Monthly Archives: January 2022

This article is part of ourreviews of AI research papers, a series of posts that explore the latest findings in artificial intelligence.

Despite seeing tremendous advances in the recent decade, artificial intelligence is still lacking sorely in basic areas such as generalizability, adaptability, and causality. Todays AI systemsmostly centered around machine learning and deep learningare limited to narrow applications, require large amounts of training data or experience, and are very sensitive to changes in their environments.

Researchers are looking to various fields of science to find solutions to the current limits of AI systems. A new concept, proposed by researchers from various organizations and universities, draws inspiration from the two-system thinking framework proposed by Nobel laureate psychologist and economist Daniel Kahneman. Introduced in a paper published online, the technique is called SlOw and Fast AI (SOFAI). SOFAI uses meta-cognition to arbitrate between different modes of inference to improve the efficiency of AI systems in using data and compute resources.

In his acclaimed book Thinking Fast and Slow, Kahneman proposes that the human mind has two systems of decision making. System 1 is fast, implicit, intuitive, and imprecise. It controls the unconscious decisions we make, such as walking or driving in a familiar neighborhood, climbing stairs, tying our shoelaces, and other tasks we can do without conscious thinking and oftentimes in parallel. System 2, on the other hand, is the slow and meticulous type of decision-making that requires logic, rational thinking, and concentration, such as solving complex mathematical equations, playing chess, or walking on a narrow ledge.

The human brain does a great job of dividing decision-making between the two modes of thinking. For example, when youre learning a new task, such as driving, your System 2 will be more engaged. Youll need to concentrate to coordinate your different muscles, shifting gears, pressing and releasing pedals, and turning the steering wheel, while at the same time watching the street and listening to the engine. As you gradually repeat the routines, you learn to perform the tasks without concentration and your brain shifts the task to your System 1. This is why an experienced driver can control the car and do something else at the same time, such as talking to the passengers, while a novice driver must concentrate fully on doing all the tasks right.

In mentally demanding tasks, such as calculus or chess, System 2 will remain the ultimate controller. But System 1 will also shoulder some of the burden over time. For example, experienced chess players who have played thousands of games use System 1 to recognize patterns of moves of formations on the chessboard. It wont give the player a perfect solution, but it will provide intuition on where the game is headed and help save the expensive System 2 crucial time when deciding the next move.

The division of labor between System 1 and System 2 is natures solution to creating a balance between speed and accuracy, learning and execution.

As the researchers note in their paper, System 1 is able to build models of the world that, although inaccurate and imprecise, can fill knowledge gaps through causal inference, allowing us to respond reasonably well to the many stimuli of our everyday life. When the problem is too complex for System 1, System 2 kicks in and solves it with access to additional computational resources, full attention, and sophisticated logical reasoning.

Most AI systems use a single architecture to solve problems. For example, machine learning engineers will design a deep neural network to perform a single task and train it until it reaches the desired level of accuracy. Classic deep learning architectures have distinct limitations that have been amply documented in recent years. Among them is the need for large amounts of training data and computational resources. For example, a deep reinforcement learning system that mastered the videogame Dota 2 required thousands of years worth of training.

On the other hand, current AI systems are very sensitive to edge cases, situations that they havent encountered during training. For example, despite having been trained on millions of miles of simulation and real-world driving, autonomous vehicles sometimes make mistakes that most average drivers would easily avoid.

Inspired from System 1 and 2, the SOFAI architecture uses multiple problem-solvers to address some of these limitations. SOFAI is composed of a pair of System 1 (S1) and System 2 (S2) models. The System 1 solver is very fast and automatically processes any new problem or input that SOFAI faces.

SOFAI has a meta-cognitive module (MC) that decides whether the System 1 solution is accurate and reliable enough or if it needs to activate the slower and more resource-intensive System 2 solver. Like the human mind, the system also has models of itself, others, and the world. As it accumulates experience, SOFAI updates these models, which helps it improve the confidence and reliability of fast decision-making with System 1.

The MC module arbitrates between the two systems by using the information it gains from the models and the solution provided by the S1 solver. Sometimes, the S1 solution might not be too accurate, but given time constraints, it might be a better option than spending additional resources on S2. In other cases, the expected gain from activating the S2 might not justify wasting extra resources, so the MC will opt to use the S1.

According to the researchers, This architecture and flow of tasks allows for minimizing time to action when there is no need for S2 processing since S1 solvers act in constant time. It also allows the MC agent to exploit the proposed action and confidence of S1 when deciding whether to activate S2, which leads to more informed and hopefully better decisions by the MC.

