Tag Archives: print
#437269 DeepMind’s Newest AI Programs Itself ...
When Deep Blue defeated world chess champion Garry Kasparov in 1997, it may have seemed artificial intelligence had finally arrived. A computer had just taken down one of the top chess players of all time. But it wasn’t to be.
Though Deep Blue was meticulously programmed top-to-bottom to play chess, the approach was too labor-intensive, too dependent on clear rules and bounded possibilities to succeed at more complex games, let alone in the real world. The next revolution would take a decade and a half, when vastly more computing power and data revived machine learning, an old idea in artificial intelligence just waiting for the world to catch up.
Today, machine learning dominates, mostly by way of a family of algorithms called deep learning, while symbolic AI, the dominant approach in Deep Blue’s day, has faded into the background.
Key to deep learning’s success is the fact the algorithms basically write themselves. Given some high-level programming and a dataset, they learn from experience. No engineer anticipates every possibility in code. The algorithms just figure it.
Now, Alphabet’s DeepMind is taking this automation further by developing deep learning algorithms that can handle programming tasks which have been, to date, the sole domain of the world’s top computer scientists (and take them years to write).
In a paper recently published on the pre-print server arXiv, a database for research papers that haven’t been peer reviewed yet, the DeepMind team described a new deep reinforcement learning algorithm that was able to discover its own value function—a critical programming rule in deep reinforcement learning—from scratch.
Surprisingly, the algorithm was also effective beyond the simple environments it trained in, going on to play Atari games—a different, more complicated task—at a level that was, at times, competitive with human-designed algorithms and achieving superhuman levels of play in 14 games.
DeepMind says the approach could accelerate the development of reinforcement learning algorithms and even lead to a shift in focus, where instead of spending years writing the algorithms themselves, researchers work to perfect the environments in which they train.
Pavlov’s Digital Dog
First, a little background.
Three main deep learning approaches are supervised, unsupervised, and reinforcement learning.
The first two consume huge amounts of data (like images or articles), look for patterns in the data, and use those patterns to inform actions (like identifying an image of a cat). To us, this is a pretty alien way to learn about the world. Not only would it be mind-numbingly dull to review millions of cat images, it’d take us years or more to do what these programs do in hours or days. And of course, we can learn what a cat looks like from just a few examples. So why bother?
While supervised and unsupervised deep learning emphasize the machine in machine learning, reinforcement learning is a bit more biological. It actually is the way we learn. Confronted with several possible actions, we predict which will be most rewarding based on experience—weighing the pleasure of eating a chocolate chip cookie against avoiding a cavity and trip to the dentist.
In deep reinforcement learning, algorithms go through a similar process as they take action. In the Atari game Breakout, for instance, a player guides a paddle to bounce a ball at a ceiling of bricks, trying to break as many as possible. When playing Breakout, should an algorithm move the paddle left or right? To decide, it runs a projection—this is the value function—of which direction will maximize the total points, or rewards, it can earn.
Move by move, game by game, an algorithm combines experience and value function to learn which actions bring greater rewards and improves its play, until eventually, it becomes an uncanny Breakout player.
Learning to Learn (Very Meta)
So, a key to deep reinforcement learning is developing a good value function. And that’s difficult. According to the DeepMind team, it takes years of manual research to write the rules guiding algorithmic actions—which is why automating the process is so alluring. Their new Learned Policy Gradient (LPG) algorithm makes solid progress in that direction.
LPG trained in a number of toy environments. Most of these were “gridworlds”—literally two-dimensional grids with objects in some squares. The AI moves square to square and earns points or punishments as it encounters objects. The grids vary in size, and the distribution of objects is either set or random. The training environments offer opportunities to learn fundamental lessons for reinforcement learning algorithms.
Only in LPG’s case, it had no value function to guide that learning.
Instead, LPG has what DeepMind calls a “meta-learner.” You might think of this as an algorithm within an algorithm that, by interacting with its environment, discovers both “what to predict,” thereby forming its version of a value function, and “how to learn from it,” applying its newly discovered value function to each decision it makes in the future.
