Canadian AI researchers win prestigious Turing Award
Three computer scientists who laid the foundations for many of the recent advances in artificial intelligence are being honoured with this year’s Turing Award, considered the field’s highest accolade.
Geoff Hinton, an emeritus professor at the University of Toronto and a senior researcher at Alphabet Inc.’s Google Brain, Yann LeCun, a professor at New York University and the chief AI scientist at Facebook Inc., and Yoshua Bengio, a professor at the University of Montreal as well as co-founder of AI company Element AI Inc., will share this year’s award, which is given annually by the Association for Computing Machinery.
The three winners will split a US$1 million prize that comes with the award, which is currently underwritten by Google. Hinton said he will donate a portion of his share to the University of Toronto’s humanities department. “They are much less well-funded and I think the humanities are very important to the future,” he said in an interview. Bengio said he may use his share to help fight climate change.
Past winners of the award, sometimes called the “Nobel Prize of Computing,” have included Tim Berners-Lee, who invented the world wide web, and Whitfield Diffie, who helped pioneer public-key cryptography.
This year’s three winners are often referred to collectively as the “Godfathers of Deep Learning” for their research into neural networks -- a kind of machine-learning software that loosely mimics the way the human brain works.
In 1983 Hinton co-invented Boltzmann machines, one of the first types of neural networks to use statistical probabilities. Three years later, he co-authored a seminal paper demonstrating that a technique for updating the strength of the connections within a neural network, known as backpropagation, could imbue this software with remarkable learning capabilities.
When Hinton began work with neural networks in the late-1970s and early-1980s, they were deeply unfashionable. At the time, most computer scientists believed the technique was a dead end. Software that explicitly encoded human expertise in a complicated set of rules was deemed a better approach to artificial intelligence.
Today, deep neural networks using backpropagation underpin most advances in artificial intelligence, from Facebook’s ability to automatically tag your friends in photos to the voice recognition capabilities of Amazon.com Inc.’s Alexa and Google’s translations from English to Mandarin.
LeCun, who did postdoctoral work under Hinton and worked to improve his backpropagation techniques, developed convolutional neural networks, the kind of software architecture that gives today’s computer vision systems their power.
Bengio, who worked with LeCun on computer vision breakthroughs when they were at Bell Labs, went on to apply neural networks to natural language processing, leading to big advances in machine translation. More recently, he has worked on a method best known for enabling neural networks to create completely novel, but highly realistic, images.
Hinton, in an interview ahead of the announcement, said he believes deep learning will eventually allow computers to have human-like or even super-human intelligence.
“I think we will discover that conscious, rational reasoning is not separate from deep learning but a high-level description of what is happening inside very large neural networks,” he said.
Bengio, who participated in the same interview, said new architectures for deep learning would be needed, however, before a neural networks could match the kind of general intelligence the human brain possesses.
Hinton disparaged the idea -- which has been advocated by New York University professor Gary Marcus among others -- that deep learning will need to be combined with older, symbolic AI techniques to achieve human-level intelligence. Hinton compared this to using electric motors only to run the fuel-injectors of gasoline engines, even though electricity is far more energy efficient.
But both were wary about offering predictions for when deep learning might yield general intelligence. Hinton noted that when he developed backpropagation, he thought practical applications of deep learning would happen almost immediately. “We were quite optimistic,” he said.
Instead, it took almost a quarter century, and the advent of much larger data sets and more powerful computers, before commercial applications of deep learning became viable.
“We had no way of estimating how much data you would need and how much computation you would need,” he said. It wasn’t until 2009, when deep learning methods made speech recognition much better than prior techniques, that Hinton said he was finally certain the method was on the verge of having a massive real world impact.
Hinton and Bengio said they were concerned about misuse of the technology they helped invent, especially regarding weapons systems that use deep learning to find and engage targets without human input. But both said the positive impact of deep learning, which they think will vastly improve business productivity, will outweigh any ill effects.