Tag Archives: operator
#433758 DeepMind’s New Research Plan to Make ...
Making sure artificial intelligence does what we want and behaves in predictable ways will be crucial as the technology becomes increasingly ubiquitous. It’s an area frequently neglected in the race to develop products, but DeepMind has now outlined its research agenda to tackle the problem.
AI safety, as the field is known, has been gaining prominence in recent years. That’s probably at least partly down to the overzealous warnings of a coming AI apocalypse from well-meaning, but underqualified pundits like Elon Musk and Stephen Hawking. But it’s also recognition of the fact that AI technology is quickly pervading all aspects of our lives, making decisions on everything from what movies we watch to whether we get a mortgage.
That’s why DeepMind hired a bevy of researchers who specialize in foreseeing the unforeseen consequences of the way we built AI back in 2016. And now the team has spelled out the three key domains they think require research if we’re going to build autonomous machines that do what we want.
In a new blog designed to provide updates on the team’s work, they introduce the ideas of specification, robustness, and assurance, which they say will act as the cornerstones of their future research. Specification involves making sure AI systems do what their operator intends; robustness means a system can cope with changes to its environment and attempts to throw it off course; and assurance involves our ability to understand what systems are doing and how to control them.
A classic thought experiment designed to illustrate how we could lose control of an AI system can help illustrate the problem of specification. Philosopher Nick Bostrom’s posited a hypothetical machine charged with making as many paperclips as possible. Because the creators fail to add what they might assume are obvious additional goals like not harming people, the AI wipes out humanity so we can’t switch it off before turning all matter in the universe into paperclips.
Obviously the example is extreme, but it shows how a poorly-specified goal can lead to unexpected and disastrous outcomes. Properly codifying the desires of the designer is no easy feat, though; often there are not neat ways to encompass both the explicit and implicit goals in ways that are understandable to the machine and don’t leave room for ambiguities, meaning we often rely on incomplete approximations.
The researchers note recent research by OpenAI in which an AI was trained to play a boat-racing game called CoastRunners. The game rewards players for hitting targets laid out along the race route. The AI worked out that it could get a higher score by repeatedly knocking over regenerating targets rather than actually completing the course. The blog post includes a link to a spreadsheet detailing scores of such examples.
Another key concern for AI designers is making their creation robust to the unpredictability of the real world. Despite their superhuman abilities on certain tasks, most cutting-edge AI systems are remarkably brittle. They tend to be trained on highly-curated datasets and so can fail when faced with unfamiliar input. This can happen by accident or by design—researchers have come up with numerous ways to trick image recognition algorithms into misclassifying things, including thinking a 3D printed tortoise was actually a gun.
Building systems that can deal with every possible encounter may not be feasible, so a big part of making AIs more robust may be getting them to avoid risks and ensuring they can recover from errors, or that they have failsafes to ensure errors don’t lead to catastrophic failure.
And finally, we need to have ways to make sure we can tell whether an AI is performing the way we expect it to. A key part of assurance is being able to effectively monitor systems and interpret what they’re doing—if we’re basing medical treatments or sentencing decisions on the output of an AI, we’d like to see the reasoning. That’s a major outstanding problem for popular deep learning approaches, which are largely indecipherable black boxes.
The other half of assurance is the ability to intervene if a machine isn’t behaving the way we’d like. But designing a reliable off switch is tough, because most learning systems have a strong incentive to prevent anyone from interfering with their goals.
The authors don’t pretend to have all the answers, but they hope the framework they’ve come up with can help guide others working on AI safety. While it may be some time before AI is truly in a position to do us harm, hopefully early efforts like these will mean it’s built on a solid foundation that ensures it is aligned with our goals.
Image Credit: cono0430 / Shutterstock.com Continue reading
#432331 $10 million XPRIZE Aims for Robot ...
Ever wished you could be in two places at the same time? The XPRIZE Foundation wants to make that a reality with a $10 million competition to build robot avatars that can be controlled from at least 100 kilometers away.
The competition was announced by XPRIZE founder Peter Diamandis at the SXSW conference in Austin last week, with an ambitious timeline of awarding the grand prize by October 2021. Teams have until October 31st to sign up, and they need to submit detailed plans to a panel of judges by the end of next January.