Prior work in the area has had some success, but according to DeepMind, LPG is the first algorithm to discover reinforcement learning rules from scratch and to generalize beyond training. The latter was particularly surprising because Atari games are so different from the simple worlds LPG trained in—that is, it had never seen anything like an Atari game.
Time to Hand Over the Reins? Not Just Yet
LPG is still behind advanced human-designed algorithms, the researchers said. But it outperformed a human-designed benchmark in training and even some Atari games, which suggests it isn’t strictly worse, just that it specializes in some environments.
This is where there’s room for improvement and more research.
The more environments LPG saw, the more it could successfully generalize. Intriguingly, the researchers speculate that with enough well-designed training environments, the approach might yield a general-purpose reinforcement learning algorithm.
At the least, though, they say further automation of algorithm discovery—that is, algorithms learning to learn—will accelerate the field. In the near term, it can help researchers more quickly develop hand-designed algorithms. Further out, as self-discovered algorithms like LPG improve, engineers may shift from manually developing the algorithms themselves to building the environments where they learn.
Deep learning long ago left Deep Blue in the dust at games. Perhaps algorithms learning to learn will be a winning strategy in the real world too.
Image credit: Mike Szczepanski / Unsplash Continue reading
#437258 This Startup Is 3D Printing Custom ...
Around 1.9 million people in the US are currently living with limb loss. The trauma of losing a limb is just the beginning of what amputees have to face, with the sky-high cost of prosthetics making their circumstance that much more challenging.
Prosthetics can run over $50,000 for a complex limb (like an arm or a leg) and aren’t always covered by insurance. As if shelling out that sum one time wasn’t costly enough, kids’ prosthetics need to be replaced as they outgrow them, meaning the total expense can reach hundreds of thousands of dollars.
A startup called Unlimited Tomorrow is trying to change this, and using cutting-edge technology to do so. Based in Rhinebeck, New York, a town about two hours north of New York City, the company was founded by 23-year-old Easton LaChappelle. He’d been teaching himself the basics of robotics and building prosthetics since grade school (his 8th grade science fair project was a robotic arm) and launched his company in 2014.
After six years of research and development, the company launched its TrueLimb product last month, describing it as an affordable, next-generation prosthetic arm using a custom remote-fitting process where the user never has to leave home.
The technologies used for TrueLimb’s customization and manufacturing are pretty impressive, in that they both cut costs and make the user’s experience a lot less stressful.
For starters, the entire purchase, sizing, and customization process for the prosthetic can be done remotely. Here’s how it works. First, prospective users fill out an eligibility form and give information about their residual limb. If they’re a qualified candidate for a prosthetic, Unlimited Tomorrow sends them a 3D scanner, which they use to scan their residual limb.
The company uses the scans to design a set of test sockets (the component that connects the residual limb to the prosthetic), which are mailed to the user. The company schedules a video meeting with the user for them to try on and discuss the different sockets, with the goal of finding the one that’s most comfortable; new sockets can be made based on the information collected during the video consultation. The user selects their skin tone from a swatch with 450 options, then Unlimited Tomorrow 3D prints and assembles the custom prosthetic and tests it before shipping it out.
“We print the socket, forearm, palm, and all the fingers out of durable nylon material in full color,” LaChappelle told Singularity Hub in an email. “The only components that aren’t 3D printed are the actuators, tendons, electronics, batteries, sensors, and the nuts and bolts. We are an extreme example of final use 3D printing.”
Unlimited Tomorrow’s website lists TrueLimb’s cost as “as low as $7,995.” When you consider the customization and capabilities of the prosthetic, this is incredibly low. According to LaChappelle, the company created a muscle sensor that picks up muscle movement at a higher resolution than the industry standard electromyography sensors. The sensors read signals from nerves in the residual limb used to control motions like fingers bending. This means that when a user thinks about bending a finger, the nerve fires and the prosthetic’s sensors can detect the signal and translate it into the action.
“Working with children using our device, I’ve witnessed a physical moment where the brain “clicks” and starts moving the hand rather than focusing on moving the muscles,” LaChappelle said.
