Tag Archives: based

#435707 AI Agents Startle Researchers With ...

Posted on September 17, 2019 by Android

After 25 million games, the AI agents playing hide-and-seek with each other had mastered four basic game strategies. The researchers expected that part.

After a total of 380 million games, the AI players developed strategies that the researchers didn’t know were possible in the game environment—which the researchers had themselves created. That was the part that surprised the team at OpenAI, a research company based in San Francisco.

The AI players learned everything via a machine learning technique known as reinforcement learning. In this learning method, AI agents start out by taking random actions. Sometimes those random actions produce desired results, which earn them rewards. Via trial-and-error on a massive scale, they can learn sophisticated strategies.

In the context of games, this process can be abetted by having the AI play against another version of itself, ensuring that the opponents will be evenly matched. It also locks the AI into a process of one-upmanship, where any new strategy that emerges forces the opponent to search for a countermeasure. Over time, this “self-play” amounted to what the researchers call an “auto-curriculum.”

According to OpenAI researcher Igor Mordatch, this experiment shows that self-play “is enough for the agents to learn surprising behaviors on their own—it’s like children playing with each other.”

Reinforcement is a hot field of AI research right now. OpenAI’s researchers used the technique when they trained a team of bots to play the video game Dota 2, which squashed a world-champion human team last April. The Alphabet subsidiary DeepMind has used it to triumph in the ancient board game Go and the video game StarCraft.

Aniruddha Kembhavi, a researcher at the Allen Institute for Artificial Intelligence (AI2) in Seattle, says games such as hide-and-seek offer a good way for AI agents to learn “foundational skills.” He worked on a team that taught their AllenAI to play Pictionary with humans, viewing the gameplay as a way for the AI to work on common sense reasoning and communication. “We are, however, quite far away from being able to translate these preliminary findings in highly simplified environments into the real world,” says Kembhavi.

Illustration: OpenAI

AI agents construct a fort during a hide-and-seek game developed by OpenAI.

In OpenAI’s game of hide-and-seek, both the hiders and the seekers received a reward only if they won the game, leaving the AI players to develop their own strategies. Within a simple 3D environment containing walls, blocks, and ramps, the players first learned to run around and chase each other (strategy 1). The hiders next learned to move the blocks around to build forts (2), and then the seekers learned to move the ramps (3), enabling them to jump inside the forts. Then the hiders learned to move all the ramps into their forts before the seekers could use them (4).

The two strategies that surprised the researchers came next. First the seekers learned that they could jump onto a box and “surf” it over to a fort (5), allowing them to jump in—a maneuver that the researchers hadn’t realized was physically possible in the game environment. So as a final countermeasure, the hiders learned to lock all the boxes into place (6) so they weren’t available for use as surfboards.

Illustration: OpenAI

An AI agent uses a nearby box to surf its way into a competitor’s fort.

In this circumstance, having AI agents behave in an unexpected way wasn’t a problem: They found different paths to their rewards, but didn’t cause any trouble. However, you can imagine situations in which the outcome would be rather serious. Robots acting in the real world could do real damage. And then there’s Nick Bostrom’s famous example of a paper clip factory run by an AI, whose goal is to make as many paper clips as possible. As Bostrom told IEEE Spectrum back in 2014, the AI might realize that “human bodies consist of atoms, and those atoms could be used to make some very nice paper clips.”

Bowen Baker, another member of the OpenAI research team, notes that it’s hard to predict all the ways an AI agent will act inside an environment—even a simple one. “Building these environments is hard,” he says. “The agents will come up with these unexpected behaviors, which will be a safety problem down the road when you put them in more complex environments.”

AI researcher Katja Hofmann at Microsoft Research Cambridge, in England, has seen a lot of gameplay by AI agents: She started a competition that uses Minecraft as the playing field. She says the emergent behavior seen in this game, and in prior experiments by other researchers, shows that games can be a useful for studies of safe and responsible AI.

“I find demonstrations like this, in games and game-like settings, a great way to explore the capabilities and limitations of existing approaches in a safe environment,” says Hofmann. “Results like these will help us develop a better understanding on how to validate and debug reinforcement learning systems–a crucial step on the path towards real-world applications.”

Baker says there’s also a hopeful takeaway from the surprises in the hide-and-seek experiment. “If you put these agents into a rich enough environment they will find strategies that we never knew were possible,” he says. “Maybe they can solve problems that we can’t imagine solutions to.” Continue reading →

Posted in Human Robots

#435703 FarmWise Raises $14.5 Million to Teach ...

