Archive for the ‘artificial intelligence’ Category

MICROSOFT IS BUYING a deep learning startup based in Montreal, a global hub for deep learning research. But two years ago, this startup wasn’t based in Montreal, and it had nothing to do with deep learning. Which just goes to show: striking it big in the world of tech is all about being in the right place at the right time with the right idea.

Sam Pasupalak and Kaheer Suleman founded Maluuba in 2011 as students at the University of Waterloo, about 400 miles from Montreal. The company’s name is an insider’s nod to one of their undergraduate computer science classes. From an office in Waterloo, they started building something like Siri, the digital assistant that would soon arrive on the iPhone, and they built it in much the same way Apple built the original, using techniques that had driven the development of conversational computing for years—techniques that require extremely slow and meticulous work, where engineers construct AI one tiny piece at a time. But as they toiled away in Waterloo, companies like Google and Facebook embraced deep neural networks, and this technology reinvented everything from image recognition to machine translations, rapidly learning these tasks by analyzing vast amounts of data. Soon, Pasupalak and Suleman realized they should change tack.

In December 2015, the two founders opened a lab in Montreal, and they started recruiting deep learning specialists from places like McGill University and the University of Montreal. Just thirteen months later, after growing to a mere 50 employees, the company sold itself to Microsoft. And that’s not an unusual story. The giants of tech are buying up deep learning startups almost as quickly as they’re created. At the end of December, Uber acquired Geometric Logic, a two-year old AI startup spanning fifteen academic researchers that offered no product and no published research. The previous summer, Twitter paid a reported $150 million for Magic Pony, a two-year-old deep learning startup based in the UK. And in recent months, similarly small, similarly young deep learning companies have disappeared into the likes of General Electric, Salesforce, and Apple.

Microsoft did not disclose how much it paid for Maluuba, but some of these deep learning acquisitions have reached hefty sums, including Intel’s $400 million purchase of Nervana and Google’s $650 million acquisition of DeepMind, the British AI lab that made headlines last spring when it cracked the ancient game of Go, a feat experts didn’t expect for another decade.

At the same time, Microsoft’s buy is a little different than the rest. Maluuba is a deep learning company that focuses on natural language understanding, the ability to not just recognize the words that come out of our mouths but actually understand them and respond in kind—the breed of AI needed to build a good chatbot. Now that deep learning has proven so effective with speech recognition, image recognition, and translation, natural language is the next frontier. “In the past, people had to build large lexicons, dictionaries, ontologies,” Suleman says. “But with neural nets, we no longer need to do that. A neural net can learn from raw data.”

The acquisition is part of an industry-wide race towards digital assistants and chatbots that can converse like a human. Yes, we already have digital assistants like Microsoft Cortana, the Google Search Assistant, Facebook M, and Amazon Alexa. And chatbots are everywhere. But none of these services know how to chat (a particular problem for the chatbots). So, Microsoft, Google, Facebook, and Amazon are now looking at deep learning as a way of improving the state of the art.

Two summers ago, Google published a research paper describing a chatbot underpinned by deep learning that could debate the meaning of life (in a way). Around the same time, Facebook described an experimental system that could read a shortened form of The Lord of the Rings and answer questions about the Tolkien trilogy. Amazon is gathering data for similar work. And, none too surprisingly, Microsoft is gobbling up a startup that only just moved into the same field.

Winning the Game
Deep neural networks are complex mathematical systems that learn to perform discrete tasks by recognizing patterns in vast amounts of digital data. Feed millions of photos into a neural network, for instance, and it can learn to identify objects and people in photos. Pairing these systems with the enormous amounts of computing power inside their data centers, companies like Google, Facebook, and Microsoft have pushed artificial intelligence far further, far more quickly, than they ever could in the past.

Now, these companies hope to reinvent natural language understanding in much the same way. But there are big caveats: It’s a much harder task, and the work has only just begun. “Natural language is an area where more research needs to be done in terms of research, even basic research,” says University of Montreal professor Yoshua Bengio, one of the founding fathers of the deep learning movement and an advisor to Maluuba.

