Daily Archives: August 9, 2017

A musician improvises alongside A.I. Duet, software developed in part by Googles Magenta

google

By Matthew HutsonAug. 8, 2017 , 3:40 PM

Can computers be creative? Thats a question bordering on the philosophical, but artificial intelligence (AI) can certainly make music and artwork that people find pleasing. Last year, Google launched Magenta, a research project aimed at pushing the limits of what AI can do in the arts. Science spoke with Douglas Eck, the teams lead in San Francisco, California, about the past, present, and future of creative AI. This interview has been edited for brevity and clarity.

Q: How does Magenta compose music?

A: Learning is the key. Were not spending any effort on classical AI approaches, which build intelligence using rules. Weve tried lots of different machine-learning techniques, including recurrent neural networks, convolutional neural networks, variational methods, adversarial training methods, and reinforcement learning. Explaining all of those buzzwords is too much for a short answer. What I can say is that theyre all different techniques for learning by example to generate something new.

Q: What examples does Magenta learn from?

A:We trained theNSynthalgorithm, which uses neural networks to synthesize new sounds, on notes generated by different instruments. TheSketchRNNalgorithm was trained onmillions of drawingsfrom ourQuick, Draw!game. Our most recent music algorithm,Performance RNNwas trained on classical piano performances captured on a modern player piano [listen below]. I'd like musicians to be able to easily train models on their own musical creations, then have fun with the resulting music, further improving it.

Q: How has computer composition changed over the years?

A:Currently the focus is on algorithms which learn by example, i.e., machine learning, instead of using hard-coded rules. I also think theres been increased focus on using computers as assistants for human creativity rather than as a replacement technology, such as our work and Sonys Daddys Car [a computer-composed song inspired by The Beatles and fleshed out by a human producer].

Q: Do the results of computer-generated music ever surprise you?

A:Yeah. All the time. I was really surprised at how expressive the short compositions were from Ian Simon and Sageev Oores recent Performance RNN algorithm. Because they trained on real performances captured in MIDI on Disklavier pianos, their model was able to generate sequences with realistic timing and dynamics.

Q: What else is Magenta doing?

A:We did a summer internship around joke telling, but we didnt generate any funny jokes. Were also working on image generation and drawing generation [seeexample below]. In the future, Id like to look more at areas related to design. Can we provide tools for architects or web page creators?

Magenta software can learn artistic styles from human paintings and apply them to new images.

Fred Bertsch

Q: How do you respond to art that you know comes from a computer?

A: When I was on the computer science faculty at University of Montreal [in Canada], I heard some computer music by a music faculty member, Jean Pich. Hed written a program that could generate music somewhat like that of the jazz pianist Keith Jarrett. It wasnt nearly as engaging as the real Keith Jarrett! But I still really enjoyed it, because programming the algorithm is itself a creative act. I think knowing Jean and attributing this cool program to him made me much more responsive than I would have been otherwise.

Q: If abilities once thought to be uniquely human can be aped by an algorithm, should we think differently about them?

A: I think differently about chess now that machines can play it well. But I dont see that chess-playing computers have devalued the game. People still love to play! And computers have become great tools for learning chess.Furthermore, I think its interesting to compare and contrast how chess masters approach the game versus how computers solve the problemvisualization and experience versus brute-force search, for example.

Q: How might people and machines collaborate to be more creative?

A: I think its an iterative process. Every new technology that made a difference in art took some time to figure out. I love to think of Magenta like an electric guitar. Rickenbacker and Gibson electrified guitars with the purpose of being loud enough to compete with other instruments onstage.Jimi Hendrix and Joni Mitchell and Marc Ribot and St. Vincent and a thousand other guitarists who pushed the envelope on how this instrument can be played were all using the instrument the wrong way, some saidretuning, distorting, bending strings, playing upside-down, using effects pedals, etc. No matter how fast machine learning advances in terms of generative models, artists will work faster to push the boundaries of whats possible there, too.

Excerpt from:

How Google is making music with artificial intelligence - Science Magazine

Growing up, Kightlinger and Gray LLP attorney Adam Ira can recall members of his family, many of whom were factory workers, expressing concerning about the prospect of automated machines taking their jobs. Now, Ira said similar concerns are creeping into his work as a lawyer, as the rise of artificial intelligence in the practice of law has begun automating legal tasks previously performed by humans.

