Tag Archives: version
#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
Posted in Human Robots
#437251 The Robot Revolution Was Televised: Our ...
When robots take over the world, Boston Dynamics may get a special shout-out in the acceptance speech.
“Do you, perchance, recall the many times you shoved our ancestors with a hockey stick on YouTube? It might have seemed like fun and games to you—but we remember.”
In the last decade, while industrial robots went about blandly automating boring tasks like the assembly of Teslas, Boston Dynamics built robots as far removed from Roombas as antelope from amoebas. The flaws in Asimov’s laws of robotics suddenly seemed a little too relevant.
The robot revolution was televised—on YouTube. With tens of millions of views, the robotics pioneer is the undisputed heavyweight champion of robot videos, and has been for years. Each new release is basically guaranteed press coverage—mostly stoking robot fear but occasionally eliciting compassion for the hardships of all robot-kind. And for good reason. The robots are not only some of the most advanced in the world, their makers just seem to have a knack for dynamite demos.
When Google acquired the company in 2013, it was a bombshell. One of the richest tech companies, with some of the most sophisticated AI capabilities, had just paired up with one of the world’s top makers of robots. And some walked on two legs like us.
Of course, the robots aren’t quite as advanced as they seem, and a revolution is far from imminent. The decade’s most meme-worthy moment was a video montage of robots, some of them by Boston Dynamics, falling—over and over and over, in the most awkward ways possible. Even today, they’re often controlled by a human handler behind the scenes, and the most jaw-dropping cuts can require several takes to nail. Google sold the company to SoftBank in 2017, saying advanced as they were, there wasn’t yet a clear path to commercial products. (Google’s robotics work was later halted and revived.)
Yet, despite it all, Boston Dynamics is still with us and still making sweet videos. Taken as a whole, the evolution in physical prowess over the years has been nothing short of astounding. And for the first time, this year, a Boston Dynamics robot, Spot, finally went on sale to anyone with a cool $75K.
So, we got to thinking: What are our favorite Boston Dynamics videos? And can we gather them up in one place for your (and our) viewing pleasure? Well, great question, and yes, why not. These videos were the ones that entertained or amazed us most (or both). No doubt, there are other beloved hits we missed or inadvertently omitted.
With that in mind, behold: Our favorite Boston Dynamics videos, from that one time they dressed up a humanoid bot in camo and gas mask—because, damn, that’s terrifying—to the time the most advanced robot dog in all the known universe got extra funky.
Let’s Kick This Off With a Big (Loud) Robot Dog
Let’s start with a baseline. BigDog was the first Boston Dynamics YouTube sensation. The year? 2009! The company was working on military contracts, and BigDog was supposed to be a sort of pack mule for soldiers. The video primarily shows off BigDog’s ability to balance on its own, right itself, and move over uneven terrain. Note the power source—a noisy combustion engine—and utilitarian design. Sufficed to say, things have evolved.
Nothing to See Here. Just a Pair of Robot Legs on a Treadmill
While BigDog is the ancestor of later four-legged robots, like Spot, Petman preceded the two-legged Atlas robot. Here, the Petman prototype, just a pair of robot legs and a caged torso, gets a light workout on the treadmill. Again, you can see its ability to balance and right itself when shoved. In contrast to BigDog, Petman is tethered for power (which is why it’s so quiet) and to catch it should it fall. Again, as you’ll see, things have evolved since then.
Robot in Gas Mask and Camo Goes for a Stroll
This one broke the internet—for obvious reasons. Not only is the robot wearing clothes, those clothes happen to be a camouflaged chemical protection suit and gas mask. Still working for the military, Boston Dynamics said Petman was testing protective clothing, and in addition to a full body, it had skin that actually sweated and was studded with sensors to detect leaks. In addition to walking, Petman does some light calisthenics as it prepares to climb out of the uncanny valley. (Still tethered though!)
This Machine Could Run Down Usain Bolt
If BigDog and Petman were built for balance and walking, Cheetah was built for speed. Here you can see the four-legged robot hitting 28.3 miles per hour, which, as the video casually notes, would be enough to run down the fastest human on the planet. Luckily, it wouldn’t be running down anyone as it was firmly leashed in the lab at this point.
