Tag Archives: makes
#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
#437265 This Russian Firm’s Star Designer Is ...
Imagine discovering a new artist or designer—whether visual art, fashion, music, or even writing—and becoming a big fan of her work. You follow her on social media, eagerly anticipate new releases, and chat about her talent with your friends. It’s not long before you want to know more about this creative, inspiring person, so you start doing some research. It’s strange, but there doesn’t seem to be any information about the artist’s past online; you can’t find out where she went to school or who her mentors were.
After some more digging, you find out something totally unexpected: your beloved artist is actually not a person at all—she’s an AI.
Would you be amused? Annoyed? Baffled? Impressed? Probably some combination of all these. If you wanted to ask someone who’s had this experience, you could talk to clients of the biggest multidisciplinary design company in Russia, Art.Lebedev Studio (I know, the period confused me at first too). The studio passed off an AI designer as human for more than a year, and no one caught on.
They gave the AI a human-sounding name—Nikolay Ironov—and it participated in more than 20 different projects that included designing brand logos and building brand identities. According to the studio’s website, several of the logos the AI made attracted “considerable public interest, media attention, and discussion in online communities” due to their unique style.
So how did an AI learn to create such buzz-worthy designs? It was trained using hand-drawn vector images each associated with one or more themes. To start a new design, someone enters a few words describing the client, such as what kind of goods or services they offer. The AI uses those words to find associated images and generate various starter designs, which then go through another series of algorithms that “touch them up.” A human designer then selects the best options to present to the client.
“These systems combined together provide users with the experience of instantly converting a client’s text brief into a corporate identity design pack archive. Within seconds,” said Sergey Kulinkovich, the studio’s art director. He added that clients liked Nikolay Ironov’s work before finding out he was an AI (and liked the media attention their brands got after Ironov’s identity was revealed even more).
Ironov joins a growing group of AI “artists” that are starting to raise questions about the nature of art and creativity. Where do creative ideas come from? What makes a work of art truly great? And when more than one person is involved in making art, who should own the copyright?
Art.Lebedev is far from the first design studio to employ artificial intelligence; Mailchimp is using AI to let businesses design multi-channel marketing campaigns without human designers, and Adobe is marketing its new Sensei product as an AI design assistant.
While art made by algorithms can be unique and impressive, though, there’s one caveat that’s important to keep in mind when we worry about human creativity being rendered obsolete. Here’s the thing: AIs still depend on people to not only program them, but feed them a set of training data on which their intelligence and output are based. Depending on the size and nature of an AI’s input data, its output will look pretty different from that of a similar system, and a big part of the difference will be due to the people that created and trained the AIs.
Admittedly, Nikolay Ironov does outshine his human counterparts in a handful of ways; as the studio’s website points out, he can handle real commercial tasks effectively, he doesn’t sleep, get sick, or have “crippling creative blocks,” and he can complete tasks in a matter of seconds.
Given these superhuman capabilities, then, why even keep human designers on staff? As detailed above, it will be a while before creative firms really need to consider this question on a large scale; for now, it still takes a hard-working creative human to make a fast-producing creative AI.
Image Credit: Art.Lebedev Continue reading
Posted in Human Robots
#437216 New Report: Tech Could Fuel an Age of ...
With rapid technological progress running headlong into dramatic climate change and widening inequality, most experts agree the coming decade will be tumultuous. But a new report predicts it could actually make or break civilization as we know it.
The idea that humanity is facing a major shake-up this century is not new. The Fourth Industrial Revolution being brought about by technologies like AI, gene editing, robotics, and 3D printing is predicted to cause dramatic social, political, and economic upheaval in the coming decades.
But according to think tank RethinkX, thinking about the coming transition as just another industrial revolution is too simplistic. In a report released last week called Rethinking Humanity, the authors argue that we are about to see a reordering of our relationship with the world as fundamental as when hunter-gatherers came together to build the first civilizations.