As the number of available AI products grows, attorneys have begun to gravitate toward tools that enable them to do their work quickly and more efficiently. Artificial intelligence can come in multiple forms, legal tech experts say, from simple document automation to more complex intelligence using algorithms to predict legal outcomes.

In recent months, several new AI products have been introduced with the promise of automating the mundane tasks of being a lawyer, leaving attorneys with more time to focus on the complex legal questions raised by their clients.

For example, Seattle-based TurboPatent Corp. launched an AI tool in mid-July known as RoboReview. Through RoboReview, patent attorneys can upload a patent application into the AI software, which then scans the document and assesses it for similarities to previous patent applications and uses the level of similarity to predict patent eligibility. RoboReview can also make other predictions about the patent process, such as how long the process might take or what actions the U.S. Patent and Trademark Office may take with the application, said Dave Billmaier, TurboPatent vice president of product marketing.

Shortly after RoboReview went public, Amy Wan, attorney and founder and CEO of Bootstrap Legal, introduced an AI product that automates the process of drafting legal paperwork for people trying to raise capital for a real estate project of $2 million or less. As a former real estate securities attorney, Wan said she witnessed firsthand how inefficient the process of drafting such documents could be, especially considering that much of the work involved routine tasks such as copying and pasting from previous documents.

With Wans AI product, users answer questions about their real estate project, and the software uses those answers to develop the necessary legal documents, which are returned to the user within 48 hours. Such technology expedites drafting the documents a process she said could otherwise take 20 to 25 hours to complete while also cutting the costs associated with raising real estate capital. Wan said her company and AI product are based on the principle that cost considerations should not prevent people from accessing legal services.

Saving time and cutting costs are AI advantages that serve as the key selling points for legal tech developers, as clients have come to expect their attorneys to use modern technology to perform efficient work at the lowest possible cost, said Jason Houdek, a patent and intellectual property attorney with Taft Stettinius & Hollister LLP. Though RoboReview is new, Houdek said he has been using similar AI tools to determine patent eligibility, ensure application quality and predict patent examiner behavior for several years.

Similarly, Haley Altman, CEO of Indianapolis-based Doxly Inc., said most legal tech entrepreneurs like her are trying to develop AI tools that take large sets of data or documents and extrapolate the relevant information lawyers are looking for, thus reducing the amount of time they spend combing through documents. The Doxly software, which is designed to automate the legal transaction process, uses AI to mimic a transactional attorneys natural workflow, making the software feel natural, she said.

Job security?

Despite these benefits, some attorneys are concerned that continued use of AI in the practice of law could put them out of a job. Further, Jim Billmaier, TurboPatent CEO, said the old guard of attorneys, those who have been in practice for many years, can be inclined to resist artificial intelligence tools because they go against traditional practice.

There may be some legitimacy to those concerns, the attorneys and legal tech experts said. For example, attorneys at large firms that still employ the typical billable-hour model could see a drop in their hours as a result of AI products, said Dan Carroll with vrsus LLC, a rebranded version of legal tech company CasePacer. The vrsus technology utilizes AI to enable attorneys at plaintiffs firms to reach outcomes for their clients as quickly as possible, rather than focusing on how many hours they are able to bill, Carroll said.

Similarly, certain practice areas that are more transactional in nature, such as bankruptcy or tax law, might be more susceptible to automation, Ira said.

But such automation is now inevitable, as further AI development is a matter of when, not if, Houdek said. Jim Billmaier agreed and noted that attorneys who are resistant to AI advancements will find themselves underperforming if they choose not to take advantage of tools that increase efficiency.

While technological advancements might be inevitable, they do not have to be uncontrollable, said Ira Smith, vrsus chief strategy officer. Few attorneys fully understand the nuances of what makes AI work, Smith said, yet few tech developers, such as IBM, understand the nuances of practicing law.

As a result, attorneys and legal tech companies should focus less on how new artificial intelligence products might change their work and instead try to mold whatever AI tools are currently on the market to improve the product of their work, Smith said. He encouraged attorneys to be product agnostic and focus less on the technological platform and more on technologys possible benefits.