Posted on September 17, 2019 by Android

We humans spend most of our time getting hungry or eating, which must be really inconvenient for the people who have to produce food for everyone. For a sustainable and tasty future, we’ll need to make the most of what we’ve got by growing more food with less effort, and that’s where the robots can help us out a little bit.

FarmWise, a California-based startup, is looking to enhance farming efficiency by automating everything from seeding to harvesting, starting with the worst task of all: weeding. And they’ve just raised US $14.5 million to do it.

FarmWise’s autonomous, AI-enabled robots are designed to solve farmers’ most pressing challenges by performing a variety of farming functions – starting with weeding, and providing personalized care to every plant they touch. Using machine learning models, computer vision and high-precision mechanical tools, FarmWise’s sophisticated robots cleanly pick weeds from fields, leaving crops with the best opportunity to thrive while eliminating harmful chemical inputs. To date, FarmWise’s robots have efficiently removed weeds from more than 10 million plants.

FarmWise is not the first company to work on large mobile farming robots. A few years ago, we wrote about DeepField Robotics and their giant weed-punching robot. But considering how many humans there are, and how often we tend to get hungry, it certainly seems like there’s plenty of opportunity to go around.

Photo: FarmWise

FarmWise is collecting massive amounts of data about every single plant in an entire field, which is something that hasn’t been possible before. Above, one of the robots at a farm in Salinas Valley, Calif.

Weeding is just one thing that farm robots are able to do. FarmWise is collecting massive amounts of data about every single plant in an entire field, practically on the per-leaf level, which is something that hasn’t been possible before. Data like this could be used for all sorts of things, but generally, the long-term hope is that robots could tend to every single plant individually—weeding them, fertilizing them, telling them what good plants they are, and then mercilessly yanking them out of the ground at absolute peak ripeness. It’s not realistic to do this with human labor, but it’s the sort of data-intensive and monotonous task that robots could be ideal for.

The question with robots like this is not necessarily whether they can do the job that they were created for, because generally, they can—farms are structured enough environments that they lend themselves to autonomous robots, and the tasks are relatively well defined. The issue right now, I think, is whether robots are really time- and cost-effective for farmers. Capable robots are an expensive investment, and even if there is a shortage of human labor, will robots perform well enough to convince farmers to adopt the technology? That’s a solid maybe, and here’s hoping that FarmWise can figure out how to make it work.

[ FarmWise ] Continue reading →

Posted in Human Robots

#435683 How High Fives Help Us Get in Touch With ...

Posted on September 16, 2019 by Android

The human sense of touch is so naturally ingrained in our everyday lives that we often don’t notice its presence. Even so, touch is a crucial sensing ability that helps people to understand the world and connect with others. As the market for robots grows, and as robots become more ingrained into our environments, people will expect robots to participate in a wide variety of social touch interactions. At Oregon State University’s Collaborative Robotics and Intelligent Systems (CoRIS) Institute, I research how to equip everyday robots with better social-physical interaction skills—from playful high-fives to challenging physical therapy routines.

Some commercial robots already possess certain physical interaction skills. For example, the videoconferencing feature of mobile telepresence robots can keep far-away family members connected with one another. These robots can also roam distant spaces and bump into people, chairs, and other remote objects. And my Roomba occasionally tickles my toes before turning to vacuum a different area of the room. As a human being, I naturally interpret this (and other Roomba behaviors) as social, even if they were not intended as such. At the same time, for both of these systems, social perceptions of the robots’ physical interaction behaviors are not well understood, and these social touch-like interactions cannot be controlled in nuanced ways.

Before joining CoRIS early this year, I was a postdoc at the University of Southern California’s Interaction Lab, and prior to that, I completed my doctoral work at the GRASP Laboratory’s Haptics Group at the University of Pennsylvania. My dissertation focused on improving the general understanding of how robot control and planning strategies influence perceptions of social touch interactions. As part of that research, I conducted a study of human-robot hand-to-hand contact, focusing on an interaction somewhere between a high five and a hand-clapping game. I decided to study this particular interaction because people often high five, and they will likely expect robots in everyday spaces to high five as well!

I conducted a study of human-robot hand-to-hand contact, focusing on an interaction somewhere between a high five and a hand-clapping game. I decided to study this particular interaction because people often high five, and they will likely expect robots to high five as well!