The prize, sponsored by Japanese airline ANA, has given contestants little guidance on how they expect them to solve the challenge other than saying their solutions need to let users see, hear, feel, and interact with the robot’s environment as well as the people in it.
XPRIZE has also not revealed details of what kind of tasks the robots will be expected to complete, though they’ve said tasks will range from “simple” to “complex,” and it should be possible for an untrained operator to use them.
That’s a hugely ambitious goal that’s likely to require teams to combine multiple emerging technologies, from humanoid robotics to virtual reality high-bandwidth communications and high-resolution haptics.
If any of the teams succeed, the technology could have myriad applications, from letting emergency responders enter areas too hazardous for humans to helping people care for relatives who live far away or even just allowing tourists to visit other parts of the world without the jet lag.
“Our ability to physically experience another geographic location, or to provide on-the-ground assistance where needed, is limited by cost and the simple availability of time,” Diamandis said in a statement.
“The ANA Avatar XPRIZE can enable creation of an audacious alternative that could bypass these limitations, allowing us to more rapidly and efficiently distribute skill and hands-on expertise to distant geographic locations where they are needed, bridging the gap between distance, time, and cultures,” he added.
Interestingly, the technology may help bypass an enduring hand break on the widespread use of robotics: autonomy. By having a human in the loop, you don’t need nearly as much artificial intelligence analyzing sensory input and making decisions.
Robotics software is doing a lot more than just high-level planning and strategizing, though. While a human moves their limbs instinctively without consciously thinking about which muscles to activate, controlling and coordinating a robot’s components requires sophisticated algorithms.
The DARPA Robotics Challenge demonstrated just how hard it was to get human-shaped robots to do tasks humans would find simple, such as opening doors, climbing steps, and even just walking. These robots were supposedly semi-autonomous, but on many tasks they were essentially tele-operated, and the results suggested autonomy isn’t the only problem.
There’s also the issue of powering these devices. You may have noticed that in a lot of the slick web videos of humanoid robots doing cool things, the machine is attached to the roof by a large cable. That’s because they suck up huge amounts of power.
Possibly the most advanced humanoid robot—Boston Dynamics’ Atlas—has a battery, but it can only run for about an hour. That might be fine for some applications, but you don’t want it running out of juice halfway through rescuing someone from a mine shaft.
When it comes to the link between the robot and its human user, some of the technology is probably not that much of a stretch. Virtual reality headsets can create immersive audio-visual environments, and a number of companies are working on advanced haptic suits that will let people “feel” virtual environments.
Motion tracking technology may be more complicated. While even consumer-grade devices can track peoples’ movements with high accuracy, you will probably need to don something more like an exoskeleton that can both pick up motion and provide mechanical resistance, so that when the robot bumps into an immovable object, the user stops dead too.
How hard all of this will be is also dependent on how the competition ultimately defines subjective terms like “feel” and “interact.” Will the user need to be able to feel a gentle breeze on the robot’s cheek or be able to paint a watercolor? Or will simply having the ability to distinguish a hard object from a soft one or shake someone’s hand be enough?
Whatever the fidelity they decide on, the approach will require huge amounts of sensory and control data to be transmitted over large distances, most likely wirelessly, in a way that’s fast and reliable enough that there’s no lag or interruptions. Fortunately 5G is launching this year, with a speed of 10 gigabits per second and very low latency, so this problem should be solved by 2021.
And it’s worth remembering there have already been some tentative attempts at building robotic avatars. Telepresence robots have solved the seeing, hearing, and some of the interacting problems, and MIT has already used virtual reality to control robots to carry out complex manipulation tasks.
South Korean company Hankook Mirae Technology has also unveiled a 13-foot-tall robotic suit straight out of a sci-fi movie that appears to have made some headway with the motion tracking problem, albeit with a human inside the robot. Toyota’s T-HR3 does the same, but with the human controlling the robot from a “Master Maneuvering System” that marries motion tracking with VR.
Combining all of these capabilities into a single machine will certainly prove challenging. But if one of the teams pulls it off, you may be able to tick off trips to the Seven Wonders of the World without ever leaving your house.
Image Credit: ANA Avatar XPRIZE Continue reading