McCarthy felt that machines did not need to simulate human thought, but should instead try to find the essence of abstract reasoning and problem solving, regardless of whether people used the same algorithms.[14] His laboratory at Stanford (SAIL) focused on using formal logic to solve a wide variety of problems, including knowledge representation, planning and learning.[147] Logic was also the focus of the work at the University of Edinburgh and elsewhere in Europe which led to the development of the programming language Prolog and the science of logic programming.[148]

Researchers at MIT (such as Marvin Minsky and Seymour Papert)[149] found that solving difficult problems in vision and natural language processing required ad-hoc solutions they argued that there was no simple and general principle (like logic) that would capture all the aspects of intelligent behavior. Roger Schank described their “anti-logic” approaches as “scruffy” (as opposed to the “neat” paradigms at CMU and Stanford).[15] Commonsense knowledge bases (such as Doug Lenat’s Cyc) are an example of “scruffy” AI, since they must be built by hand, one complicated concept at a time.[150]

When computers with large memories became available around 1970, researchers from all three traditions began to build knowledge into AI applications.[151] This “knowledge revolution” led to the development and deployment of expert systems (introduced by Edward Feigenbaum), the first truly successful form of AI software.[37] The knowledge revolution was also driven by the realization that enormous amounts of knowledge would be required by many simple AI applications.

By the 1980s progress in symbolic AI seemed to stall and many believed that symbolic systems would never be able to imitate all the processes of human cognition, especially perception, robotics, learning and pattern recognition. A number of researchers began to look into “sub-symbolic” approaches to specific AI problems.[16] Sub-symbolic methods manage to approach intelligence without specific representations of knowledge.

This includes embodied, situated, behavior-based, and nouvelle AI. Researchers from the related field of robotics, such as Rodney Brooks, rejected symbolic AI and focused on the basic engineering problems that would allow robots to move and survive.[152] Their work revived the non-symbolic viewpoint of the early cybernetics researchers of the 1950s and reintroduced the use of control theory in AI. This coincided with the development of the embodied mind thesis in the related field of cognitive science: the idea that aspects of the body (such as movement, perception and visualization) are required for higher intelligence.

Within developmental robotics, developmental learning approaches are elaborated upon to allow robots to accumulate repertoires of novel skills through autonomous self-exploration, social interaction with human teachers, and the use of guidance mechanisms (active learning, maturation, motor synergies, etc.).[153][154]

Interest in neural networks and “connectionism” was revived by David Rumelhart and others in the middle of the 1980s.[157] Artificial neural networks are an example of soft computing — they are solutions to problems which cannot be solved with complete logical certainty, and where an approximate solution is often sufficient. Other soft computing approaches to AI include fuzzy systems, evolutionary computation and many statistical tools. The application of soft computing to AI is studied collectively by the emerging discipline of computational intelligence.[158]

Much of GOFAI got bogged down on ad hoc patches to symbolic computation that worked on their own toy models but failed to generalize to real-world results. However, around the 1990s, AI researchers adopted sophisticated mathematical tools, such as hidden Markov models (HMM), information theory, and normative Bayesian decision theory to compare or to unify competing architectures. The shared mathematical language permitted a high level of collaboration with more established fields (like mathematics, economics or operations research).[d] Compared with GOFAI, new “statistical learning” techniques such as HMM and neural networks were gaining higher levels of accuracy in many practical domains such as data mining, without necessarily acquiring semantic understanding of the datasets. The increased successes with real-world data led to increasing emphasis on comparing different approaches against shared test data to see which approach performed best in a broader context than that provided by idiosyncratic toy models; AI research was becoming more scientific. Nowadays results of experiments are often rigorously measurable, and are sometimes (with difficulty) reproducible.[38][159] Different statistical learning techniques have different limitations; for example, basic HMM cannot model the infinite possible combinations of natural language. Critics note that the shift from GOFAI to statistical learning is often also a shift away from Explainable AI. In AGI research, some scholars caution against over-reliance on statistical learning, and argue that continuing research into GOFAI will still be necessary to attain general intelligence.