Part of the problem is that researchers don’t yet have the data needed to train neural networks for true conversation, and Maluuba is among those working to fill the void. Like Facebook and Amazon, it’s building brand new datasets for training natural language models: One involves questions and answers, and the other focuses on conversational dialogue. What’s more, the company is sharing this data with the larger community of researchers and encouraging then\m to share their own—a common strategy that seeks to accelerate the progress of AI research.

But even with adequate data, the task is quite different from image recognition or translation. Natural language isn’t necessarily something that neural networks can solve on their own. Dialogue isn’t a single task. It’s a series of tasks, each building on the one before. A neural network can’t just identify a pattern in a single piece of data. It must somehow identify patterns across an endless stream of data—and a keep a “memory” of this stream. That’s why Maluuba is exploring AI beyond neural networks, including a technique called reinforcement learning.

With reinforcement learning, a system repeats the same task over and over again, while carefully keeping tabs on what works and what doesn’t. Engineers at Google’s DeepMind lab used this method in building AlphaGo, the system that topped Korean grandmaster Lee Sedol at the ancient game of Go. In essence, the machine learned to play Go at a higher level than any human by playing game after game against itself, tracking which moves won the most territory on the board. In similar fashion, reinforcement learning can help machines learn to carry on a conversation. Like a game, Bengio says, dialogue is interactive. It’s a back and forth.

For Microsoft, winning the game of conversation means winning an enormous market. Natural language could streamline practically any computer interface. With this in mind, the company is already building an army of chatbots, but so far, the results are mixed. In China, the company says, its Xiaoice chatbot has been used by 40 million people. But when it first unleashed a similar bot in the US, the service was coaxed into spewing racism, and the replacement is flawed in so many other ways. That’s why Microsoft acquired Maluuba. The startup was in the right place at the right time. And it may carry the right idea.

https://www.wired.com/2017/01/microsoft-thinks-machines-can-learn-converse-chats-become-game/

Google AI computers have created their own secret language, creating a fascinating and existentially challenging development.

In September, Google announced that its Neural Machine Translation system had gone live. It uses deep learning to produce better, more natural translations between languages.

Following on this success, GNMT’s creators were curious about something. If you teach the translation system to translate English to Korean and vice versa, and also English to Japanese and vice versa… could it translate Korean to Japanese, without resorting to English as a bridge between them?

This is called zero-shot translation, illustrated below.

Indeed, Google’s AI has evolves to produce reasonable translations between two languages that it has not explicitly linked in any way.

But this raised a second question. If the computer is able to make connections between concepts and words that have not been formally linked… does that mean that the computer has formed a concept of shared meaning for those words, meaning at a deeper level than simply that one word or phrase is the equivalent of another?

n other words, has the computer developed its own internal language to represent the concepts it uses to translate between other languages? Based on how various sentences are related to one another in the memory space of the neural network, Google’s language and AI boffins think that it has.

This “interlingua” seems to exist as a deeper level of representation that sees similarities between a sentence or word in all three languages. Beyond that, it’s hard to say, since the inner processes of complex neural networks are infamously difficult to describe.

It could be something sophisticated, or it could be something simple. But the fact that it exists at all — an original creation of the system’s own to aid in its understanding of concepts it has not been trained to understand — is, philosophically speaking, pretty powerful stuff.

Google’s AI translation tool seems to have invented its own secret internal language

Thanks to Kebmodee for bringing this to the attention of the It’s Interesting community.

by Jeremy Kahn

Google’s DeepMind AI unit, which earlier this year achieved a breakthrough in computer intelligence by creating software that beat the world’s best human player at the strategy game Go, is turning its attention to the sci-fi video game Starcraft II.

The company said it had reached a deal with Blizzard Entertainment Inc., the Irvine, California-based division of Activision Blizzard, which makes the Starcraft game series, to create an interface to let artificial intelligence researchers connect machine-learning software to the game.