The implications of motion and planning on the social touch experience in these interactions is also crucial—think about a disappointingly wimpy (or triumphantly amazing) high five that you’ve experienced in the past. This great or terrible high-fiving experience could be fleeting, but it could also influence who you interact with, who you’re friends with, and even how you perceive the character or personalities of those around you. This type of perception, judgement, and response could extend to personal robots, too!

An investigation like this requires a mixture of more traditional robotics research (e.g., understanding how to move and control a robot arm, developing models of the desired robot motion) along with techniques from design and psychology (e.g., performing interviews with research participants, using best practices from experimental methods in perception). Enabling robots with social touch abilities also comes with many challenges, and even skilled humans can have trouble anticipating what another person is about to do. Think about trying to make satisfying hand contact during a high five—you might know the classic adage “watch the elbow,” but if you’re like me, even this may not always work.

I conducted a research study involving eight different types of human-robot hand contact, with different combinations of the following: interactions with a facially reactive or non-reactive robot, a physically reactive or non-reactive planning strategy, and a lower or higher robot arm stiffness. My robotic system could become facially reactive by changing its facial expression in response to hand contact, or physically reactive by updating its plan of where to move next after sensing hand contact. The stiffness of the robot could be adjusted by changing a variable that controlled how quickly the robot’s motors tried to pull its arm to the desired position. I knew from previous research that fine differences in touch interactions can have a big impact on perceived robot character. For example, if a robot grips an object too tightly or for too long while handing an object to a person, it might be perceived as greedy, possessive, or perhaps even Sméagol-like. A robot that lets go too soon might appear careless or sloppy.

In the example cases of robot grip, it’s clear that understanding people’s perceptions of robot characteristics and personality can help roboticists choose the right robot design based on the proposed operating environment of the robot. I likewise wanted to learn how the facial expressions, physical reactions, and stiffness of a hand-clapping robot would influence human perceptions of robot pleasantness, energeticness, dominance, and safety. Understanding this relationship can help roboticists to equip robots with personalities appropriate for the task at hand. For example, a robot assisting people in a grocery store may need to be designed with a high level of pleasantness and only moderate energy, while a maximally effective robot for comedy roast battles may need high degrees of energy and dominance above all else.

After many a late night at the GRASP Lab clapping hands with a big red robot, I was ready to conduct the study. Twenty participants visited the lab to clap hands with our Baxter Research Robot and help me begin to understand how characteristics of this humanoid robot’s social touch influenced its pleasantness, energeticness, dominance, and apparent safety. Baxter interacted with participants using a custom 3D-printed hand that was inlaid with silicone inserts.

The study showed that a facially reactive robot seemed more pleasant and energetic. A physically reactive robot seemed less pleasant, energetic, and dominant for this particular study design and interaction. I thought contact with a stiffer robot would seem harder (and therefore more dominant and less safe), but counter to my expectations, a stiffer-armed robot seemed safer and less dominant to participants. This may be because the stiffer robot was more precise in following its pre-programmed trajectory, therefore seeming more predictable and less free-spirited.

Safety ratings of the robot were generally high, and several participants commented positively on the robot’s facial expressions. Some participants attributed inventive (and non-existent) intelligences to the robot—I used neither computer vision nor the Baxter robot’s cameras in this study, but more than one participant complimented me on how well the robot tracked their hand position. While interacting with the robot, participants displayed happy facial expressions more than any other analyzed type of expression.

Photo: Naomi Fitter

Participants were asked to clap hands with Baxter and describe how they perceived the robot in terms of its pleasantness, energeticness, dominance, and apparent safety.

Circling back to the idea of how people might interpret even rudimentary and practical robot behaviors as social, these results show that this type of social perception isn’t just true for my lovable (but sometimes dopey) Roomba, but also for collaborative industrial robots, and generally, any robot capable of physical human-robot interaction. In designing the motion of Baxter, the adjustment of a single number in the equation that controls joint stiffness can flip the robot from seeming safe and docile to brash and commanding. These implications are sometimes predictable, but often unexpected.

The results of this particular study give us a partial guide to manipulating the emotional experience of robot users by adjusting aspects of robot control and planning, but future work is needed to fully understand the design space of social touch. Will materials play a major role? How about personalized machine learning? Do results generalize over all robot arms, or even a specialized subset like collaborative industrial robot arms? I’m planning to continue answering these questions, and when I finally solve human-robot social touch, I’ll high five all my robots to celebrate.