AI has developed a large number of tools to solve the most difficult problems in computer science. A few of the most general of these methods are discussed below.

Many problems in AI can be solved in theory by intelligently searching through many possible solutions:[168] Reasoning can be reduced to performing a search. For example, logical proof can be viewed as searching for a path that leads from premises to conclusions, where each step is the application of an inference rule.[169] Planning algorithms search through trees of goals and subgoals, attempting to find a path to a target goal, a process called means-ends analysis.[170] Robotics algorithms for moving limbs and grasping objects use local searches in configuration space.[115] Many learning algorithms use search algorithms based on optimization.

Simple exhaustive searches[171] are rarely sufficient for most real world problems: the search space (the number of places to search) quickly grows to astronomical numbers. The result is a search that is too slow or never completes. The solution, for many problems, is to use “heuristics” or “rules of thumb” that prioritize choices in favor of those that are more likely to reach a goal, and to do so in a shorter number of steps. In some search methodologies heuristics can also serve to entirely eliminate some choices that are unlikely to lead to a goal (called “pruning the search tree”). Heuristics supply the program with a “best guess” for the path on which the solution lies.[172] Heuristics limit the search for solutions into a smaller sample size.

A very different kind of search came to prominence in the 1990s, based on the mathematical theory of optimization. For many problems, it is possible to begin the search with some form of a guess and then refine the guess incrementally until no more refinements can be made. These algorithms can be visualized as blind hill climbing: we begin the search at a random point on the landscape, and then, by jumps or steps, we keep moving our guess uphill, until we reach the top. Other optimization algorithms are simulated annealing, beam search and random optimization.[173]

Evolutionary computation uses a form of optimization search. For example, they may begin with a population of organisms (the guesses) and then allow them to mutate and recombine, selecting only the fittest to survive each generation (refining the guesses). Classic evolutionary algorithms include genetic algorithms, gene expression programming, and genetic programming.[174] Alternatively, distributed search processes can coordinate via swarm intelligence algorithms. Two popular swarm algorithms used in search are particle swarm optimization (inspired by bird flocking) and ant colony optimization (inspired by ant trails).[175][176]

Logic[177] is used for knowledge representation and problem solving, but it can be applied to other problems as well. For example, the satplan algorithm uses logic for planning[178] and inductive logic programming is a method for learning.[179]

Several different forms of logic are used in AI research. Propositional logic[180] involves truth functions such as “or” and “not”. First-order logic[181] adds quantifiers and predicates, and can express facts about objects, their properties, and their relations with each other. Fuzzy set theory assigns a “degree of truth” (between 0 and 1) to vague statements such as “Alice is old” (or rich, or tall, or hungry) that are too linguistically imprecise to be completely true or false. Fuzzy logic is successfully used in control systems to allow experts to contribute vague rules such as “if you are close to the destination station and moving fast, increase the train’s brake pressure”; these vague rules can then be numerically refined within the system. Fuzzy logic fails to scale well in knowledge bases; many AI researchers question the validity of chaining fuzzy-logic inferences.[e][183][184]

Default logics, non-monotonic logics and circumscription[91] are forms of logic designed to help with default reasoning and the qualification problem. Several extensions of logic have been designed to handle specific domains of knowledge, such as: description logics;[79] situation calculus, event calculus and fluent calculus (for representing events and time);[80] causal calculus;[81] belief calculus;[185] and modal logics.[82]

Overall, qualitiative symbolic logic is brittle and scales poorly in the presence of noise or other uncertainty. Exceptions to rules are numerous, and it is difficult for logical systems to function in the presence of contradictory rules.[187]

Many problems in AI (in reasoning, planning, learning, perception and robotics) require the agent to operate with incomplete or uncertain information. AI researchers have devised a number of powerful tools to solve these problems using methods from probability theory and economics.[188]