London-based DeepMind, which Google purchased in 2014, has not said it has created software that can play Starcraft expertly — at least not yet. “We’re still a long way from being able to challenge a professional human player,” DeepMind research scientist Oriol Vinyals said in a blog post Friday. But the company’s announcement shows it’s looking seriously at Starcraft as a candidate for a breakthrough in machine intelligence.

Starcraft fascinates artificial intelligence researchers because it comes closer to simulating “the messiness of the real world” than games like chess or Go, Vinyals said. “An agent that can play Starcraft will need to demonstrate effective use of memory, an ability to plan over a long time and the capacity to adapt plans to new information,” he said, adding that techniques required to create a machine-learning system that mastered these skills in order to play Starcraft “could ultimately transfer to real-world tasks.”

Virtual Mining

In the game, which is played in real-time over the internet, players choose one of three character types, each of which has distinct strengths and weaknesses. Players must run an in-game economy, discovering and mining minerals and other commodities in order to conquer new territory.A successful player needs to remember large volumes of information about places they’ve scouted in the past, even when those places are not immediately observable on their screen.

The player’s view of what an opposing player is doing is limited — unlike chess or Go where opponents can observe the whole board at one time. Furthermore,unlike in a game where players take turns, a machine-learning system has to deal with an environment that is constantly in flux. Starcraft in particular also requires an ability to plan both a long-term strategy and make very quick tactical decisions to stay ahead of an opponent — and designing software that is good at both types of decision-making is difficult.

Facebook, Microsoft

Researchers at Facebook Inc. and Microsoft Corp. have also published papers on ways to interface artificial intelligence systems with earlier versions of Starcraft. And some Starcraft-playing bots have already been created, but so far these systems have not been able to defeat talented human players.

Microsoft Chief Executive Officer Satya Nadella has taken swipes at Google’s focus on games in its AI research, telling the audience at a company event in Atlanta in September that Microsoft was “not pursuing AI to beat humans at games” and that Microsoft wanted to build AI “to solve the most pressing problems of our society and economy.”

Games have long-served as important tests and milestones for artificial intelligence research. In the mid-1990s, International Business Machines Corp.’s supercomputer Deep Blue defeated world chess champion Garry Kasparov on several occasions. IBM’s Watson artificial intelligence beat top human players in the game show Jeopardy in 2011, an achievement that showcased IBM’s strides in natural language processing. In 2015, DeepMind developed machine learning software that taught itself how to play dozens of retro Atari games, such as Breakout, as well or better than a human. Then, in March of 2016, DeepMind’s Alpha Go program, trained in a different way, defeated Go world champion Lee Sodol.

In the twenty years since Starcraft debuted, the game has acquired a massive and devoted following. More than 9.5 million copies of the original game were sold within the first decade of its release, with more than half of those being sold in Korea, where the game was especially popular. Starcraft II shattered sales records for a strategy game when it was released in 2010, selling 1.5 million copies within 48 hours. Pitting two players against one another in real-time, Starcraft was a pioneer in professional video game competitions and remains an important game in the world of e-sports, although its prominence has since been eclipsed by other games.

http://www.detroitnews.com/story/business/2016/11/05/deepmind-master-go-takes-video-game-starcraft/93370028/

by Bryan Nelson

Quantum physics has some spooky, anti-intuitive effects, but it could also be essential to how actual intuition works, at least in regards to artificial intelligence.

In a new study, researcher Vedran Dunjko and co-authors applied a quantum analysis to a field within artificial intelligence called reinforcement learning, which deals with how to program a machine to make appropriate choices to maximize a cumulative reward. The field is surprisingly complex and must take into account everything from game theory to information theory.

Dunjko and his team found that quantum effects, when applied to reinforcement learning in artificial intelligence systems, could provide quadratic improvements in learning efficiency, reports Phys.org. Exponential improvements might even be possible over short-term performance tasks. The study was published in the journal Physical Review Letters.