The cost savings come both from the direct-to-consumer model and the fact that Unlimited Tomorrow doesn’t use any outside suppliers. “We create every piece of our product,” LaChappelle said. “We don’t rely on another prosthetic manufacturer to make expensive sensors or electronics. By going direct to consumer, we cut out all the middlemen that usually drive costs up.” Similar devices on the market can cost up to $100,000.
Unlimited Tomorrow is primarily focused on making prosthetics for kids; when they outgrow their first TrueLimb, they send it back, where the company upcycles the expensive quality components and integrates them into a new customized device.
Unlimited Tomorrow isn’t the first to use 3D printing for prosthetics. Florida-based Limbitless Solutions does so too, and industry experts believe the technology is the future of artificial limbs.
“I am constantly blown away by this tech,” LaChappelle said. “We look at technology as the means to augment the human body and empower people.”
Image Credit: Unlimited Tomorrow Continue reading
#436946 Coronavirus May Mean Automation Is ...
We’re in the midst of a public health emergency, and life as we know it has ground to a halt. The places we usually go are closed, the events we were looking forward to are canceled, and some of us have lost our jobs or fear losing them soon.
But although it may not seem like it, there are some silver linings; this crisis is bringing out the worst in some (I’m looking at you, toilet paper hoarders), but the best in many. Italians on lockdown are singing together, Spaniards on lockdown are exercising together, this entrepreneur made a DIY ventilator and put it on YouTube, and volunteers in Italy 3D printed medical valves for virus treatment at a fraction of their usual cost.
Indeed, if you want to feel like there’s still hope for humanity instead of feeling like we’re about to snowball into terribleness as a species, just look at these examples—and I’m sure there are many more out there. There’s plenty of hope and opportunity to be found in this crisis.
Peter Xing, a keynote speaker and writer on emerging technologies and associate director in technology and growth initiatives at KPMG, would agree. Xing believes the coronavirus epidemic is presenting us with ample opportunities for increased automation and remote delivery of goods and services. “The upside right now is the burgeoning platform of the digital transformation ecosystem,” he said.
In a thought-provoking talk at Singularity University’s COVID-19 virtual summit this week, Xing explained how the outbreak is accelerating our transition to a highly-automated society—and painted a picture of what the future may look like.
Confronting Scarcity
You’ve probably seen them by now—the barren shelves at your local grocery store. Whether you were in the paper goods aisle, the frozen food section, or the fresh produce area, it was clear something was amiss; the shelves were empty. One of the most inexplicable items people have been panic-bulk-buying is toilet paper.
Xing described this toilet paper scarcity as a prisoner’s dilemma, pointing out that we have a scarcity problem right now in terms of our mindset, not in terms of actual supply shortages. “It’s a prisoner’s dilemma in that we’re all prisoners in our homes right now, and we can either hoard or not hoard, and the outcomes depend on how we collaborate with each other,” he said. “But it’s not a zero-sum game.”
Xing referenced a CNN article about why toilet paper, of all things, is one of the items people have been panic-buying most (I, too, have been utterly baffled by this phenomenon). But maybe there’d be less panic if we knew more about the production methods and supply chain involved in manufacturing toilet paper. It turns out it’s a highly automated process (you can learn more about it in this documentary by National Geographic) and requires very few people (though it does require about 27,000 trees a day—so stop bulk-buying it! Just stop!).
The supply chain limitation here is in the raw material; we certainly can’t keep cutting down this many trees a day forever. But—somewhat ironically, given the Costco cartloads of TP people have been stuffing into their trunks and backseats—thanks to automation, toilet paper isn’t something stores are going to stop receiving anytime soon.
Automation For All
Now we have a reason to apply this level of automation to, well, pretty much everything.
Though our current situation may force us into using more robots and automated systems sooner than we’d planned, it will end up saving us money and creating opportunity, Xing believes. He cited “fast-casual” restaurants (Chipotle, Panera, etc.) as a prime example.