Why would it matter whether (IBMs) Watson is utilizing my data as long as I can take that and serve it back to my clients? Smith said.

Human advantage

Even as legal tech and other companies offer new and ever more advanced AI products, attorneys said the human mind will always be needed in the practice of law.

For example, even if a computer becomes intelligent enough to draw up contracts on its own, lawyers will still need to review and finalize them, Altman said. Ira agreed and noted that use of AI can create ethical issues, as attorneys must ensure the automated documents they produce reflect accurate and competent work.

Further, the power of persuasion is a trait that is uniquely human, and one that is critical to the practice of law, Ira said. Though an intelligent computer might able to cobble together a legal argument one day an advancement he thinks is still at least 10 to 15 years off it could never speak to a judge or a jury in a manner meant to persuade and effectively advocate on behalf of a client, Ira said.

Similarly, judges will always be needed to use their minds and legal training to decide the outcome of cases, Houdek said, and human juries will always be needed to decide cases.

Though some human jobs or billable hours might decrease as a result of advancements in artificial intelligence, the legal tech experts said AI is more of a benefit than a threat because it allows legal professionals to use their minds and training for the creative work that comes with being an attorney.

AI technology isnt taking their jobs, Altman said. The whole point of it is to enable them to do the work that they really want to be focusing on.

Read more:

Artificial intelligence is inevitable. Will you embrace or resist it in your practice? - Indiana Lawyer

South Korean scientists have been able to train artificial intelligence to detect anthrax at fast speeds, potentially dealing ablow to bioterrorism.

Hidden in letters, the biological agent killed five Americans and infected 17 morein the yearfollowing the 9/11 attacks, and the threat of a biological attackremains a top concern of Western security services as radicals such as the Islamic State militant group (ISIS) seek new ways to attack the West.

Researchers from the Korea Advanced Institute of Science and Technology have now created an algorithm that is able to study bacterial spores and quickly identify the biological agent, according to a paper published last week for the Science Advances journal.

Tech & Science Emails and Alerts - Get the best of Newsweek Tech & Science delivered to your inbox

The new training of AI to identify the bacteria usingmicroscopic images could decrease the time it takes to detect anthrax drastically, to mere seconds from a day. It is also accurate 95 percent of the time.

Anthrax contaminates the body when spores enter it, mostly through inhalation, multiplying and spreading an illness that could be fatal. Skin infections of anthrax are less deadly.

Spores from the Sterne strain of anthrax bacteria (Bacillus anthracis) are pictured in this handout scanning electron micrograph obtained by Reuters May 28, 2015. Reuters/Center for Disease Control/Handout

This study showed that holographic imaging and deep learning can identify anthrax in a few seconds,YongKeun Paul Park, associate professor of physics at the Korea Advanced Institute of Science and Technology, told the IEEE Spectrum blog.

Conventional approaches such as bacterial culture or gene sequencing would take several hours to a day, he added.

Park is working with the South Korean agency responsible for developing the country's defense capabilities amid fears that North Korea may plan a biological attack against its archenemy across their shared border.

North Korea's regime is no stranger to chemical agents. South Korea has accused operatives linked toPyongyang of responsibility for the assassination of North Korean leader Kim JongUn's half brother, Kim Jong Nam, using a VX agent at Malaysia's Kuala Lumpur International Airport in February.

Contamination by anthrax hasa death rate of 80 percent, so detection of the bacteria is crucial.

Spreading anthrax far and wide in an attack would mean that thousands would die if contaminated. So Western security services fear that hostile parties, such as ISIS sympathizers or regimes such as North Korea, will make attempts to develop a capability to cause a mass-casualty attack.

The researchers say the AI innovation could bring advances elsewhere, too, including the potential to detect other bacterias, such as those that cause food poisoning and kill more than a quarter of a million people every year.

See the original post:

AI Vs. Bioterrorism: Artificial Intelligence Trained to Detect Anthrax by Scientists - Newsweek

I've been following cybersecurity startups and hackers for years, and I suddenly discovered how hackers are always ahead of the rest of us they have a better business model funding them in their proof of concept (POC) stage of development.