At the core of their argument is the fact that since the first large human settlements appeared 10,000 years ago, civilization has been built on the back of our ability to extract resources from nature, be they food, energy, or materials. This led to a competitive landscape where the governing logic is grow or die, which has driven all civilizations to date.
That could be about to change thanks to emerging technologies that will fundamentally disrupt the five foundational sectors underpinning society: information, energy, food, transportation, and materials. They predict that across all five, costs will fall by 10 times or more, while production processes will become 10 times more efficient and will use 90 percent fewer natural resources with 10 to 100 times less waste.
They say that this transformation has already happened in information, where the internet has dramatically reduced barriers to communication and knowledge. They predict the combination of cheap solar and grid storage will soon see energy costs drop as low as one cent per kilowatt hour, and they envisage widespread adoption of autonomous electric vehicles and the replacement of car ownership with ride-sharing.
The authors laid out their vision for the future of food in another report last year, where they predicted that traditional agriculture would soon be replaced by industrial-scale brewing of single-celled organisms genetically modified to produce all the nutrients we need. In a similar vein, they believe the same processes combined with additive manufacturing and “nanotechnologies” will allow us to build all the materials required for the modern world from the molecule up rather than extracting scarce natural resources.
They believe this could allow us to shift from a system of production based on extraction to one built on creation, as limitless renewable energy makes it possible to build everything we need from scratch and barriers to movement and information disappear. As a result, a lifestyle worthy of the “American Dream” could be available to anyone for as little as $250/month by 2030.
This will require a fundamental reimagining of our societies, though. All great civilizations have eventually hit fundamental limits on their growth and we are no different, as demonstrated by our growing impact on the environment and the increasing concentration of wealth. Historically this stage of development has lead to a doubling down on old tactics in search of short-term gains, but this invariably leads to the collapse of the civilization.
The authors argue that we’re in a unique position. Because of the technological disruption detailed above, we have the ability to break through the limits on our growth. But only if we change what the authors call our “Organizing System.” They describe this as “the prevailing models of thought, belief systems, myths, values, abstractions, and conceptual frameworks that help explain how the world works and our relationship to it.”
They say that the current hierarchical, centralized system based on nation-states is unfit for the new system of production that is emerging. The cracks are already starting to appear, with problems like disinformation campaigns, fake news, and growing polarization demonstrating how ill-suited our institutions are for dealing with the distributed nature of today’s information systems. And as this same disruption comes to the other foundational sectors the shockwaves could lead to the collapse of civilization as we know it.
Their solution is a conscious shift towards a new way of organizing the world. As emerging technology allows communities to become self-sufficient, flows of physical resources will be replaced by flows of information, and we will require a decentralized but highly networked Organizing System.
The report includes detailed recommendations on how to usher this in. Examples include giving individuals control and ownership of data rights; developing new models for community ownership of energy, information, and transportation networks; and allowing states and cities far greater autonomy on policies like immigration, taxation, education, and public expenditure.
How easy it will be to get people on board with such a shift is another matter. The authors say it may require us to re-examine the foundations of our society, like representative democracy, capitalism, and nation-states. While they acknowledge that these ideas are deeply entrenched, they appear to believe we can reason our way around them.
That seems optimistic. Cultural and societal change can be glacial, and efforts to impose it top-down through reason and logic are rarely successful. The report seems to brush over many of the messy realities of humanity, such as the huge sway that tradition and religion hold over the vast majority of people.
It also doesn’t deal with the uneven distribution of the technology that is supposed to catapult us into this new age. And while the predicted revolutions in transportation, energy, and information do seem inevitable, the idea that in the next decade or two we’ll be able to produce any material we desire using cheap and abundant stock materials seems like a stretch.
Despite the techno-utopianism though, many of the ideas in the report hold promise for building societies that are better adapted for the disruptive new age we are about to enter.
Image Credit: Futuristic Society/flickr Continue reading
Posted in Human Robots