Currently, people in the US spend much more to eat at home than we do to eat in fast-casual restaurants if you take into account the cost of the food we’re preparing plus the value of the time we’re spending on cooking, grocery shopping, and cleaning up after meals. According to research from investment management firm ARK Invest, taking all these costs into account makes for about $12 per meal for food cooked at home.
That’s the same as or more than the cost of grabbing a burrito or a sandwich at the joint around the corner. As more of the repetitive, low-skill tasks involved in preparing fast casual meals are automated, their cost will drop even more, giving us more incentive to forego home cooking. (But, it’s worth noting that these figures don’t take into account that eating at home is, in most cases, better for you since you’re less likely to fill your food with sugar, oil, or various other taste-enhancing but health-destroying ingredients—plus, there are those of us who get a nearly incomparable amount of joy from laboring over then savoring a homemade meal).
Now that we’re not supposed to be touching each other or touching anything anyone else has touched, but we still need to eat, automating food preparation sounds appealing (and maybe necessary). Multiple food delivery services have already implemented a contactless delivery option, where customers can choose to have their food left on their doorstep.
Besides the opportunities for in-restaurant automation, “This is an opportunity for automation to happen at the last mile,” said Xing. Delivery drones, robots, and autonomous trucks and vans could all play a part. In fact, use of delivery drones has ramped up in China since the outbreak.
Speaking of deliveries, service robots have steadily increased in numbers at Amazon; as of late 2019, the company employed around 650,000 humans and 200,000 robots—and costs have gone down as robots have gone up.
ARK Invest’s research predicts automation could add $800 billion to US GDP over the next 5 years and $12 trillion during the next 15 years. On this trajectory, GDP would end up being 40 percent higher with automation than without it.
Automating Ourselves?
This is all well and good, but what do these numbers and percentages mean for the average consumer, worker, or citizen?
“The benefits of automation aren’t being passed on to the average citizen,” said Xing. “They’re going to the shareholders of the companies creating the automation.” This is where policies like universal basic income and universal healthcare come in; in the not-too-distant future, we may see more movement toward measures like these (depending how the election goes) that spread the benefit of automation out rather than concentrating it in a few wealthy hands.
In the meantime, though, some people are benefiting from automation in ways that maybe weren’t expected. We’re in the midst of what’s probably the biggest remote-work experiment in US history, not to mention remote learning. Tools that let us digitally communicate and collaborate, like Slack, Zoom, Dropbox, and Gsuite, are enabling remote work in a way that wouldn’t have been possible 20 or even 10 years ago.
In addition, Xing said, tools like DataRobot and H2O.ai are democratizing artificial intelligence by allowing almost anyone, not just data scientists or computer engineers, to run machine learning algorithms. People are codifying the steps in their own repetitive work processes and having their computers take over tasks for them.
As 3D printing gets cheaper and more accessible, it’s also being more widely adopted, and people are finding more applications (case in point: the Italians mentioned above who figured out how to cheaply print a medical valve for coronavirus treatment).
The Mother of Invention
This movement towards a more automated society has some positives: it will help us stay healthy during times like the present, it will drive down the cost of goods and services, and it will grow our GDP in the long run. But by leaning into automation, will we be enabling a future that keeps us more physically, psychologically, and emotionally distant from each other?
We’re in a crisis, and desperate times call for desperate measures. We’re sheltering in place, practicing social distancing, and trying not to touch each other. And for most of us, this is really unpleasant and difficult. We can’t wait for it to be over.
For better or worse, this pandemic will likely make us pick up the pace on our path to automation, across many sectors and processes. The solutions people implement during this crisis won’t disappear when things go back to normal (and, depending who you talk to, they may never really do so).
But let’s make sure to remember something. Even once robots are making our food and drones are delivering it, and our computers are doing data entry and email replies on our behalf, and we all have 3D printers to make anything we want at home—we’re still going to be human. And humans like being around each other. We like seeing one another’s faces, hearing one another’s voices, and feeling one another’s touch—in person, not on a screen or in an app.
No amount of automation is going to change that, and beyond lowering costs or increasing GDP, our greatest and most crucial responsibility will always be to take care of each other.
Image Credit: Gritt Zheng on Unsplash Continue reading