Bayesian networks[189] are a very general tool that can be used for a large number of problems: reasoning (using the Bayesian inference algorithm),[190] learning (using the expectation-maximization algorithm),[f][192] planning (using decision networks)[193] and perception (using dynamic Bayesian networks).[194] Probabilistic algorithms can also be used for filtering, prediction, smoothing and finding explanations for streams of data, helping perception systems to analyze processes that occur over time (e.g., hidden Markov models or Kalman filters).[194] Compared with symbolic logic, formal Bayesian inference is computationally expensive. For inference to be tractable, most observations must be conditionally independent of one another. Complicated graphs with diamonds or other “loops” (undirected cycles) can require a sophisticated method such as Markov Chain Monte Carlo, which spreads an ensemble of random walkers throughout the Bayesian network and attempts to converge to an assessment of the conditional probabilities. Bayesian networks are used on XBox Live to rate and match players; wins and losses are “evidence” of how good a player is. AdSense uses a Bayesian network with over 300 million edges to learn which ads to serve.

A key concept from the science of economics is “utility”: a measure of how valuable something is to an intelligent agent. Precise mathematical tools have been developed that analyze how an agent can make choices and plan, using decision theory, decision analysis,[196] and information value theory.[97] These tools include models such as Markov decision processes,[197] dynamic decision networks,[194] game theory and mechanism design.[198]

The simplest AI applications can be divided into two types: classifiers (“if shiny then diamond”) and controllers (“if shiny then pick up”). Controllers do, however, also classify conditions before inferring actions, and therefore classification forms a central part of many AI systems. Classifiers are functions that use pattern matching to determine a closest match. They can be tuned according to examples, making them very attractive for use in AI. These examples are known as observations or patterns. In supervised learning, each pattern belongs to a certain predefined class. A class can be seen as a decision that has to be made. All the observations combined with their class labels are known as a data set. When a new observation is received, that observation is classified based on previous experience.[199]

A classifier can be trained in various ways; there are many statistical and machine learning approaches. The decision tree[200] is perhaps the most widely used machine learning algorithm. Other widely used classifiers are the neural network,[202] k-nearest neighbor algorithm,[g][204] kernel methods such as the support vector machine (SVM),[h][206] Gaussian mixture model,[207] the extremely popular naive Bayes classifier[i][209] and improved version of decision tree – decision stream.[210] Classifier performance depends greatly on the characteristics of the data to be classified, such as the dataset size, the dimensionality, and the level of noise. Model-based classifiers perform well if the assumed model is an extremely good fit for the actual data. Otherwise, if no matching model is available, and if accuracy (rather than speed or scalability) is the sole concern, conventional wisdom is that discriminative classifiers (especially SVM) tend to be more accurate than model-based classifiers such as “naive Bayes” on most practical data sets.[211]

Neural networks, or neural nets, were inspired by the architecture of neurons in the human brain. A simple “neuron” N accepts input from multiple other neurons, each of which, when activated (or “fired”), cast a weighted “vote” for or against whether neuron N should itself activate. Learning requires an algorithm to adjust these weights based on the training data; one simple algorithm (dubbed “fire together, wire together”) is to increase the weight between two connected neurons when the activation of one triggers the successful activation of another. The net forms “concepts” that are distributed among a subnetwork of shared[j] neurons that tend to fire together; a concept meaning “leg” might be coupled with a subnetwork meaning “foot” that includes the sound for “foot”. Neurons have a continuous spectrum of activation; in addition, neurons can process inputs in a nonlinear way rather than weighing straightforward votes. Modern neural nets can learn both continuous functions and, surprisingly, digital logical operations. Neural networks’ early successes included predicting the stock market and (in 1995) a mostly self-driving car.[k] In the 2010s, advances in neural networks using deep learning thrust AI into widespread public consciousness and contributed to an enormous upshift in corporate AI spending; for example, AI-related M&A in 2017 was over 25 times as large as in 2015.[214][215]

The study of non-learning artificial neural networks[202] began in the decade before the field of AI research was founded, in the work of Walter Pitts and Warren McCullouch. Frank Rosenblatt invented the perceptron, a learning network with a single layer, similar to the old concept of linear regression. Early pioneers also include Alexey Grigorevich Ivakhnenko, Teuvo Kohonen, Stephen Grossberg, Kunihiko Fukushima, Christoph von der Malsburg, David Willshaw, Shun-Ichi Amari, Bernard Widrow, John Hopfield, Eduardo R. Caianiello, and others.