“This is, to our knowledge, the first work which shows that quantum improvements are possible in more general, interactive learning tasks,” explained Dunjko. “Thus, it opens up a new frontier of research in quantum machine learning.”

One of the key quantum effects in regards to learning is quantum superposition, which potentially allows a machine to perform many steps simultaneously. Such a system has vastly improved processing power, which allows it to compute more variables when making decisions.

The research is tantalizing, in part because it mirrors some theories about how biological brains might produce higher cognitive states, possibly even being related to consciousness. For instance, some scientists have proposed the idea that our brains pull off their complex calculations by making use of quantum computation.

Could quantum effects unlock consciousness in our machines? Quantum physics isn’t likely to produce HAL from “2001: A Space Odyssey” right away; the most immediate improvements in artificial intelligence will likely come in complex fields such as climate modeling or automated cars. But eventually, who knows?

You probably won’t want to be taking a joyride in an automated vehicle the moment it becomes conscious, if HAL is an example of what to expect.

“While the initial results are very encouraging, we have only begun to investigate the potential of quantum machine learning,” said Dunjko. “We plan on furthering our understanding of how quantum effects can aid in aspects of machine learning in an increasingly more general learning setting. One of the open questions we are interested in is whether quantum effects can play an instrumental role in the design of true artificial intelligence.”

http://www.mnn.com/green-tech/research-innovations/stories/quantum-artificial-intelligence-could-lead-super-smart-machines

Elon Musk has said that there is only a “one in billions” chance that we’re not living in a computer simulation.

Our lives are almost certainly being conducted within an artificial world powered by AI and highly-powered computers, like in The Matrix, the Tesla and SpaceX CEO suggested at a tech conference in California.

Mr Musk, who has donated huge amounts of money to research into the dangers of artificial intelligence, said that he hopes his prediction is true because otherwise it means the world will end.

“The strongest argument for us probably being in a simulation I think is the following,” he told the Code Conference. “40 years ago we had Pong – two rectangles and a dot. That’s where we were.

“Now 40 years later we have photorealistic, 3D simulations with millions of people playing simultaneously and it’s getting better every year. And soon we’ll have virtual reality, we’ll have augmented reality.

“If you assume any rate of improvement at all, then the games will become indistinguishable from reality, just indistinguishable.”

He said that even if the speed of those advancements dropped by 1000, we would still be moving forward at an intense speed relative to the age of life.

Since that would lead to games that would be indistinguishable from reality that could be played anywhere, “it would seem to follow that the odds that we’re in ‘base reality’ is one in billions”, Mr Musk said.

Asked whether he was saying that the answer to the question of whether we are in a simulated computer game was “yes”, he said the answer is “probably”.

He said that arguably we should hope that it’s true that we live in a simulation. “Otherwise, if civilisation stops advancing, then that may be due to some calamitous event that stops civilisation.”

He said that either we will make simulations that we can’t tell apart from the real world, “or civilisation will cease to exist”.

Mr Musk said that he has had “so many simulation discussions it’s crazy”, and that it got to the point where “every conversation [he had] was the AI/simulation conversation”.

The question of whether what we see is real or simulated has perplexed humans since at least the Ancient philosophers. But it has been given a new and different edge in recent years with the development of powerful computers and artificial intelligence, which some have argued shows how easily such a simulation could be created.

http://www.independent.co.uk/life-style/gadgets-and-tech/news/elon-musk-ai-artificial-intelligence-computer-simulation-gaming-virtual-reality-a7060941.html

t’s been almost 20 years since IBM’s Deep Blue supercomputer beat the reigning world chess champion, Gary Kasparov, for the first time under standard tournament rules. Since then, chess-playing computers have become significantly stronger, leaving the best humans little chance even against a modern chess engine running on a smartphone.

But while computers have become faster, the way chess engines work has not changed. Their power relies on brute force, the process of searching through all possible future moves to find the best next one.