While the researchers present SOFAI as a concept, they also experimented with a real implementation of the system in a grid-navigation problem. The goal of the AI system was to generate a trajectory that went from the initial state to a goal state.

The environment had a reward for achieving the goal and penalties for each move. There are additional constraints such as extra penalties for squares with black, green, and blue color codes. Basically, the AI agent must find the shortest trajectory to the goal while avoiding states that result in penalties. The researchers added some randomness to the environment to prevent it from becoming deterministic.

The SOFAI agent was composed of a simple System 1 solver that chose the move with the highest expected reward. The S1 starts with no knowledge of the environment and gradually improves as it collects experience and the SOFAI updates its model of the world (the grid environment) and itself (trajectories and moves). The System 2 component is created based on the Multi-alternative Decision Field Theory (MDFT), an inference model that can reason between different choices. MDFT can provide better results than the untrained S1 solver, but it is slower and computationally more expensive. In their experiments, the researchers tested three variations of the MDFT, each tuned for different trajectory preferences.

At every step, the SOFAIs meta-cognition unit decides whether it can trust the S1s solution or if it needs to switch to the S2 solver.

The researchers tested the different variations of SOFAI against solo S1 and S2 (MDFT) agents. Their experiments show that when used alone, the S1 system generates poor reward, trajectory length, and timing results. The S2 solver generates good trajectories and rewards but is computationally expensive and has poor timing results. In contrast, SOFAI found the right balance between reward and efficiency.

They then aggregated the results over 1,000 trajectories to see how the SOFAI model evolves its behavior and balances the use of the S1 and S2 agents. The results show that as SOFAI goes through more and more trajectories, its timing decreases, which means it becomes more compute-efficient, and its evolving behavior is very similar to how the human mind distributes cognitive labor between System 1 and System 2.

In the beginning, the SOFAI mostly uses S2 because its S1 module does not have enough experience and its decisions are not trustable. As the S2 model goes through multiple trajectories, the SOFAI updates its environment and self models, which results in better decisions by the S1. Consequently, the MC module gradually starts to shift decisions to the faster S1 module instead of relying on the compute-intensive S2. After about 450 trajectories S1 is used more often than S2. This evolving behavior allows SOFAI to be faster without degrading the quality of the trajectories it generates.

This behavior is similar to what happens in humans we first tackle a non-familiar problem with our System 2, until we have enough experience that it becomes familiar and we pass to using System 1, the researchers write.

SOFAI is one of several directions of research that have been inspired by the System 1 and 2 thinking theory. In 2019, deep learning pioneer Yoshua Bengio discussed System 2 deep learning, an area of research that aims to improve neural networks toward developing symbolic reasoning capabilities. Other related efforts are being made in developing hybrid AI systems that combine neural networks and symbolic AI.

And there are notable efforts in self-supervised learning systems that can develop behavior without the need for large amounts of data. The intersection of self-supervised learning and reinforcement learning is particularly interesting as it aims to develop memory and data-efficient AI systems that can be applied to the real world.

Though SOFAI is not the only game in town, it looks promising. The researchers plan to expand on the idea and create SOFAI systems that have multiple S1 and S2 modules and can tackle several problems with the same architecture.

Continue reading here:

An AI system that thinks fast and slow - TechTalks

Most businesses today have a great deal of data at their fingertips. They also have the tools to mine this information. But with this power comes responsibility. Before using data, technologists need to step back and evaluate the need. In todays data-driven, virtual age, it's not a question of whether you have the information, but if you should use it and how.

Artificial intelligence (AI) tools have revolutionized the processing of data, turning huge amounts of information into actionable insights. Its tempting to believe that all data is good, and that AI makes it even better. Spreadsheets, graphs, and visualizations make data real. But as any good technologist knows, the old computing sentiment, garbage in, garbage out still applies. Now more than ever, organizations need to question where the data originates and how the algorithms interpret that data. Buried in all those graphs are potential ethical risks, biases and unintended consequences.

Its easy to ask your technology partners to develop new features or capabilities, but as more and more businesses adopt machine learning (ML) operations and tools to streamline and inform processes, there is potential for bias. For instance, are the algorithms discriminating unknowingly against people of color or women? What is the source of the data? Is there permission to use the data? All these considerations need to be transparent and closely monitored.

The first step in this journey is to develop data privacy guidelines. This includes, for example, policies and procedures that address considerations such as notice and transparency that data is used for AI, policies on how information is protected and kept up to date, and how sharing data with third parties is governed. These guidelines hopefully build on an existing overarching framework of data privacy.