Naomi Fitter is an assistant professor in the Collaborative Robotics and Intelligent Systems (CoRIS) Institute at Oregon State University, where her Social Haptics, Assistive Robotics, and Embodiment (SHARE) research group aims to equip robots with the ability to engage and empower people in interactions from playful high-fives to challenging physical therapy routines. She completed her doctoral work in the GRASP Laboratory’s Haptics Group and was a postdoctoral scholar in the University of Southern California’s Interaction Lab from 2017 to 2018. Naomi’s not-so-secret pastime is performing stand-up and improv comedy. Continue reading →

Posted in Human Robots

#435676 Intel’s Neuromorphic System Hits 8 ...

Posted on September 16, 2019 by Android

At the DARPA Electronics Resurgence Initiative Summit today in Detroit, Intel plans to unveil an 8-million-neuron neuromorphic system comprising 64 Loihi research chips—codenamed Pohoiki Beach. Loihi chips are built with an architecture that more closely matches the way the brain works than do chips designed to do deep learning or other forms of AI. For the set of problems that such “spiking neural networks” are particularly good at, Loihi is about 1,000 times as fast as a CPU and 10,000 times as energy efficient. The new 64-Loihi system represents the equivalent of 8-million neurons, but that’s just a step to a 768-chip, 100-million-neuron system that the company plans for the end of 2019.

Intel and its research partners are just beginning to test what massive neural systems like Pohoiki Beach can do, but so far the evidence points to even greater performance and efficiency, says Mike Davies, director of neuromorphic research at Intel.

“We’re quickly accumulating results and data that there are definite benefits… mostly in the domain of efficiency. Virtually every one that we benchmark…we find significant gains in this architecture,” he says.

Going from a single-Loihi to 64 of them is more of a software issue than a hardware one. “We designed scalability into the Loihi chip from the beginning,” says Davies. “The chip has a hierarchical routing interface…which allows us to scale to up to 16,000 chips. So 64 is just the next step.”

Photo: Tim Herman/Intel Corporation

One of Intel’s Nahuku boards, each of which contains 8 to 32 Intel Loihi neuromorphic chips, shown here interfaced to an Intel Arria 10 FPGA development kit. Intel’s latest neuromorphic system, Pohoiki Beach, is made up of multiple Nahuku boards and contains 64 Loihi chips.

Finding algorithms that run well on an 8-million-neuron system and optimizing those algorithms in software is a considerable effort, he says. Still, the payoff could be huge. Neural networks that are more brain-like, such as Loihi, could be immune to some of the artificial intelligence’s—for lack of a better word—dumbness.

For example, today’s neural networks suffer from something called catastrophic forgetting. If you tried to teach a trained neural network to recognize something new—a new road sign, say—by simply exposing the network to the new input, it would disrupt the network so badly that it would become terrible at recognizing anything. To avoid this, you have to completely retrain the network from the ground up. (DARPA’s Lifelong Learning, or L2M, program is dedicated to solving this problem.)

(Here’s my favorite analogy: Say you coached a basketball team, and you raised the net by 30 centimeters while nobody was looking. The players would miss a bunch at first, but they’d figure things out quickly. If those players were like today’s neural networks, you’d have to pull them off the court and teach them the entire game over again—dribbling, passing, everything.)

Loihi can run networks that might be immune to catastrophic forgetting, meaning it learns a bit more like a human. In fact, there’s evidence through a research collaboration with Thomas Cleland’s group at Cornell University, that Loihi can achieve what’s called one-shot learning. That is, learning a new feature after being exposed to it only once. The Cornell group showed this by abstracting a model of the olfactory system so that it would run on Loihi. When exposed to a new virtual scent, the system not only didn't catastrophically forget everything else it had smelled, it learned to recognize the new scent just from the single exposure.

Loihi might also be able to run feature-extraction algorithms that are immune to the kinds of adversarial attacks that befuddle today’s image recognition systems. Traditional neural networks don’t really understand the features they’re extracting from an image in the way our brains do. “They can be fooled with simplistic attacks like changing individual pixels or adding a screen of noise that wouldn’t fool a human in any way,” Davies explains. But the sparse-coding algorithms Loihi can run work more like the human visual system and so wouldn’t fall for such shenanigans. (Disturbingly, humans are not completely immune to such attacks.)