Of course, no human can match that or come anywhere close. While Deep Blue was searching some 200 million positions per second, Kasparov was probably searching no more than five a second. And yet he played at essentially the same level. Clearly, humans have a trick up their sleeve that computers have yet to master.

This trick is in evaluating chess positions and narrowing down the most profitable avenues of search. That dramatically simplifies the computational task because it prunes the tree of all possible moves to just a few branches.

Computers have never been good at this, but today that changes thanks to the work of Matthew Lai at Imperial College London. Lai has created an artificial intelligence machine called Giraffe that has taught itself to play chess by evaluating positions much more like humans and in an entirely different way to conventional chess engines.

Straight out of the box, the new machine plays at the same level as the best conventional chess engines, many of which have been fine-tuned over many years. On a human level, it is equivalent to FIDE International Master status, placing it within the top 2.2 percent of tournament chess players.

The technology behind Lai’s new machine is a neural network. This is a way of processing information inspired by the human brain. It consists of several layers of nodes that are connected in a way that change as the system is trained. This training process uses lots of examples to fine-tune the connections so that the network produces a specific output given a certain input, to recognize the presence of face in a picture, for example.

In the last few years, neural networks have become hugely powerful thanks to two advances. The first is a better understanding of how to fine-tune these networks as they learn, thanks in part to much faster computers. The second is the availability of massive annotated datasets to train the networks.

That has allowed computer scientists to train much bigger networks organized into many layers. These so-called deep neural networks have become hugely powerful and now routinely outperform humans in pattern recognition tasks such as face recognition and handwriting recognition.

So it’s no surprise that deep neural networks ought to be able to spot patterns in chess and that’s exactly the approach Lai has taken. His network consists of four layers that together examine each position on the board in three different ways.

The first looks at the global state of the game, such as the number and type of pieces on each side, which side is to move, castling rights and so on. The second looks at piece-centric features such as the location of each piece on each side, while the final aspect is to map the squares that each piece attacks and defends.

Lai trains his network with a carefully generated set of data taken from real chess games. This data set must have the correct distribution of positions. “For example, it doesn’t make sense to train the system on positions with three queens per side, because those positions virtually never come up in actual games,” he says.

It must also have plenty of variety of unequal positions beyond those that usually occur in top level chess games. That’s because although unequal positions rarely arise in real chess games, they crop up all the time in the searches that the computer performs internally.

And this data set must be huge. The massive number of connections inside a neural network have to be fine-tuned during training and this can only be done with a vast dataset. Use a dataset that is too small and the network can settle into a state that fails to recognize the wide variety of patterns that occur in the real world.

Lai generated his dataset by randomly choosing five million positions from a database of computer chess games. He then created greater variety by adding a random legal move to each position before using it for training. In total he generated 175 million positions in this way.

The usual way of training these machines is to manually evaluate every position and use this information to teach the machine to recognize those that are strong and those that are weak.

But this is a huge task for 175 million positions. It could be done by another chess engine but Lai’s goal was more ambitious. He wanted the machine to learn itself.

Instead, he used a bootstrapping technique in which Giraffe played against itself with the goal of improving its prediction of its own evaluation of a future position. That works because there are fixed reference points that ultimately determine the value of a position—whether the game is later won, lost or drawn.

In this way, the computer learns which positions are strong and which are weak.

Having trained Giraffe, the final step is to test it and here the results make for interesting reading. Lai tested his machine on a standard database called the Strategic Test Suite, which consists of 1,500 positions that are chosen to test an engine’s ability to recognize different strategic ideas. “For example, one theme tests the understanding of control of open files, another tests the understanding of how bishop and knight’s values change relative to each other in different situations, and yet another tests the understanding of center control,” he says.

The results of this test are scored out of 15,000.

Lai uses this to test the machine at various stages during its training. As the bootstrapping process begins, Giraffe quickly reaches a score of 6,000 and eventually peaks at 9,700 after only 72 hours. Lai says that matches the best chess engines in the world.