Beyond privacy, other relevant bodies of law may impact your development and deployment of AI. For example, in the HR space, it is critical that you refer to federal, state, and local employment and anti-discrimination laws. Likewise, in the financial sector, there are a range of applicable rules and regulations that have to be taken into account. Existing law continues to apply, just as it does outside the AI context.

Beyond existing law, with the acceleration of technology, including AI and ML, the considerations become more complex. In particular, AI and ML introduce new opportunities to discern insights from data that were previously unachievable and can do so in many ways better than humans. But AI and ML are created ultimately by humans, and without careful oversight there are risks of introducing unwanted bias and outcomes. Creating an AI and Data Ethics Board can help businesses anticipate issues in these new technologies.

Begin by establishing guiding principles to govern the use of AI, ML and automation specifically in your company. The goal is to ensure that your models are relevant and functional, and do not drift from their intended goal unknowingly or inappropriately. Consider these five guidelines:

1. Accountability and transparency. Conduct audit and risk assessments to test your models, and actively monitor and improve your models and systems to ensure that changes in the underlying data or model conditions do not inappropriately affect the desired results.

2. Privacy by design. Ensure that your enterprise-wide approach incorporates privacy and data security into ML and associated data processing systems. For example, do your ML models seek to minimize access to identifiable information to ensure that you are using only the personal data you need to generate insights? Are you providing individuals with a reasonable opportunity to examine their own personal data and to update it if its inaccurate?

3. Clarity. Design AI solutions that are explainable and direct. Are your ML data discovery and data usage models designed with understanding as a key attribute, measured against an expressed desired outcome?

4. Data governance. Understanding how you use data and the sources from which you obtain it should be key to your AI and ML principles. Maintain processes and systems to track and manage data usage and retention. If you use external information in your models, such as government reports or industry terminologies, understand the processes and impact of that information in your models.

5. Ethical and practical use of data. Establish governance to provide guidance and oversight on the development of products, systems and applications that involve AI and data.

Principles like these can both guide discussion about these issues and help to create policies and procedures about how data is handled in your business. More broadly, they will set the tone for the entire organization.

Guidelines are great -- but they need to be enforced to be effective. An AI and data ethics board is one way to ensure these principles are woven into product development and uses of internal data. But how can companies go about doing this?

Begin by bringing together an interdisciplinary team. Consider including both internal and external experts such as IT, product development, legal and compliance, privacy, security, audit, diversity and inclusion, industry analysts, external legal and/or an expert in consumer affairs, for instance. The more diverse and knowledgeable the team, the more effective your discussions can be around potential implications and the viability of different use cases.

Next, spend time discussing the larger issues. Its important here to step away from process for a minute and immerse yourselves in live, productive discussion. What are your organizations core values? How should they inform your policies around development and deployment of AI and ML? All this discussion sets the foundation for the procedures and processes you outline.

Setting a regular meeting cadence to review projects can be helpful as well. Again, the bigger issues should drive the discussion. For instance, most product developers will present the technical aspects -- such as how the data is protected or encrypted. The boards role should aim to analyze the project on a more fundamental level. Some questions to consider for guiding discussion could be:

Because AI and ethics has become an increasingly important issue, there are many resources to help your organization navigate these waters. Reach out to your vendors, consulting firms or trade groups and consortiums, like the Enterprise Data Management (EDM) Council. Implement the pieces that are appropriate for your business but remember that tools, checklists, processes, and procedures should not replace the value of the discussion.

The ultimate goal is to make these considerations a part of the company culture so that every employee that touches a project, works with a vendor or consults with a client, keeps data privacy front of mind.

Read more:

Why You Need an AI & Ethics Board - InformationWeek

AUSTIN, Texas, Jan. 24, 2022 /PRNewswire/ -- Silicon Labs, a leader in secure, intelligent wireless technology for a more connected world, today announced the BG24 and MG24 families of 2.4 GHz wireless SoCs for Bluetooth and Multiple-protocol operations, respectively, and a new software toolkit. This new co-optimized hardware and software platform will help bring AI/ML applications and wireless high performance to battery-powered edge devices. Matter-ready, the ultra-low-power BG24 and MG24 families support multiple wireless protocols and incorporate PSA Level 3 Secure Vaultprotection, ideal for diverse smart home, medical and industrial applications. The SoC and software solution for the Internet of Things (IoT) announced today includes:

"The BG24 and MG24 wireless SoCs represent an awesome combination of industry capabilities including broad wireless multiprotocol support, battery life, machine learning, and security for IoT Edge applications," said Matt Johnson, CEO of Silicon Labs.