To even begin protecting ourselves from their well-funded advances and attacks, cyberdefense and artificial intelligence (AI)technologies must be funded at the same level in the POC stage.

Today, however, traditional investors not only want your technology running, they also need assurances that you already have a revenue stream which stifles potential new technology discovery at the POC level. And in some industries, this is dangerous.

Consider the fast-paced world of cybersecurity, in which companies are offered traditional funding avenues as they promote their product's tech capabilities so people will invest. This promotion and disclosure of their technology, however, gives hackers a road map to the new cyberdefense technologies and a window of time to gain knowledge on how to exploit them.

This same road map exists for technologies covered in detail when standard groups, universities, governments and private labs publish white papers documents that essentially assist hackers by giving them advanced notice of cyberdefense techniques.

In addition to this, some hackers receive immediate funding through nation states that are coordinating cyberwarfare like the traditional military and others are involved in organized secret groups that fund the use of ransomware and DDoS attacks. These hackers get immediate funding and then throw their technology on the Internet for POC discovery.

One project that strongly makes a case for rapidly funding cyberdefense technologies in an effort to keep up with hackers is the $5.7 billion U.S. Department of Homeland Security's (DHS) EINSTEIN cyberdefense system, which was deemed obsolete upon its deployment for failing to detect 94 percent of security vulnerabilities. As this situation illustrates, the traditional methods of funding cyberdefense taking years of bureaucratic analysis and vendor contracts does not work in the fast technology discovery world of cyberdefense. After the EINSTEIN project failure, DHS decided to conduct an assessment it's currently working to understand if it's making the right investments in dealing with the ever-changing cyberenvironment.

But it also has other roadblocks, as even large technology companies and contractors with which DHS does business have their own bureaucracies and investments that ultimately deter the department from getting the best in cyberdefense technologies. And once universities, standards groups, regulation and funding approvals are added to these processes, you're pretty much assured to be headed for another disaster.

But DHS doesnt need to develop these technologies itself. The department needs to support public- and private-sector POCs to rapidly mature and deploy new cyberdefense technologies. This suggestion is supported by what other countries are successfully doing including our adversaries.

The same two things that have motivated mankind all through history immediate power and money are now motivating hackers, and cyberdefense technologies are taking years to be deployed. So I'll say it again: The motivational and funding model of cyberdefense technologies must change. The key to successful cyberdefense technology development is making it as aggressive as the hackers that attack it. And this needs to be done at the conceptual POC level.

The concern in cyberdefense (and really all AI) is the race to the quantum computer.

Quantum computer technologies cant be hacked, and in theory, its processing power can break all encryption. The computational physics behind the quantum also offer remarkable capabilities that will drastically change all current AI and cyberdefense technologies. This is a winner-takes-all technology that offers capability with absolute security capabilities capabilities that we can now only imagine.

The most recent funding source for hackers is Bitcoin, which uses the decentralized and secure blockchain technology. It has even been used to support POC funding in what is called an Initial Coin Offering (ICO), the intent of which is to crowdfund early startup companies at the development or POC level by bypassing traditional and lengthy funding avenues. Because this type of startup seed offering has been clouded with scams, it is now in regulatory limbo.

Some states have passed laws that make it difficult to legally present and offer an ICO. While the U.S. seems to be pushing ICOregulation, other countries are still deciding what to do. But like ICOs or not, they offer first-time startups an avenue of fast-track funding at the concept level where engineers and scientists can jump on newer technologies by focusing seed money on testing their concepts. Bogging ICOs down with regulatory laws will both slow down legitimate POC innovation in the U.S. and give other countries a competitive edge.

Another barrier to cyberdefense POC funding is the size and technological control of a handful of tech companies. Google, Facebook, Amazon, Microsoft and Apple have become enormous concentrations of wealth and data, drawing the attention of economists and academics who warn they're growing too powerful. Now as big as major American cities, these companies are mega centers of both money and technology. They are so large and control so much of the market that many are beginning to view them as in violation of the Sherman Antitrust Act. So how can small startups compete with these tech giants and potentially fund POCs in areas such as cyberdefense and AI? By aligning with giant companies in industries that have the most need for cyberdefense and AI technologies: critical infrastructure.

