Why 2015 Was a Breakthrough Year in Artificial Intelligence
After a half-decade of quiet breakthroughs in artificial intelligence, 2015 has been a landmark year. Computers are smarter and learning faster than ever.
The pace of advancement in AI is "actually speeding up," said Jeff Dean, a senior fellow at Google. To celebrate their achievements and plot the year ahead, Dean and many of the other top minds in AI are convening in Montreal this week at the Neural Information Processing Systems conference. It started in 1987 and has become a must-attend event for many Silicon Valley companies in the last few years, thanks to the explosion in AI. NIPS was where Facebook Chief Executive Officer Mark Zuckerberg chose in 2013 to announce the company's plans to form an AI laboratory and where a startup named DeepMind showed off an AI that could learn to play computer games before it was acquired by Google.
There should be plenty to discuss this week. The unprecedented advancements in AI research this year can be attributed to a confluence of nerdy factors. For one, cloud computing infrastructure is vastly more powerful and affordable, with the ability to process complex information. There are also more plentiful datasets and free or inexpensive software development tools for researchers to work with. Thanks to this, a crucial class of learning technology, known as neural networks, have gone from being prohibitively expensive to relatively cheap.
That's led to rapid uptake by the tech industry's largest companies, including Google, Facebook, and Microsoft. Each operates its own AI lab that conducts important research in the field and publishes much of it for the academic community to build upon. This year, Google researchers nabbed the cover of scientific journal Nature with a system that can learn to play and master old Atari games without directions. Facebook built a way to let computers describe images to blind people; Microsoft showed off a new Skype system that can automatically translate from one language to another; and IBM singled out AI as one of its greatest potential growth areas.
Startups are also contributing meaningfully to AI. Preferred Networks is making AI systems that will go into industrial robots made by Japan's Fanuc, and Indico Data Labs worked with a Facebook researcher to teach a computer how to paint faces using its own sort of imagination.
For a look at how far computer intelligence has come this year, here are six charts that should give you a clearer picture.
Computers have become a lot better at figuring out what's in a photo. In 2012, a team of University of Toronto researchers won the world's top image-recognition competition. The entire team was eventually recruited by Google, and its approach was quickly adopted by the company and its peers. In 2015, AI systems based on the project's approach, which relies on a technique called deep learning, have become much more accurate. In tests, error rates are down to less than 5 percent, making them better than some humans' performances.
Lots of companies are embracing AI, perhaps none more than Google. The Internet giant went from sporadic usage of deep learning in 2012 to applying it to thousands of projects this year.
Startups are adopting AI in big ways, too. CrowdFlower, which supplies structured data to companies, said it has seen a dramatic uptick in the amount of data being requested by businesses to help them conduct AI research. DiffBot, another startup, is using AI to improve its automated data-scraping tools.
A main focus of AI research is in teaching computers to think for themselves and improvise solutions to common problems. One way to do that is to give them a slimmed-down version of the real world, such as the simplified environments presented in video games, then ask them to explore it and record the results. (Check out the chart above for a look at how far Google's Atari project has come since 2013.) But the potential goes beyond games: Similar software could be used to teach things to AI computers and help them more quickly learn such new things as medical diagnostics, environmental science, or improved personal recommendations.
Google's Dean likens recent advancements in AI capabilities to evolution. "We're at this point in actual evolution where, previously, animals didn't have eyes, and now they have eyes," he said. "That's going to change a lot of stuff. Computers used to not be able to see very well, and now they're starting to open their eyes."