Archive for the ‘AI’ Category


Bob: “I can can I I everything else.”

Alice: “Balls have zero to me to me to me to me to me to me to me to me to.”

To you and I, that passage looks like nonsense. But what if I told you this nonsense was the discussion of what might be the most sophisticated negotiation software on the planet? Negotiation software that had learned, and evolved, to get the best deal possible with more speed and efficiency–and perhaps, hidden nuance–than you or I ever could? Because it is.

This conversation occurred between two AI agents developed inside Facebook. At first, they were speaking to each other in plain old English. But then researchers realized they’d made a mistake in programming.

“There was no reward to sticking to English language,” says Dhruv Batra, visiting research scientist from Georgia Tech at Facebook AI Research (FAIR). As these two agents competed to get the best deal–a very effective bit of AI vs. AI dogfighting researchers have dubbed a “generative adversarial network”–neither was offered any sort of incentive for speaking as a normal person would. So they began to diverge, eventually rearranging legible words into seemingly nonsensical sentences.

“Agents will drift off understandable language and invent codewords for themselves,” says Batra, speaking to a now-predictable phenomenon that Facebook as observed again, and again, and again. “Like if I say ‘the’ five times, you interpret that to mean I want five copies of this item. This isn’t so different from the way communities of humans create shorthands.”

Indeed. Humans have developed unique dialects for everything from trading pork bellies on the floor of the Mercantile Exchange to hunting down terrorists as Seal Team Six–simply because humans sometimes perform better by not abiding to normal language conventions.
So should we let our software do the same thing? Should we allow AI to evolve its dialects for specific tasks that involve speaking to other AIs? To essentially gossip out of our earshot? Maybe; it offers us the possibility of a more interoperable world, a more perfect place where iPhones talk to refrigerators that talk to your car without a second thought.

The tradeoff is that we, as humanity, would have no clue what those machines were actually saying to one another.

Facebook ultimately opted to require its negotiation bots to speak in plain old English. “Our interest was having bots who could talk to people,” says Mike Lewis, research scientist at FAIR. Facebook isn’t alone in that perspective. When I inquired to Microsoft about computer-to-computer languages, a spokesperson clarified that Microsoft was more interested in human-to-computer speech. Meanwhile, Google, Amazon, and Apple are all also focusing incredible energies on developing conversational personalities for human consumption. They’re the next wave of user interface, like the mouse and keyboard for the AI era.

The other issue, as Facebook admits, is that it has no way of truly understanding any divergent computer language. “It’s important to remember, there aren’t bilingual speakers of AI and human languages,” says Batra. We already don’t generally understand how complex AIs think because we can’t really see inside their thought process. Adding AI-to-AI conversations to this scenario would only make that problem worse.

But at the same time, it feels shortsighted, doesn’t it? If we can build software that can speak to other software more efficiently, shouldn’t we use that? Couldn’t there be some benefit?

Because, again, we absolutely can lead machines to develop their own languages. Facebook has three published papers proving it. “It’s definitely possible, it’s possible that [language] can be compressed, not just to save characters, but compressed to a form that it could express a sophisticated thought,” says Batra. Machines can converse with any baseline building blocks they’re offered. That might start with human vocabulary, as with Facebook’s negotiation bots. Or it could start with numbers, or binary codes. But as machines develop meanings, these symbols become “tokens”–they’re imbued with rich meanings. As Dauphin points out, machines might not think as you or I do, but tokens allow them to exchange incredibly complex thoughts through the simplest of symbols. The way I think about it is with algebra: If A + B = C, the “A” could encapsulate almost anything. But to a computer, what “A” can mean is so much bigger than what that “A” can mean to a person, because computers have no outright limit on processing power.

“It’s perfectly possible for a special token to mean a very complicated thought,” says Batra. “The reason why humans have this idea of decomposition, breaking ideas into simpler concepts, it’s because we have a limit to cognition.” Computers don’t need to simplify concepts. They have the raw horsepower to process them.

But how could any of this technology actually benefit the world, beyond these theoretical discussions? Would our servers be able to operate more efficiently with bots speaking to one another in shorthand? Could microsecond processes, like algorithmic trading, see some reasonable increase? Chatting with Facebook, and various experts, I couldn’t get a firm answer.

However, as paradoxical as this might sound, we might see big gains in such software better understanding our intent. While two computers speaking their own language might be more opaque, an algorithm predisposed to learn new languages might chew through strange new data we feed it more effectively. For example, one researcher recently tried to teach a neural net to create new colors and name them. It was terrible at it, generating names like Sudden Pine and Clear Paste (that clear paste, by the way, was labeled on a light green). But then they made a simple change to the data they were feeding the machine to train it. They made everything lowercase–because lowercase and uppercase letters were confusing it. Suddenly, the color-creating AI was working, well, pretty well! And for whatever reason, it preferred, and performed better, with RGB values as opposed to other numerical color codes.

