Geoffrey Hinton, a University of Toronto computer scientist whose work laid the foundations for a revolution in artificial intelligence and who has since been vocal about the technology’s dangers, has been named a co-recipient of this year’s Nobel Prize in physics.
Dr. Hinton, 76, was awarded the prize together with John Hopfield, 91, a physicist and professor emeritus at Princeton University. The two will share equally in the 11 million Swedish krona ($1.4-million) prize, announced on Tuesday by the Royal Swedish Academy of Sciences.
Beginning in the 1980s, Dr. Hopfield and Dr. Hinton made separate and crucial contributions to the development of computer algorithms known as neural networks. Based on the way neurons connect and interact in the brain, the networks can adjust themselves in response to their own performance, gradually optimizing their ability to solve various kinds of calculations.
While their progress came slowly at first, neural networks have since proved to be a watershed for machine learning. Aided by blindingly fast computer processors and vast repositories of training data that were not available decades ago, systems based on neural networks first became powerful and then transformative.
“I want to emphasize that AI is going to do tremendous good,” Dr. Hinton told The Globe and Mail after learning of his Nobel win. “In areas like health care, it’s going to be amazing. That’s why its development is never going to be stopped. The real question is, can we keep it safe?”
For Geoffrey Hinton, the godfather of AI, machines are closer to humans than we think
Neural networks lie at the heart of a range of applications that feature AI, including image search, speech recognition, forecasting, natural language processing and the ability to generate new content based on prompts, as evidenced by an array of commercial AI products such as ChatGPT.
In recent years, the advent of machine learning based on neural networks has been alternately hailed as a boon to discovery and productivity and feared as a disruptive force that will eliminate jobs, undermine creativity and potentially threaten human existence.
Indeed, so rapid has been the technology’s ascent, with AI-powered tools now rolling out across a wide range of industries and fields, that Dr. Hinton has become just as well known for his warnings about AI as for the technology he helped to create.
As AI becomes more ubiquitous, the opportunities for its misuse are also growing, including its use in weapons of war, Dr. Hinton said. A more existential threat is the longer-term possibility that AI systems will eventually grow so intelligent relative to human decision makers that they effectively take control.
“What we care about is people and we’d like people to stay in charge,” he said.
Dr. Hinton has received multiple accolades for his work, including the Turing Award, considered the most prestigious prize in computer science. But as a non-physicist he said he was taken completely by surprise when he was informed via a telephone call from Stockholm that he had won the physics Nobel.
When asked what the impact of becoming a Nobel laureate might have on an already celebrated career, Dr. Hinton said: “I think it will give me more credibility in talking about the risks of AI.”
Born in Wimbledon, near London, in 1947, Dr. Hinton completed his undergraduate studies at the University of Cambridge and then went on to earn a PhD at the University of Edinburgh in 1978.
After working at various universities in Britain and the United States, he came to the University of Toronto in 1987. It has been his academic home base for most of his career since then, including during a 10-year stint at Google. Dr. Hinton, who is both a Canadian and British citizen, ended his position with the U.S.-based company in 2023 so that he could speak freely about the potential dangers of artificial intelligence.
Dr. Hinton’s co-winner, Dr. Hopfield, was born in Chicago in 1933 and received his PhD from Cornell University in 1958.
After making important contributions to solid state physics and to the study of biochemical reactions, Dr. Hopfield became interested in the questions related to neuroscience, including how neurons in the brain act together to store and retrieve information. This led him to discover a computer-based analogue known as a Hopfield network, which allows a machine to retrieve a stored pattern that is most similar to a given input. His work was published in 1982 in what is now considered a landmark paper in the field.
Soon after, Dr. Hinton made use of Hopfield networks when developing a data-recognition algorithm. Using methods drawn from statistical physics, he and Terrence Sejnowski of Johns Hopkins University came up with a method in which a category of network called a Boltzmann machine could be trained to classify an image or similar input based on examples.
Dr. Hinton then continued to explore networks constructed in layers that could break down a task such as image or language recognition into bite-sized pieces that together led to an answer that the system calculated was most likely to be correct.
A key feature of this work was the use of a technique called back-propagation by Dr. Hinton and his collaborators, including U.S. psychologist and computer scientist David Rumelhart. This method amounts to feeding the network information about its own accuracy so that it can adjust how much weight to give to each of its subcomponents, or nodes, to keep improving through experience.
Today’s AI systems have their basis in this kind of interactive version of machine learning.
Yoshua Bengio, another machine-learning pioneer and scientific director of the Montreal-based research hub MILA, said the foundational work by Dr. Hinton helped to inspire his own trajectory in the field. The influence went beyond Dr. Hinton’s specific methods, Dr. Bengio said, and extended to “the more fundamental idea that maybe we can understand intelligence as something beautiful and simple, mathematically, which was not the way of thinking at that time.”
Through the 1990s, neural networks also had their detractors. Many experts favoured efforts that gave computer systems access to large amounts of information and predetermined rules for solving problems. But where those approaches enjoyed success in some specific domains, neural networks eventually proved more versatile in the era of fast computers and big data.
Dr. Hinton said that by 2006 it was apparent to him that the approach was working, but he was nevertheless astonished by the speed at which neural networks took off after that.
“What’s really surprised me is how good they are at understanding natural language – that happened much faster than I thought,” he said. “And I’m still amazed that they really do understand what they’re saying.”
While the Nobel citation underscores the early years of the field, Dr. Hinton said that his most important contribution was doggedly sticking with neural networks as a research focus for half a century, during which time he developed a variety of methods for improving them.
Dr. Bengio said that Dr. Hinton’s more recent outspokenness about the risks of AI is not a contradiction but a recognition of reality now that learning algorithms have proved surprisingly adept at mastering conversational language among other complex tasks.
“We have to be rational about it, and ‘be rational’ here is accept our lack of knowledge about how it could unfold, and then apply the precautionary principle to make sure nothing catastrophic happens,” Dr. Bengio said.
In his role as one of the “godfathers” of AI, Dr. Hinton is also known for fostering a circle of graduate students at the University of Toronto who have been instrumental in AI’s development, including Ilya Sutskever, former chief scientist at OpenAI, the company that created ChatGPT.
An emeritus professor at the University of Toronto since 2014, Dr. Hinton is chief scientific adviser at the UoT-adjacent Vector Institute for Artificial Intelligence. He also advises the Learning in Machines & Brains program at CIFAR (formerly the Canadian Institute for Advanced Research).
In a statement, Stephen Toope, CIFAR’s president and chief executive, praised Dr. Hinton’s persistence in the development of neural networks and his leadership in the organization’s AI-related research programs.
With Tuesday’s announcement, Dr. Hinton becomes one of seven Canadian-born or Canadian-based researchers to have won the world’s top physics prize.
The others are Richard E. Taylor (1990), Bertram Brockhouse (1994), Willard Boyle (2009), Arthur McDonald (2015), Donna Strickland (2018) and James Peebles (2019).
Dr. Hinton said that he made his decision to come to Canada not just to do science but because of the country’s social system and multicultural diversity.
“It doesn’t have as much money as places like the United States,” he said, but Canada’s federal granting system has a track record of supporting curiosity-driven basic research. “And that’s what led to all of this.”
Ivan Semeniuk
Science Reporter
The Globe and Mail, October 8, 2024