The main categories of networks are acyclic or feedforward neural networks (where the signal passes in only one direction) and recurrent neural networks (which allow feedback and short-term memories of previous input events). Among the most popular feedforward networks are perceptrons, multi-layer perceptrons and radial basis networks.[216] Neural networks can be applied to the problem of intelligent control (for robotics) or learning, using such techniques as Hebbian learning (“fire together, wire together”), GMDH or competitive learning.[217]

Today, neural networks are often trained by the backpropagation algorithm, which had been around since 1970 as the reverse mode of automatic differentiation published by Seppo Linnainmaa,[218][219] and was introduced to neural networks by Paul Werbos.[220][221][222]

Hierarchical temporal memory is an approach that models some of the structural and algorithmic properties of the neocortex.[223]

In short, most neural networks use some form of gradient descent on a hand-created neural topology. However, some research groups, such as Uber, argue that simple neuroevolution to mutate new neural network topologies and weights may be competitive with sophisticated gradient descent approaches. One advantage of neuroevolution is that it may be less prone to get caught in “dead ends”.[224]

Deep learning is any artificial neural network that can learn a long chain of causal links. For example, a feedforward network with six hidden layers can learn a seven-link causal chain (six hidden layers + output layer) and has a “credit assignment path” (CAP) depth of seven. Many deep learning systems need to be able to learn chains ten or more causal links in length.[225] Deep learning has transformed many important subfields of artificial intelligence, including computer vision, speech recognition, natural language processing and others.[226][227][225]

According to one overview,[228] the expression “Deep Learning” was introduced to the Machine Learning community by Rina Dechter in 1986[229] and gained traction after Igor Aizenberg and colleagues introduced it to Artificial Neural Networks in 2000.[230] The first functional Deep Learning networks were published by Alexey Grigorevich Ivakhnenko and V. G. Lapa in 1965.[231][pageneeded] These networks are trained one layer at a time. Ivakhnenko’s 1971 paper[232] describes the learning of a deep feedforward multilayer perceptron with eight layers, already much deeper than many later networks. In 2006, a publication by Geoffrey Hinton and Ruslan Salakhutdinov introduced another way of pre-training many-layered feedforward neural networks (FNNs) one layer at a time, treating each layer in turn as an unsupervised restricted Boltzmann machine, then using supervised backpropagation for fine-tuning. Similar to shallow artificial neural networks, deep neural networks can model complex non-linear relationships. Over the last few years, advances in both machine learning algorithms and computer hardware have led to more efficient methods for training deep neural networks that contain many layers of non-linear hidden units and a very large output layer.[234]

Deep learning often uses convolutional neural networks (CNNs), whose origins can be traced back to the Neocognitron introduced by Kunihiko Fukushima in 1980.[235] In 1989, Yann LeCun and colleagues applied backpropagation to such an architecture. In the early 2000s, in an industrial application CNNs already processed an estimated 10% to 20% of all the checks written in the US.[236] Since 2011, fast implementations of CNNs on GPUs have won many visual pattern recognition competitions.[225]

CNNs with 12 convolutional layers were used in conjunction with reinforcement learning by Deepmind’s “AlphaGo Lee”, the program that beat a top Go champion in 2016.[237]