Why did these simple data changes matter? Basically, the researcher did a better job at speaking the computer’s language. As one coder put it to me, “Getting the data into a format that makes sense for machine learning is a huge undertaking right now and is more art than science. English is a very convoluted and complicated language and not at all amicable for machine learning.”

In other words, machines allowed to speak and generate machine languages could somewhat ironically allow us to communicate with (and even control) machines better, simply because they’d be predisposed to have a better understanding of the words we speak.
As one insider at a major AI technology company told me: No, his company wasn’t actively interested in AIs that generated their own custom languages. But if it were, the greatest advantage he imagined was that it could conceivably allow software, apps, and services to learn to speak to each other without human intervention.

Right now, companies like Apple have to build APIs–basically a software bridge–involving all sorts of standards that other companies need to comply with in order for their products to communicate. However, APIs can take years to develop, and their standards are heavily debated across the industry in decade-long arguments. But software, allowed to freely learn how to communicate with other software, could generate its own shorthands for us. That means our “smart devices” could learn to interoperate, no API required.

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



Behind a locked door in a white-walled basement in a research building in Tempe, Ariz., a monkey sits stone-still in a chair, eyes locked on a computer screen. From his head protrudes a bundle of wires; from his mouth, a plastic tube. As he stares, a picture of a green cursor on the black screen floats toward the corner of a cube. The monkey is moving it with his mind.

The monkey, a rhesus macaque named Oscar, has electrodes implanted in his motor cortex, detecting electrical impulses that indicate mental activity and translating them to the movement of the ball on the screen. The computer isn’t reading his mind, exactly — Oscar’s own brain is doing a lot of the lifting, adapting itself by trial and error to the delicate task of accurately communicating its intentions to the machine. (When Oscar succeeds in controlling the ball as instructed, the tube in his mouth rewards him with a sip of his favorite beverage, Crystal Light.) It’s not technically telekinesis, either, since that would imply that there’s something paranormal about the process. It’s called a “brain-computer interface” (BCI). And it just might represent the future of the relationship between human and machine.

Stephen Helms Tillery’s laboratory at Arizona State University is one of a growing number where researchers are racing to explore the breathtaking potential of BCIs and a related technology, neuroprosthetics. The promise is irresistible: from restoring sight to the blind, to helping the paralyzed walk again, to allowing people suffering from locked-in syndrome to communicate with the outside world. In the past few years, the pace of progress has been accelerating, delivering dazzling headlines seemingly by the week.

At Duke University in 2008, a monkey named Idoya walked on a treadmill, causing a robot in Japan to do the same. Then Miguel Nicolelis stopped the monkey’s treadmill — and the robotic legs kept walking, controlled by Idoya’s brain. At Andrew Schwartz’s lab at the University of Pittsburgh in December 2012, a quadriplegic woman named Jan Scheuermann learned to feed herself chocolate by mentally manipulating a robotic arm. Just last month, Nicolelis’ lab set up what it billed as the first brain-to-brain interface, allowing a rat in North Carolina to make a decision based on sensory data beamed via Internet from the brain of a rat in Brazil.

So far the focus has been on medical applications — restoring standard-issue human functions to people with disabilities. But it’s not hard to imagine the same technologies someday augmenting capacities. If you can make robotic legs walk with your mind, there’s no reason you can’t also make them run faster than any sprinter. If you can control a robotic arm, you can control a robotic crane. If you can play a computer game with your mind, you can, theoretically at least, fly a drone with your mind.

It’s tempting and a bit frightening to imagine that all of this is right around the corner, given how far the field has already come in a short time. Indeed, Nicolelis — the media-savvy scientist behind the “rat telepathy” experiment — is aiming to build a robotic bodysuit that would allow a paralyzed teen to take the first kick of the 2014 World Cup. Yet the same factor that has made the explosion of progress in neuroprosthetics possible could also make future advances harder to come by: the almost unfathomable complexity of the human brain.

From I, Robot to Skynet, we’ve tended to assume that the machines of the future would be guided by artificial intelligence — that our robots would have minds of their own. Over the decades, researchers have made enormous leaps in artificial intelligence (AI), and we may be entering an age of “smart objects” that can learn, adapt to, and even shape our habits and preferences. We have planes that fly themselves, and we’ll soon have cars that do the same. Google has some of the world’s top AI minds working on making our smartphones even smarter, to the point that they can anticipate our needs. But “smart” is not the same as “sentient.” We can train devices to learn specific behaviors, and even out-think humans in certain constrained settings, like a game of Jeopardy. But we’re still nowhere close to building a machine that can pass the Turing test, the benchmark for human-like intelligence. Some experts doubt we ever will.

Philosophy aside, for the time being the smartest machines of all are those that humans can control. The challenge lies in how best to control them. From vacuum tubes to the DOS command line to the Mac to the iPhone, the history of computing has been a progression from lower to higher levels of abstraction. In other words, we’ve been moving from machines that require us to understand and directly manipulate their inner workings to machines that understand how we work and respond readily to our commands. The next step after smartphones may be voice-controlled smart glasses, which can intuit our intentions all the more readily because they see what we see and hear what we hear.