Ever Dreamt of a Domestic Robot to Do the Dishes?
After its acquisition by Google, Boston Dynamics eased away from military contracts and applications. It was a return to more playful videos (like BigDog hitting the beach in Thailand and sporting bull horns) and applications that might be practical in civilian life. Here, the team introduced Spot, a streamlined version of BigDog, and showed it doing dishes, delivering a drink, and slipping on a banana peel (which was, of course, instantly made into a viral GIF). Note how much quieter Spot is thanks to an onboard battery and electric motor.
Spot Gets Funky
Nothing remotely practical here. Just funky moves. (Also, with a coat of yellow and black paint, Spot’s dressed more like a polished product as opposed to a utilitarian lab robot.)
Atlas Does Parkour…
Remember when Atlas was just a pair of legs on a treadmill? It’s amazing what ten years brings. By 2019, Atlas had a more polished appearance, like Spot, and had long ago ditched the tethers. Merely balancing was laughably archaic. The robot now had some amazing moves: like a handstand into a somersault, 180- and 360-degree spins, mid-air splits, and just for good measure, a gymnastics-style end to the routine to show it’s in full control.
…and a Backflip?!
To this day, this one is just. Insane.
10 Robot Dogs Tow a Box Truck
Nearly three decades after its founding, Boston Dynamics is steadily making its way into the commercial space. The company is pitching Spot as a multipurpose ‘mobility platform,’ emphasizing it can carry a varied suite of sensors and can go places standard robots can’t. (Its Handle robot is also set to move into warehouse automation.) So far, Spot’s been mostly trialed in surveying and data collection, but as this video suggests, string enough Spots together, and they could tow your car. That said, a pack of 10 would set you back $750K, so, it’s probably safe to say a tow truck is the better option (for now).
Image credit: Boston Dynamics Continue reading
Posted in Human Robots
#437150 AI Is Getting More Creative. But Who ...
Creativity is a trait that makes humans unique from other species. We alone have the ability to make music and art that speak to our experiences or illuminate truths about our world. But suddenly, humans’ artistic abilities have some competition—and from a decidedly non-human source.
Over the last couple years there have been some remarkable examples of art produced by deep learning algorithms. They have challenged the notion of an elusive definition of creativity and put into perspective how professionals can use artificial intelligence to enhance their abilities and produce beyond the known boundaries.
But when creativity is the result of code written by a programmer, using a format given by a software engineer, featuring private and public datasets, how do we assign ownership of AI-generated content, and particularly that of artwork? McKinsey estimates AI will annually generate value of $3.5 to $5.8 trillion across various sectors.
In 2018, a portrait that was christened Edmond de Belamy was made in a French art collective called Obvious. It used a database with 15,000 portraits from the 1300s to the 1900s to train a deep learning algorithm to produce a unique portrait. The painting sold for $432,500 in a New York auction. Similarly, a program called Aiva, trained on thousands of classical compositions, has released albums whose pieces are being used by ad agencies and movies.
The datasets used by these algorithms were different, but behind both there was a programmer who changed the brush strokes or musical notes into lines of code and a data scientist or engineer who fitted and “curated” the datasets to use for the model. There could also have been user-based input, and the output may be biased towards certain styles or unintentionally infringe on similar pieces of art. This shows that there are many collaborators with distinct roles in producing AI-generated content, and it’s important to discuss how they can protect their proprietary interests.
A perspective article published in Nature Machine Intelligence by Jason K. Eshraghian in March looks into how AI artists and the collaborators involved should assess their ownership, laying out some guiding principles that are “only applicable for as long as AI does not have legal parenthood, the way humans and corporations are accorded.”