Early on, deep learning was also applied to sequence learning with recurrent neural networks (RNNs)[238] which are in theory Turing complete[239] and can run arbitrary programs to process arbitrary sequences of inputs. The depth of an RNN is unlimited and depends on the length of its input sequence; thus, an RNN is an example of deep learning.[225] RNNs can be trained by gradient descent[240][241][242] but suffer from the vanishing gradient problem.[226][243] In 1992, it was shown that unsupervised pre-training of a stack of recurrent neural networks can speed up subsequent supervised learning of deep sequential problems.[244]

Numerous researchers now use variants of a deep learning recurrent NN called the long short-term memory (LSTM) network published by Hochreiter & Schmidhuber in 1997.[245] LSTM is often trained by Connectionist Temporal Classification (CTC).[246] At Google, Microsoft and Baidu this approach has revolutionised speech recognition.[247][248][249] For example, in 2015, Google’s speech recognition experienced a dramatic performance jump of 49% through CTC-trained LSTM, which is now available through Google Voice to billions of smartphone users.[250] Google also used LSTM to improve machine translation,[251] Language Modeling[252] and Multilingual Language Processing.[253] LSTM combined with CNNs also improved automatic image captioning[254] and a plethora of other applications.

AI, like electricity or the steam engine, is a general purpose technology. There is no consensus on how to characterize which tasks AI tends to excel at.[255] While projects such as AlphaZero have succeeded in generating their own knowledge from scratch, many other machine learning projects require large training datasets.[256][257] Researcher Andrew Ng has suggested, as a “highly imperfect rule of thumb”, that “almost anything a typical human can do with less than one second of mental thought, we can probably now or in the near future automate using AI.”[258] Moravec’s paradox suggests that AI lags humans at many tasks that the human brain has specifically evolved to perform well.[121]

Games provide a well-publicized benchmark for assessing rates of progress. AlphaGo around 2016 brought the era of classical board-game benchmarks to an close. Games of imperfect knowledge provide new challenges to AI in the area of game theory.[259][260] E-sports such as StarCraft continue to provide additional public benchmarks.[261][262] There are many competitions and prizes, such as the Imagenet Challenge, to promote research in artificial intelligence. The main areas of competition include general machine intelligence, conversational behavior, data-mining, robotic cars, and robot soccer as well as conventional games.[citation needed]

The “imitation game” (an interpretation of the 1950 Turing test that assesses whether a computer can imitate a human) is nowadays considered too exploitable to be a meaningful benchmark.[263] A derivative of the Turing test is the Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA). As the name implies, this helps to determine that a user is an actual person and not a computer posing as a human. In contrast to the standard Turing test, CAPTCHA is administered by a machine and targeted to a human as opposed to being administered by a human and targeted to a machine. A computer asks a user to complete a simple test then generates a grade for that test. Computers are unable to solve the problem, so correct solutions are deemed to be the result of a person taking the test. A common type of CAPTCHA is the test that requires the typing of distorted letters, numbers or symbols that appear in an image undecipherable by a computer.

Proposed “universal intelligence” tests aim to compare how well machines, humans, and even non-human animals perform on problem sets that are generic as possible. At an extreme, the test suite can contain every possible problem, weighted by Kolmogorov complexity; unfortunately, these problem sets tend to be dominated by impoverished pattern-matching exercises where a tuned AI can easily exceed human performance levels.[265][266]

AI is relevant to any intellectual task. Modern artificial intelligence techniques are pervasive and are too numerous to list here. Frequently, when a technique reaches mainstream use, it is no longer considered artificial intelligence; this phenomenon is described as the AI effect.