The logical endpoint of this progression would be computers that read our minds, computers we can control without any physical action on our part at all. That sounds impossible. After all, if the human brain is so hard to compute, how can a computer understand what’s going on inside it?

It can’t. But as it turns out, it doesn’t have to — not fully, anyway. What makes brain-computer interfaces possible is an amazing property of the brain called neuroplasticity: the ability of neurons to form new connections in response to fresh stimuli. Our brains are constantly rewiring themselves to allow us to adapt to our environment. So when researchers implant electrodes in a part of the brain that they expect to be active in moving, say, the right arm, it’s not essential that they know in advance exactly which neurons will fire at what rate. When the subject attempts to move the robotic arm and sees that it isn’t quite working as expected, the person — or rat or monkey — will try different configurations of brain activity. Eventually, with time and feedback and training, the brain will hit on a solution that makes use of the electrodes to move the arm.

That’s the principle behind such rapid progress in brain-computer interface and neuroprosthetics. Researchers began looking into the possibility of reading signals directly from the brain in the 1970s, and testing on rats began in the early 1990s. The first big breakthrough for humans came in Georgia in 1997, when a scientist named Philip Kennedy used brain implants to allow a “locked in” stroke victim named Johnny Ray to spell out words by moving a cursor with his thoughts. (It took him six exhausting months of training to master the process.) In 2008, when Nicolelis got his monkey at Duke to make robotic legs run a treadmill in Japan, it might have seemed like mind-controlled exoskeletons for humans were just another step or two away. If he succeeds in his plan to have a paralyzed youngster kick a soccer ball at next year’s World Cup, some will pronounce the cyborg revolution in full swing.

Schwartz, the Pittsburgh researcher who helped Jan Scheuermann feed herself chocolate in December, is optimistic that neuroprosthetics will eventually allow paralyzed people to regain some mobility. But he says that full control over an exoskeleton would require a more sophisticated way to extract nuanced information from the brain. Getting a pair of robotic legs to walk is one thing. Getting robotic limbs to do everything human limbs can do may be exponentially more complicated. “The challenge of maintaining balance and staying upright on two feet is a difficult problem, but it can be handled by robotics without a brain. But if you need to move gracefully and with skill, turn and step over obstacles, decide if it’s slippery outside — that does require a brain. If you see someone go up and kick a soccer ball, the essential thing to ask is, ‘OK, what would happen if I moved the soccer ball two inches to the right?'” The idea that simple electrodes could detect things as complex as memory or cognition, which involve the firing of billions of neurons in patterns that scientists can’t yet comprehend, is far-fetched, Schwartz adds.

That’s not the only reason that companies like Apple and Google aren’t yet working on devices that read our minds (as far as we know). Another one is that the devices aren’t portable. And then there’s the little fact that they require brain surgery.

A different class of brain-scanning technology is being touted on the consumer market and in the media as a way for computers to read people’s minds without drilling into their skulls. It’s called electroencephalography, or EEG, and it involves headsets that press electrodes against the scalp. In an impressive 2010 TED Talk, Tan Le of the consumer EEG-headset company Emotiv Lifescience showed how someone can use her company’s EPOC headset to move objects on a computer screen.

Skeptics point out that these devices can detect only the crudest electrical signals from the brain itself, which is well-insulated by the skull and scalp. In many cases, consumer devices that claim to read people’s thoughts are in fact relying largely on physical signals like skin conductivity and tension of the scalp or eyebrow muscles.

Robert Oschler, a robotics enthusiast who develops apps for EEG headsets, believes the more sophisticated consumer headsets like the Emotiv EPOC may be the real deal in terms of filtering out the noise to detect brain waves. Still, he says, there are limits to what even the most advanced, medical-grade EEG devices can divine about our cognition. He’s fond of an analogy that he attributes to Gerwin Schalk, a pioneer in the field of invasive brain implants. The best EEG devices, he says, are “like going to a stadium with a bunch of microphones: You can’t hear what any individual is saying, but maybe you can tell if they’re doing the wave.” With some of the more basic consumer headsets, at this point, “it’s like being in a party in the parking lot outside the same game.”

It’s fairly safe to say that EEG headsets won’t be turning us into cyborgs anytime soon. But it would be a mistake to assume that we can predict today how brain-computer interface technology will evolve. Just last month, a team at Brown University unveiled a prototype of a low-power, wireless neural implant that can transmit signals to a computer over broadband. That could be a major step forward in someday making BCIs practical for everyday use. Meanwhile, researchers at Cornell last week revealed that they were able to use fMRI, a measure of brain activity, to detect which of four people a research subject was thinking about at a given time. Machines today can read our minds in only the most rudimentary ways. But such advances hint that they may be able to detect and respond to more abstract types of mental activity in the always-changing future.