“[That] is remarkable because their evaluation functions are all carefully hand-designed behemoths with hundreds of parameters that have been tuned both manually and automatically over several years, and many of them have been worked on by human grandmasters,” he adds.

Lai goes on to use the same kind of machine learning approach to determine the probability that a given move is likely to be worth pursuing. That’s important because it prevents unnecessary searches down unprofitable branches of the tree and dramatically improves computational efficiency.

Lai says this probabilistic approach predicts the best move 46 percent of the time and places the best move in its top three ranking, 70 percent of the time. So the computer doesn’t have to bother with the other moves.

That’s interesting work that represents a major change in the way chess engines work. It is not perfect, of course. One disadvantage of Giraffe is that neural networks are much slower than other types of data processing. Lai says Giraffe takes about 10 times longer than a conventional chess engine to search the same number of positions.

But even with this disadvantage, it is competitive. “Giraffe is able to play at the level of an FIDE International Master on a modern mainstream PC,” says Lai. By comparison, the top engines play at super-Grandmaster level.

That’s still impressive. “Unlike most chess engines in existence today, Giraffe derives its playing strength not from being able to see very far ahead, but from being able to evaluate tricky positions accurately, and understanding complicated positional concepts that are intuitive to humans, but have been elusive to chess engines for a long time,” says Lai. “This is especially important in the opening and end game phases, where it plays exceptionally well.”

And this is only the start. Lai says it should be straightforward to apply the same approach to other games. One that stands out is the traditional Chinese game of Go, where humans still hold an impressive advantage over their silicon competitors. Perhaps Lai could have a crack at that next.

http://www.technologyreview.com/view/541276/deep-learning-machine-teaches-itself-chess-in-72-hours-plays-at-international-master/

Thanks to Kebmodee for bringing this to the It’s Interesting community.

defense-large

The Office of Naval Research will award $7.5 million in grant money over five years to university researchers from Tufts, Rensselaer Polytechnic Institute, Brown, Yale and Georgetown to explore how to build a sense of right and wrong and moral consequence into autonomous robotic systems.

“Even though today’s unmanned systems are ‘dumb’ in comparison to a human counterpart, strides are being made quickly to incorporate more automation at a faster pace than we’ve seen before,” Paul Bello, director of the cognitive science program at the Office of Naval Research told Defense One. “For example, Google’s self-driving cars are legal and in-use in several states at this point. As researchers, we are playing catch-up trying to figure out the ethical and legal implications. We do not want to be caught similarly flat-footed in any kind of military domain where lives are at stake.”

The United States military prohibits lethal fully autonomous robots. And semi-autonomous robots can’t “select and engage individual targets or specific target groups that have not been previously selected by an authorized human operator,” even in the event that contact with the operator is cut off, according to a 2012 Department of Defense policy directive.

“Even if such systems aren’t armed, they may still be forced to make moral decisions,” Bello said. For instance, in a disaster scenario, a robot may be forced to make a choice about whom to evacuate or treat first, a situation where a bot might use some sense of ethical or moral reasoning. “While the kinds of systems we envision have much broader use in first-response, search-and-rescue and in the medical domain, we can’t take the idea of in-theater robots completely off the table,” Bello said.

Some members of the artificial intelligence, or AI, research and machine ethics communities were quick to applaud the grant. “With drones, missile defines, autonomous vehicles, etc., the military is rapidly creating systems that will need to make moral decisions,” AI researcher Steven Omohundro told Defense One. “Human lives and property rest on the outcomes of these decisions and so it is critical that they be made carefully and with full knowledge of the capabilities and limitations of the systems involved. The military has always had to define ‘the rules of war’ and this technology is likely to increase the stakes for that.”

“We’re talking about putting robots in more and more contexts in which we can’t predict what they’re going to do, what kind of situations they’ll encounter. So they need to do some kind of ethical reasoning in order to sort through various options,” said Wendell Wallach, the chair of the Yale Technology and Ethics Study Group and author of the book Moral Machines: Teaching Robots Right From Wrong.