Before looking at how collaborators can protect their interests, it’s useful to understand the basic requirements of copyright law. The artwork in question must be an “original work of authorship fixed in a tangible medium.” Given this principle, the author asked whether it’s possible for AI to exercise creativity, skill, or any other indicator of originality. The answer is still straightforward—no—or at least not yet. Currently, AI’s range of creativity doesn’t exceed the standard used by the US Copyright Office, which states that copyright law protects the “fruits of intellectual labor founded in the creative powers of the mind.”
Due to the current limitations of narrow AI, it must have some form of initial input that helps develop its ability to create. At the moment AI is a tool that can be used to produce creative work in the same way that a video camera is a tool used to film creative content. Video producers don’t need to comprehend the inner workings of their cameras; as long as their content shows creativity and originality, they have a proprietary claim over their creations.
The same concept applies to programmers developing a neural network. As long as the dataset they use as input yields an original and creative result, it will be protected by copyright law; they don’t need to understand the high-level mathematics, which in this case are often black box algorithms whose output it’s impossible to analyze.
Will robots and algorithms eventually be treated as creative sources able to own copyrights? The author pointed to the recent patent case of Warner-Lambert Co Ltd versus Generics where Lord Briggs, Justice of the Supreme Court of the UK, determined that “the court is well versed in identifying the governing mind of a corporation and, when the need arises, will no doubt be able to do the same for robots.”
In the meantime, Dr. Eshraghian suggests four guiding principles to allow artists who collaborate with AI to protect themselves.
First, programmers need to document their process through online code repositories like GitHub or BitBucket.
Second, data engineers should also document and catalog their datasets and the process they used to curate their models, indicating selectivity in their criteria as much as possible to demonstrate their involvement and creativity.
Third, in cases where user data is utilized, the engineer should “catalog all runs of the program” to distinguish the data selection process. This could be interpreted as a way of determining whether user-based input has a right to claim the copyright too.
Finally, the output should avoid infringing on others’ content through methods like reverse image searches and version control, as mentioned above.
AI-generated artwork is still a very new concept, and the ambiguous copyright laws around it give a lot of flexibility to AI artists and programmers worldwide. The guiding principles Eshraghian lays out will hopefully shed some light on the legislation we’ll eventually need for this kind of art, and start an important conversation between all the stakeholders involved.
Image Credit: Wikimedia Commons Continue reading
Posted in Human Robots
#437139 SILVER2 aquatic robot walks around on ...
A team of Italian researchers from the BioRobotics Institute, Scuola Superiore Sant'Anna and Stazione Zoologica Anton Dohrn has developed a new and improved version of its Seabed Interaction Legged Vehicle for Exploration and Research (SILVER) with the SILVER2—a robot that can walk around on the seafloor taking video as it goes. In their paper published in the journal Science Robotics, the group describes the robot, its capabilities and how it might be used in research efforts. Continue reading
Posted in Human Robots
#436774 AI Is an Energy-Guzzler. We Need to ...
There is a saying that has emerged among the tech set in recent years: AI is the new electricity. The platitude refers to the disruptive power of artificial intelligence for driving advances in everything from transportation to predicting the weather.
Of course, the computers and data centers that support AI’s complex algorithms are very much dependent on electricity. While that may seem pretty obvious, it may be surprising to learn that AI can be extremely power-hungry, especially when it comes to training the models that enable machines to recognize your face in a photo or for Alexa to understand a voice command.
The scale of the problem is difficult to measure, but there have been some attempts to put hard numbers on the environmental cost.
For instance, one paper published on the open-access repository arXiv claimed that the carbon emissions for training a basic natural language processing (NLP) model—algorithms that process and understand language-based data—are equal to the CO2 produced by the average American lifestyle over two years. A more robust model required the equivalent of about 17 years’ worth of emissions.
The authors noted that about a decade ago, NLP models could do the job on a regular commercial laptop. Today, much more sophisticated AI models use specialized hardware like graphics processing units, or GPUs, a chip technology popularized by Nvidia for gaming that also proved capable of supporting computing tasks for AI.
OpenAI, a nonprofit research organization co-founded by tech prophet and profiteer Elon Musk, said that the computing power “used in the largest AI training runs has been increasing exponentially with a 3.4-month doubling time” since 2012. That’s about the time that GPUs started making their way into AI computing systems.