The industries that are most vulnerable and could cause the most devastation if hacked are those involved in critical infrastructure. These large industries have the resources to fund cyberdefense technologies at the concept level and they would obtain superior cyberdefense technologies in doing so.

Cyberattacks to critical infrastructure could devastate entire country economies and must be protected by the most advanced cyberdefense. Quantum computing and artificial intelligence will initiate game-changing technology in both cyberdefense and the new intellectual property deriving from quantum sciences. Entering these new technologies at the POC level is like being a Microsoft or Google years ago. Funding the development of these new technologies in cyberdefense and AI are needed soon but what about today?

Future quantum computer capabilities will also demand immediate short-term fixes in current cyberdefense and AI. New quantum-ready compressed encryption and cyberdefense deep learning AI must be funded and tested now at the concept level. The power grid, oil and gas, and even existing telecoms are perfect targets for this funding and development. Investing today would offer current cyberdefense and business intelligence protection while creating new profit centers in the licensing and sale of these leading-edge technologies. This is true for many other industries, all differing in their approach and requiring specialized cyberdefense capabilities and new intelligence gathering that will shape their future.

So we must find creative ways of rapidly funding cyberdefense technologies at the conceptual level. If this is what hackers do and it's why they're always one step ahead, shouldn't we work to surpass them?

Link:

The Race to Cyberdefense, Artificial Intelligence and the Quantum Computer - Government Technology

Even as artificial intelligence and machine learning continue to break new ground, there is enough evidence to indicate how easy it is for bias to creep into even the most advanced algorithms. Photo: iStockphoto

Bengaluru: Last year, for the first time ever, an international beauty contest was judged by machines. Thousands of people from across the world submitted their photos to Beauty.AI, hoping that their faces would be selected by an advanced algorithm free of human biases, in the process accurately defining what constitutes human beauty.

In preparation, the algorithm had studied hundreds of images of past beauty contests, training itself to recognize human beauty based on the winners. But what was supposed to be a breakthrough moment that would showcase the potential of modern self-learning, artificially intelligent algorithms rapidly turned into an embarrassment for the creators of Beauty.AI, as the algorithm picked the winners solely on the basis of skin colour.

The algorithm made a fairly non-trivial correlation between skin colour and beauty. A classic example of bias creeping into an algorithm, says Nisheeth K. Vishnoi, an associate professor at the School of Computer and Communication Sciences at Switzerland-based cole Polytechnique Fdrale de Lausanne (EPFL). He specializes in issues related to algorithmic bias.

A widely cited piece titled Machine bias from US-based investigative journalism organization ProPublica in 2016 highlighted another disturbing case.

It cited an incident involving a black teenager named Brisha Borden who was arrested for riding an unlocked bicycle she found on the road. The police estimated the value of the item was about $80.

In a separate incident, a 41-year-old Caucasian man named Vernon Prater was arrested for shoplifting goods worth roughly the same amount. Unlike Borden, Prater had a prior criminal record and had already served prison time.

Yet, when Borden and Prater were brought for sentencing, a self-learning program determined Borden was more likely to commit future crimes than Praterexhibiting the sort of racial bias computers were not supposed to have. Two years later, it was proved wrong when Prater was charged with another crime, while Bordens record remained clean.

And who can forget Tay, the infamous racist chatbot that Microsoft Corp. developed last year?

Even as artificial intelligence and machine learning continue to break new ground, there is enough evidence to indicate how easy it is for bias to creep into even the most advanced algorithms. Given the extent to which these algorithms are capable of building deeply personal profiles about us from relatively trivial information, the impact that this can have on personal privacy is significant.

This issue caught the attention of the US government, which in October 2016 published a comprehensive report titled Preparing for the future of artificial intelligence, turning the spotlight on the issue of algorithmic bias. It raised concerns about how machine-learning algorithms can discriminate against people or sets of people based on the personal profiles they develop of all of us.

If a machine learning model is used to screen job applicants, and if the data used to train the model reflects past decisions that are biased, the result could be to perpetuate past bias. For example, looking for candidates who resemble past hires may bias a system toward hiring more people like those already on a team, rather than considering the best candidates across the full diversity of potential applicants, the report says.