First Integrated AI/ML Acceleration Improves Performance and Energy Efficiency

IoT product designers see the tremendous potential of AI and machine learning to bring even greater intelligence to edge applications like home security systems, wearable medical monitors, sensors monitoring commercial facilities and industrial equipment, and more. But today, those considering deploying AI or machine learning at the edge are faced with steep penalties in performance and energy use that may outweigh the benefits.

The BG24 and MG24 alleviate those penalties as the first ultra-low powered devices with dedicated AI/ML accelerators built-in. This specialized hardware is designed to handle complex calculations quickly and efficiently, with internal testing showing up to a 4x improvement in performance along with up to a 6x improvement in energy efficiency. Because the ML calculations are happening on the local device rather than in the cloud, network latency is eliminated for faster decision-making and actions.

The BG24 and MG24 families also have the largest Flash and random access memory (RAM) capacities in the Silicon Labs portfolio. This means that the device can evolve for multi-protocol support, Matter, and trained ML algorithms for large datasets. PSA Level3-Certified Secure VaultTM,the highest level of security certification for IoT devices, provides the security needed in products like door locks, medical equipment, and other sensitive deployments where hardening the device from external threats is paramount.

To learn more about the capabilities of the BG24 and MG24 SoCs and view a demo on how to get started, register for the instructional Tech Talk "Unboxing the new BG24 and MG24 SoCs" here: https://www.silabs.com/tech-talks.

AI/ML Software and Matter-Support Help Designers Create for New Innovative Applications

In addition to natively supporting TensorFlow, Silicon Labs has partnered with some of the leading AI and ML tools providers, like SensiML and Edge Impulse, to ensure that developers have an end-to-end toolchain that simplifies the development of machine learning models optimized for embedded deployments of wireless applications. Using this new AI/ML toolchain with Silicon Labs's Simplicity Studio and the BG24 and MG24 families of SoCs, developers can create applications that draw information from various connected devices, all communicating with each other using Matter to then make intelligent machine learning-driven decisions.

For example, in a commercial office building, many lights are controlled by motion detectors that monitor occupancy to determine if the lights should be on or off. However, when typing at a desk with motion limited to hands and fingers, workers may be left in the dark when motion sensors alone cannot recognize their presence. By connecting audio sensors with motion detectors through the Matter application layer, the additional audio data, such as the sound of typing, can be run through a machine-learning algorithm to allow the lighting system to make a more informed decision about whether the lights should be on or off.

ML computing at the edge enables other intelligent industrial and home applications, including sensor-data processing for anomaly detection, predictive maintenance, audio pattern recognition for improved glass-break detection, simple-command word recognition, and vision use cases like presence detection or people counting with low-resolution cameras.

Alpha Program Highlights Variety of Deployment Options

More than 40 companies representing various industries and applications have already begun developing and testing this new platform solution in a closed Alpha program. These companies have been drawn to the BG24 and MG24 platforms by their ultra-low power, advanced features, including AI/ML capabilities and support for Matter. Global retailers are looking to improve the in-store shopping experience with more accurate asset tracking, real-time price updating, and other uses. Participants from the commercial building management sector are exploring how to make their building systems, including lighting and HVAC, more intelligent to lower owners' costs and reduce their environmental footprint. Finally, consumer and smart home solution providers are working to make it easier to connect various devices and expand the way they interact to bring innovative new features and services to consumers.

Silicon Labs' Most Capable Family of SoCs

The single-die BG24 and MG24 SoCs combine a 78 MHz ARM Cortex-M33 processor, high-performance 2.4 GHz radio, industry-leading 20-bit ADC, an optimized combination of Flash (up to 1536 kB) and RAM (up to 256 kB), and an AI/ML hardware accelerator for processing machine learning algorithms while offloading the ARM Cortex-M33, so applications have more cycles to do other work.Supporting a broad range of 2.4 GHz wireless IoT protocols, these SoCs incorporate the highest security with the best RF performance/energy-efficiency ratio in the market.

Availability

EFR32BG24 and EFR32MG24 SoCs in 5 mm x 5 mm QFN40 and 6 mm x 6 mm QFN48 packages are shipping today to Alpha customers and will be available for mass deployment in April 2022. Multiple evaluation boards are available to designers developing applications.Modules based on the BG24 and MG24 SoCs will be available in the second half of 2022.

To learn more about the new BG24 family, go to: http://silabs.com/bg24.

To learn more about the new MG24 family, go to: http://silabs.com/mg24.