The sophistication of cutting-edge drones like British BAE Systems’s batwing-shaped Taranis and Northrop Grumman’s X-47B reveal more self-direction creeping into ever more heavily armed systems. The X-47B, Wallach said, is “enormous and it does an awful lot of things autonomously.”

But how do you code something as abstract as moral logic into a bunch of transistors? The vast openness of the problem is why the framework approach is important, says Wallach. Some types of morality are more basic, thus more code-able, than others.

“There’s operational morality, functional morality, and full moral agency,” Wallach said. “Operational morality is what you already get when the operator can discern all the situations that the robot may come under and program in appropriate responses… Functional morality is where the robot starts to move into situations where the operator can’t always predict what [the robot] will encounter and [the robot] will need to bring some form of ethical reasoning to bear.”

It’s a thick knot of questions to work through. But, Wallach says, with a high potential to transform the battlefield.

“One of the arguments for [moral] robots is that they may be even better than humans in picking a moral course of action because they may consider more courses of action,” he said.

Ronald Arkin, an AI expert from Georgia Tech and author of the book Governing Lethal Behavior in Autonomous Robots, is a proponent of giving machines a moral compass. “It is not my belief that an unmanned system will be able to be perfectly ethical in the battlefield, but I am convinced that they can perform more ethically than human soldiers are capable of,” Arkin wrote in a 2007 research paper (PDF). Part of the reason for that, he said, is that robots are capable of following rules of engagement to the letter, whereas humans are more inconsistent.

AI robotics expert Noel Sharkey is a detractor. He’s been highly critical of armed drones in general. and has argued that autonomous weapons systems cannot be trusted to conform to international law.

“I do not think that they will end up with a moral or ethical robot,” Sharkey told Defense One. “For that we need to have moral agency. For that we need to understand others and know what it means to suffer. The robot may be installed with some rules of ethics but it won’t really care. It will follow a human designer’s idea of ethics.”

“The simple example that has been given to the press about scheduling help for wounded soldiers is a good one. My concern would be if [the military] were to extend a system like this for lethal autonomous weapons – weapons where the decision to kill is delegated to a machine; that would be deeply troubling,” he said.

This week, Sharkey and Arkin are debating the issue of whether or not morality can be built into AI systems before the U.N. where they may find an audience very sympathetic to the idea that a moratorium should be placed on the further development of autonomous armed robots.

Christof Heyns, U.N. special rapporteur on extrajudicial, summary or arbitrary executions for the Office of the High Commissioner for Human Rights, is calling for a moratorium. “There is reason to believe that states will, inter alia, seek to use lethal autonomous robotics for targeted killing,” Heyns said in an April 2013 report to the U.N.

The Defense Department’s policy directive on lethal autonomy offers little reassurance here since the department can change it without congressional approval, at the discretion of the chairman of the Joint Chiefs of Staff and two undersecretaries of Defense. University of Denver scholar Heather Roff, in an op-ed for the Huffington Post, calls that a “disconcerting” lack of oversight and notes that “fielding of autonomous weapons then does not even raise to the level of the Secretary of Defense, let alone the president.”

If researchers can prove that robots can do moral math, even if in some limited form, they may be able to diffuse rising public anger and mistrust over armed unmanned vehicles. But it’s no small task.

“This is a significantly difficult problem and it’s not clear we have an answer to it,” said Wallach. “Robots both domestic and militarily are going to find themselves in situations where there are a number of courses of actions and they are going to need to bring some kinds of ethical routines to bear on determining the most ethical course of action. If we’re moving down this road of increasing autonomy in robotics, and that’s the same as Google cars as it is for military robots, we should begin now to do the research to how far can we get in ensuring the robot systems are safe and can make appropriate decisions in the context they operate.”

Thanks to Kebmodee for bringing this to the attention of the It’s Interesting community.

http://www.defenseone.com/technology/2014/05/now-military-going-build-robots-have-morals/84325/?oref=d-topstory