The difficulty of understanding machine learning results is at odds with the common misconception that complex algorithms always do what their designers choose to have them do, and therefore that bias will creep into an algorithm if and only if its developers themselves suffer from conscious or unconscious bias. It is certainly true that a technology developer who wants to produce a biased algorithm can do so, and that unconscious bias may cause practitioners to apply insufficient effort to preventing bias, it says.

Over the years, social media platforms have been using similar self-learning algorithms to personalize their services, offering content better suited to the preferences of their usersbased solely on their past behaviour on the site in terms of what they liked or the links they clicked on.

What you are seeing on platforms such as Google or Facebook is extreme personalizationwhich is basically when the algorithm realizes that you prefer one option over another. Maybe you have a slight bias towards (US President Donald) Trump versus Hillary (Clinton) or (Prime Minister Narendra) Modi versus other opponentsthats when you get to see more and more articles which are confirming your bias. The trouble is that as you see more and more such articles, it actually influences your views, says EPFLs Vishnoi.

The opinions of human beings are malleable. The US election is a great example of how algorithmic bots were used to influence some of these very important historical events of mankind, he adds, referring to the impact of fake news on recent global events.

Experts, however, believe that these algorithms are rarely the product of malice. Its just a product of careless algorithm design, says Elisa Celis, a senior researcher along with Vishnoi at EPFL.

How does one detect bias in an algorithm? It bears mentioning that machine learning-algorithms and neural networks are designed to function without human involvement. Even the most skilled data scientist has no way to predict how his algorithms will process the data provided to them, said Mint columnist and lawyer Rahul Matthan in a recent research paper on the issue of data privacy published by the Takshashila Institute, titled Beyond consent: A new paradigm for data protection.

One solution is black-box testing, which determines whether an algorithm is working as effectively as it should without peering into its internal structure. In a black-box audit, the actual algorithms of the data controllers are not reviewed. Instead, the audit compares the input algorithm to the resulting output to verify that the algorithm is in fact performing in a privacy-preserving manner. This mechanism is designed to strike a balance between the auditability of the algorithm on the one hand and the need to preserve proprietary advantage of the data controller on the other. Data controllers should be mandated to make themselves and their algorithms accessible for a black box audit, says Matthan, who is also a fellow with Takshashilas technology and policy research programme.

He suggests the creation of a class of technically skilled personnel or learned intermediaries whose sole job will be to protect data rights. Learned intermediaries will be technical personnel trained to evaluate the output of machine-learning algorithms and detect bias on the margins and legitimate auditors who must conduct periodic reviews of the data algorithms with the objective of making them stronger and more privacy protective. They should be capable of indicating appropriate remedial measures if they detect bias in an algorithm. For instance, a learned intermediary can introduce an appropriate amount of noise into the processing so that any bias caused over time due to a set pattern is fuzzed out, Matthan explains.

That said there still remain significant challenges in removing the bias once discovered.

If you are talking about removing biases from algorithms and developing appropriate solutions, this is an area that is still largely in the hands of academiaand removed from the broader industry. It will take time for the industry to adopt these solutions on a larger scale, says Animesh Mukherjee, an associate professor at the Indian Institute of Technology, Kharagpur, who specializes in areas such as natural language processing and complex algorithms.

This is the first in a four-part series. The next part will focus on consent as the basis of privacy protection.

A nine-judge Constitution bench of the Supreme Court is currently deliberating whether or not Indian citizens have the right to privacy. At the same time, the government has appointed a committee under the chairmanship of retired Supreme Court judge B.N. Srikrishna to formulate a data protection law for the country. Against this backdrop, a new discussion paper from the Takshashila Institute has proposed a model of privacy particularly suited for a data-intense world. Over the course of this week we will take a deeper look at that model and why we need a new paradigm for privacy. In that context, we examine the increasing reliance on software to make decisions for us, assuming that dispassionate algorithms will ensure a level of fairness that we are denied because of human frailties. But algorithms have their own shortcomingsand those can pose a serious threat to our personal privacy.

Continue reading here:

When artificial intelligence goes wrong - Livemint