High-profile examples of AI include autonomous vehicles (such as drones and self-driving cars), medical diagnosis, creating art (such as poetry), proving mathematical theorems, playing games (such as Chess or Go), search engines (such as Google search), online assistants (such as Siri), image recognition in photographs, spam filtering, prediction of judicial decisions[269] and targeting online advertisements.[270][271]

With social media sites overtaking TV as a source for news for young people and news organisations increasingly reliant on social media platforms for generating distribution,[272] major publishers now use artificial intelligence (AI) technology to post stories more effectively and generate higher volumes of traffic.[273]

Artificial intelligence is breaking into the healthcare industry by assisting doctors. According to Bloomberg Technology, Microsoft has developed AI to help doctors find the right treatments for cancer.[274] There is a great amount of research and drugs developed relating to cancer. In detail, there are more than 800 medicines and vaccines to treat cancer. This negatively affects the doctors, because there are too many options to choose from, making it more difficult to choose the right drugs for the patients. Microsoft is working on a project to develop a machine called “Hanover”. Its goal is to memorize all the papers necessary to cancer and help predict which combinations of drugs will be most effective for each patient. One project that is being worked on at the moment is fighting myeloid leukemia, a fatal cancer where the treatment has not improved in decades. Another study was reported to have found that artificial intelligence was as good as trained doctors in identifying skin cancers.[275] Another study is using artificial intelligence to try and monitor multiple high-risk patients, and this is done by asking each patient numerous questions based on data acquired from live doctor to patient interactions.[276]

According to CNN, a recent study by surgeons at the Children’s National Medical Center in Washington successfully demonstrated surgery with an autonomous robot. The team supervised the robot while it performed soft-tissue surgery, stitching together a pig’s bowel during open surgery, and doing so better than a human surgeon, the team claimed.[277] IBM has created its own artificial intelligence computer, the IBM Watson, which has beaten human intelligence (at some levels). Watson not only won at the game show Jeopardy! against former champions,[278] but was declared a hero after successfully diagnosing a woman who was suffering from leukemia.[279]

Advancements in AI have contributed to the growth of the automotive industry through the creation and evolution of self-driving vehicles. As of 2016, there are over 30 companies utilizing AI into the creation of driverless cars. A few companies involved with AI include Tesla, Google, and Apple.[280]

Many components contribute to the functioning of self-driving cars. These vehicles incorporate systems such as braking, lane changing, collision prevention, navigation and mapping. Together, these systems, as well as high performance computers, are integrated into one complex vehicle.[281]

Recent developments in autonomous automobiles have made the innovation of self-driving trucks possible, though they are still in the testing phase. The UK government has passed legislation to begin testing of self-driving truck platoons in 2018.[282] Self-driving truck platoons are a fleet of self-driving trucks following the lead of one non-self-driving truck, so the truck platoons aren’t entirely autonomous yet. Meanwhile, the Daimler, a German automobile corporation, is testing the Freightliner Inspiration which is a semi-autonomous truck that will only be used on the highway.[283]

One main factor that influences the ability for a driver-less automobile to function is mapping. In general, the vehicle would be pre-programmed with a map of the area being driven. This map would include data on the approximations of street light and curb heights in order for the vehicle to be aware of its surroundings. However, Google has been working on an algorithm with the purpose of eliminating the need for pre-programmed maps and instead, creating a device that would be able to adjust to a variety of new surroundings.[284] Some self-driving cars are not equipped with steering wheels or brake pedals, so there has also been research focused on creating an algorithm that is capable of maintaining a safe environment for the passengers in the vehicle through awareness of speed and driving conditions.[285]

Another factor that is influencing the ability for a driver-less automobile is the safety of the passenger. To make a driver-less automobile, engineers must program it to handle high risk situations. These situations could include a head on collision with pedestrians. The car’s main goal should be to make a decision that would avoid hitting the pedestrians and saving the passengers in the car. But there is a possibility the car would need to make a decision that would put someone in danger. In other words, the car would need to decide to save the pedestrians or the passengers.[286] The programing of the car in these situations is crucial to a successful driver-less automobile.

Financial institutions have long used artificial neural network systems to detect charges or claims outside of the norm, flagging these for human investigation. The use of AI in banking can be traced back to 1987 when Security Pacific National Bank in US set-up a Fraud Prevention Task force to counter the unauthorised use of debit cards. Programs like Kasisto and Moneystream are using AI in financial services.