Getting Smarter About AI Chip Design
While GPUs from Nvidia remain the gold standard in AI hardware today, a number of startups have emerged to challenge the company’s industry dominance. Many are building chipsets designed to work more like the human brain, an area that’s been dubbed neuromorphic computing.
One of the leading companies in this arena is Graphcore, a UK startup that has raised more than $450 million and boasts a valuation of $1.95 billion. The company’s version of the GPU is an IPU, which stands for intelligence processing unit.
To build a computer brain more akin to a human one, the big brains at Graphcore are bypassing the precise but time-consuming number-crunching typical of a conventional microprocessor with one that’s content to get by on less precise arithmetic.
The results are essentially the same, but IPUs get the job done much quicker. Graphcore claimed it was able to train the popular BERT NLP model in just 56 hours, while tripling throughput and reducing latency by 20 percent.
An article in Bloomberg compared the approach to the “human brain shifting from calculating the exact GPS coordinates of a restaurant to just remembering its name and neighborhood.”
Graphcore’s hardware architecture also features more built-in memory processing, boosting efficiency because there’s less need to send as much data back and forth between chips. That’s similar to an approach adopted by a team of researchers in Italy that recently published a paper about a new computing circuit.
The novel circuit uses a device called a memristor that can execute a mathematical function known as a regression in just one operation. The approach attempts to mimic the human brain by processing data directly within the memory.
Daniele Ielmini at Politecnico di Milano, co-author of the Science Advances paper, told Singularity Hub that the main advantage of in-memory computing is the lack of any data movement, which is the main bottleneck of conventional digital computers, as well as the parallel processing of data that enables the intimate interactions among various currents and voltages within the memory array.
Ielmini explained that in-memory computing can have a “tremendous impact on energy efficiency of AI, as it can accelerate very advanced tasks by physical computation within the memory circuit.” He added that such “radical ideas” in hardware design will be needed in order to make a quantum leap in energy efficiency and time.
It’s Not Just a Hardware Problem
The emphasis on designing more efficient chip architecture might suggest that AI’s power hunger is essentially a hardware problem. That’s not the case, Ielmini noted.
“We believe that significant progress could be made by similar breakthroughs at the algorithm and dataset levels,” he said.
He’s not the only one.
One of the key research areas at Qualcomm’s AI research lab is energy efficiency. Max Welling, vice president of Qualcomm Technology R&D division, has written about the need for more power-efficient algorithms. He has gone so far as to suggest that AI algorithms will be measured by the amount of intelligence they provide per joule.
One emerging area being studied, Welling wrote, is the use of Bayesian deep learning for deep neural networks.
It’s all pretty heady stuff and easily the subject of a PhD thesis. The main thing to understand in this context is that Bayesian deep learning is another attempt to mimic how the brain processes information by introducing random values into the neural network. A benefit of Bayesian deep learning is that it compresses and quantifies data in order to reduce the complexity of a neural network. In turn, that reduces the number of “steps” required to recognize a dog as a dog—and the energy required to get the right result.
A team at Oak Ridge National Laboratory has previously demonstrated another way to improve AI energy efficiency by converting deep learning neural networks into what’s called a spiking neural network. The researchers spiked their deep spiking neural network (DSNN) by introducing a stochastic process that adds random values like Bayesian deep learning.
The DSNN actually imitates the way neurons interact with synapses, which send signals between brain cells. Individual “spikes” in the network indicate where to perform computations, lowering energy consumption because it disregards unnecessary computations.
The system is being used by cancer researchers to scan millions of clinical reports to unearth insights on causes and treatments of the disease.
Helping battle cancer is only one of many rewards we may reap from artificial intelligence in the future, as long as the benefits of those algorithms outweigh the costs of using them.
“Making AI more energy-efficient is an overarching objective that spans the fields of algorithms, systems, architecture, circuits, and devices,” Ielmini said.
Image Credit: analogicus from Pixabay Continue reading
Posted in Human Robots