Photo: Tim Herman/Intel Corporation

A close-up shot of Loihi, Intel’s neuromorphic research chip. Intel’s latest neuromorphic system, Pohoiki Beach, will be comprised of 64 of these Loihi chips.

Researchers have also been using Loihi to improve real-time control for robotic systems. For example, last week at the Telluride Neuromorphic Cognition Engineering Workshop—an event Davies called “summer camp for neuromorphics nerds”—researchers were hard at work using a Loihi-based system to control a foosball table. “It strikes people as crazy,” he says. “But it’s a nice illustration of neuromorphic technology. It’s fast, requires quick response, quick planning, and anticipation. These are what neuromorphic chips are good at.” Continue reading →

Posted in Human Robots

#435674 MIT Future of Work Report: We ...

Posted on September 16, 2019 by Android

Robots aren’t going to take everyone’s jobs, but technology has already reshaped the world of work in ways that are creating clear winners and losers. And it will continue to do so without intervention, says the first report of MIT’s Task Force on the Work of the Future.

The supergroup of MIT academics was set up by MIT President Rafael Reif in early 2018 to investigate how emerging technologies will impact employment and devise strategies to steer developments in a positive direction. And the headline finding from their first publication is that it’s not the quantity of jobs we should be worried about, but the quality.

Widespread press reports of a looming “employment apocalypse” brought on by AI and automation are probably wide of the mark, according to the authors. Shrinking workforces as developed countries age and outstanding limitations in what machines can do mean we’re unlikely to have a shortage of jobs.

But while unemployment is historically low, recent decades have seen a polarization of the workforce as the number of both high- and low-skilled jobs have grown at the expense of the middle-skilled ones, driving growing income inequality and depriving the non-college-educated of viable careers.

This is at least partly attributable to the growth of digital technology and automation, the report notes, which are rendering obsolete many middle-skilled jobs based around routine work like assembly lines and administrative support.

That leaves workers to either pursue high-skilled jobs that require deep knowledge and creativity, or settle for low-paid jobs that rely on skills—like manual dexterity or interpersonal communication—that are still beyond machines, but generic to most humans and therefore not valued by employers. And the growth of emerging technology like AI and robotics is only likely to exacerbate the problem.

This isn’t the first report to note this trend. The World Bank’s 2016 World Development Report noted how technology is causing a “hollowing out” of labor markets. But the MIT report goes further in saying that the cause isn’t simply technology, but the institutions and policies we’ve built around it.

The motivation for introducing new technology is broadly assumed to be to increase productivity, but the authors note a rarely-acknowledged fact: “Not all innovations that raise productivity displace workers, and not all innovations that displace workers substantially raise productivity.”

Examples of the former include computer-aided design software that makes engineers and architects more productive, while examples of the latter include self-service checkouts and automated customer support that replace human workers, often at the expense of a worse customer experience.

While the report notes that companies have increasingly adopted the language of technology augmenting labor, in reality this has only really benefited high-skilled workers. For lower-skilled jobs the motivation is primarily labor cost savings, which highlights the other major force shaping technology’s impact on employment: shareholder capitalism.

The authors note that up until the 1980s, increasing productivity resulted in wage growth across the economic spectrum, but since then average wage growth has failed to keep pace and gains have dramatically skewed towards the top earners.

The report shies away from directly linking this trend to the birth of Reaganomics (something others have been happy to do), but it notes that American veneration of the shareholder as the primary stakeholder in a business and tax policies that incentivize investment in capital rather than labor have exacerbated the negative impacts technology can have on employment.

That means the current focus on re-skilling workers to thrive in the new economy is a necessary, but not sufficient, solution to the disruptive impact technology is having on work, the authors say.

Alongside significant investment in education, fiscal policies need to be re-balanced away from subsidizing investment in physical capital and towards boosting investment in human capital, the authors write, and workers need to have a greater say in corporate decision-making.

The authors point to other developed economies where productivity growth, income growth, and equality haven’t become so disconnected thanks to investments in worker skills, social safety nets, and incentives to invest in human capital. Whether such a radical reshaping of US economic policy is achievable in today’s political climate remains to be seen, but the authors conclude with a call to arms.

“The failure of the US labor market to deliver broadly shared prosperity despite rising productivity is not an inevitable byproduct of current technologies or free markets,” they write. “We can and should do better.”

Image Credit: Simon Abrams / Unsplash/a> Continue reading →

Posted in Human Robots