Banks use artificial intelligence systems today to organize operations, maintain book-keeping, invest in stocks, and manage properties. AI can react to changes overnight or when business is not taking place.[287] In August 2001, robots beat humans in a simulated financial trading competition.[288] AI has also reduced fraud and financial crimes by monitoring behavioral patterns of users for any abnormal changes or anomalies.[289]

The use of AI machines in the market in applications such as online trading and decision making has changed major economic theories.[290] For example, AI based buying and selling platforms have changed the law of supply and demand in that it is now possible to easily estimate individualized demand and supply curves and thus individualized pricing. Furthermore, AI machines reduce information asymmetry in the market and thus making markets more efficient while reducing the volume of trades. Furthermore, AI in the markets limits the consequences of behavior in the markets again making markets more efficient. Other theories where AI has had impact include in rational choice, rational expectations, game theory, Lewis turning point, portfolio optimization and counterfactual thinking.

In video games, artificial intelligence is routinely used to generate dynamic purposeful behavior in non-player characters (NPCs). In addition, well-understood AI techniques are routinely used for pathfinding. Some researchers consider NPC AI in games to be a “solved problem” for most production tasks. Games with more atypical AI include the AI director of Left 4 Dead (2008) and the neuroevolutionary training of platoons in Supreme Commander 2 (2010).[291][292]

Worldwide annual military spending on robotics rose from 5.1 billion USD in 2010 to 7.5 billion USD in 2015.[293][294] Military drones capable of autonomous action are widely considered a useful asset. In 2017, Vladimir Putin stated that “Whoever becomes the leader in (artificial intelligence) will become the ruler of the world”.[295][296] Many artificial intelligence researchers seek to distance themselves from military applications of AI.[297]

For financial statements audit, AI makes continuous audit possible. AI tools could analyze many sets of different information immediately. The potential benefit would be the overall audit risk will be reduced, the level of assurance will be increased and the time duration of audit will be reduced.[298]

A report by the Guardian newspaper in the UK in 2018 found that online gambling companies were using AI to predict the behavior of customers in order to target them with personalized promotions.[299]

There are three philosophical questions related to AI:

Can a machine be intelligent? Can it “think”?

Widespread use of artificial intelligence could have unintended consequences that are dangerous or undesirable. Scientists from the Future of Life Institute, among others, described some short-term research goals to see how AI influences the economy, the laws and ethics that are involved with AI and how to minimize AI security risks. In the long-term, the scientists have proposed to continue optimizing function while minimizing possible security risks that come along with new technologies.[310]

Read more:

Artificial intelligence – Wikipedia

What is Artificial Intelligence (AI)? – Definition from …

Artificial intelligence is a branch of computer science that aims to create intelligent machines. It has become an essential part of the technology industry.

Research associated with artificial intelligence is highly technical and specialized. The core problems of artificial intelligence include programming computers for certain traits such as:

Knowledge engineering is a core part of AI research. Machines can often act and react like humans only if they have abundant information relating to the world. Artificial intelligence must have access to objects, categories, properties and relations between all of them to implement knowledge engineering. Initiating common sense, reasoning and problem-solving power in machines is a difficult and tedious task.

Machine learning is also a core part of AI. Learning without any kind of supervision requires an ability to identify patterns in streams of inputs, whereas learning with adequate supervision involves classification and numerical regressions. Classification determines the category an object belongs to and regression deals with obtaining a set of numerical input or output examples, thereby discovering functions enabling the generation of suitable outputs from respective inputs. Mathematical analysis of machine learning algorithms and their performance is a well-defined branch of theoretical computer science often referred to as computational learning theory.

Machine perception deals with the capability to use sensory inputs to deduce the different aspects of the world, while computer vision is the power to analyze visual inputs with a few sub-problems such as facial, object and gesture recognition.

Robotics is also a major field related to AI. Robots require intelligence to handle tasks such as object manipulation and navigation, along with sub-problems of localization, motion planning and mapping.

Read more:

What is Artificial Intelligence (AI)? – Definition from …