To learn more about how Silicon Labs supports AI and ML, go to: http://silabs.com/ai-ml.

About Silicon Labs

Silicon Labs (NASDAQ: SLAB) is a leader in secure, intelligent wireless technology for a more connected world. Our integrated hardware and software platform, intuitive development tools, unmatched ecosystem, and robust support make us an ideal long-term partner in building advanced industrial, commercial, home, and life applications. We make it easy for developers to solve complex wireless challenges throughout the product lifecycle and get to market quickly with innovative solutions that transform industries, grow economies, and improve lives.Silabs.com

View original content to download multimedia:https://www.prnewswire.com/news-releases/silicon-labs-brings-ai-and-machine-learning-to-the-edge-with-matter-ready-platform-301466032.html

SOURCE Silicon Labs

See original here:

Silicon Labs Brings AI and Machine Learning to the Edge with Matter-Ready Platform - inForney.com

There is great promise and potential in artificial intelligence (AI), but if such technologies are built and trained by humans, are they capable of bias?

Absolutely, says William Wang, the Duncan and Suzanne Mellichamp Chair in Artificial Intelligence and Designs at UC Santa Barbara, who will give the virtual talk What is Responsible AI, at 4 p.m. Tuesday, Jan. 25, as part of the UCSB Librarys Pacific Views speaker series (register here).

The key challenge for building AI and machine learning systems is that when such asystem is trained on datasets with limited samples from history, they may gain knowledge from the protected variables (e.g., gender, race, income, etc.), and they are prone to produce biased outputs, said Wang, also director of UC Santa Barbaras Center for Responsible Machine Learning.

Sometimes these biases could lead to the rich getting richer phenomenon after the AI systems are deployed, he added.Thats why in addition to accuracy, it is important to conduct research in fair and responsible AI systems, including the definition of fairness, measurement, detection and mitigation of biases in AI systems.

Wangs examination of the topic serves as the kickoff event for UCSB Reads 2022, the campus and community-wide reading program run by UCSB Library. Their new season is centered on Ted Chiangs Exhalation, a short story collection thataddressesessential questions about human and computer interaction, including the use of artificial intelligence.

Copies of Exhalation will be distributed free to students (while supplies last) Tuesday, Feb. 1 outside the Librarys WestPaseo entrance. Additional events announced so far include on-air readings from the book on KCSB, a faculty book discussion moderated by physicist and professor David Weld and a sci-fi writing workshop. It all culminates May 10 with a free lecture by Ted Chiang in Campbell Hall.

First though: William Wang, an associate professor of computer science and co-director of the Natural Language Processing Group.

In this talk, my hope is to summarize the key advances of artificialintelligence technologies in the last decade, and share how AI can bring us an exciting future, he noted. I will also describe the key challenges of AI: how we should consider the research and development of responsible AI systems,which not only optimize their accuracy performance,but also provide a human-centric view to consider fairness, bias, transparency and energy efficiency of AI systems.

How do we build AI models that are transparent? How do we write AI system descriptions that meet disclosive transparency guidelines?How do we consider energy efficiency when building AI models? he asked. The future of AI is bright, but all of these are key aspects of responsible AI that we need to address.

See original here:

Getting a Read on Responsible AI | The UCSB Current - The UCSB Current

In the last decade, almost every argument putting forward the idea of artificial intelligence (AI) taking up white-collar jobs from humans was downplayed by critics with a rather simplistic counter machines lack the creativity and intuition of a human worker.

For instance, Rytr, a popular AI copywriting app with 6,00,000 users, is helping businesses with text that is indistinguishable from human writing. Unlike content writers, this app doesnt tire, doesnt ask for payment and can generate an unlimited amount of content.

Similarly, AI-powered customer service is slowly taking over the market. In the United States, an estimated 85 percent of customer interaction is taking place with AI-backed tools. Given the brisk pace at which such tools are gaining popularity, there would be huge implications for nearly 3 million customer service representatives employed in the US.

The landscape has clearly changed. The businesses which dont use AI tools are at a disadvantage. Their input cost is higher and their product (or services) is probably inferior to their counterparts who have deployed modern technologies.

While AI-backed technologies are a boon for nations like Japan where the workforce is ageing, they can lead to a massive disruption in countries like India, where millions of people enter the job market every year.

Nonetheless, AI eating up jobs is surely not new but a reality that we all have to come to terms with. Of course, there is this argument of new job avenues opening up with the rise of AI but the question is are we prepared?

Read Also |

(Edited by : Thomas Abraham)

Read the original:

How is AI slowly eating up white-collar jobs? - CNBCTV18