As the United Nations declares AI the “defining technology of our time,” forecasting earlier this year it will grow 25-fold in a decade to be worth $4.8 trillion (USD) by 2033, it’s difficult to imagine two AI winters when interest dwindled and programs in which millions were invested were simply abandoned. And it may be even more difficult to fathom that the “godfather of AI” — Nobel Laureate Geoffrey Hinton — was once thought to be wasting his time.
When Hinton first landed in Toronto in 1987, his pursuit of neural networks had fallen out of favour. “The predominant view in AI in those days was that neural networks were going nowhere,” recalls Graeme Hirst, Professor Emeritus and Hinton’s colleague at the University of Toronto’s Department of Computer Science, who specializes in computational linguistics and natural language processing. “Geoff was surprisingly dogged in working against the majority view.”
Pursuing research uninfluenced by trends is a key advantage academics had in Toronto. It’s what attracted great minds like Hinton, Hirst and others from the 1980s onwards, creating a world-class research centre in Toronto. Under Hinton alone, Toronto attracted and developed world-leading talent: Ruslan Salakhutdinov, who became director of AI research at Apple and later head of research at Meta; Yann LeCun who is now Meta’s chief AI scientist and shared the Turing Award with Hinton and Yoshua Bengio in 2018; and Ilya Sutskever, who went on to become co-founder and chief scientist at OpenAI.
This investment in talent ultimately laid the groundwork for the robust ecosystem that would follow — one that connects research to commercialization, and was backed by the first national strategy on AI globally.
Canada’s bet on curiosity-driven research
After a burst of interest and funding for AI in the 1960s and into the early 1970s, led by the United States and United Kingdom, the first AI winter set in the latter half of the decade, with enthusiasm waning until 1980. While the U.S. government invested heavily in open-ended AI research through the Defense Advanced Research Project Agency (DARPA, previously ARPA) throughout the 1960s, spending slowed in the early 1970s, when funding became more restricted, with DARPA abandoning open-ended pursuits for applied research. At the same time, the U.K. pulled back funding after mathematician James Lighthill’s 1973 report stating that the results from pouring millions into AI research had not lived up to the promise of AI’s potential.
Fraser Mustard was this incredibly inspiring and forward-thinking leader.– Sheila McIlraith, Professor of Computer Science, University of Toronto & Canada CIFAR AI Chair
As research agendas narrowed and the funding dried up abroad, an opportunity opened up for Canada. Dr. Fraser Mustard led a new drive to recruit top-tier talent to Canada to pursue interdisciplinary research as the founding president of Canadian Institute for Advanced Research (CIFAR, previously CIAR) in 1982. Described by the Globe and Mail as a “man in whom intellectual curiosity raged like a virus,” Dr. Mustard turned down the presidency at McMaster University to launch, coordinate and fundraise for ambitious projects.
This built on efforts that the University of Toronto had made in the future of technology more broadly — investing in the first electronic computer in Canada in 1952, and launching the country’s first computer science department in 1964.
CIFAR’s very first program focused on AI and robotics, and part of Dr. Mustard’s strategic efforts attracted both rising stars and established talent to Toronto, including Hinton, and two scholars focused on knowledge representation and reasoning, Raymond Reiter and Hector Levesque.
“Fraser Mustard was this incredibly inspiring and forward-thinking leader,” says Sheila McIlraith, who is a Professor of Computer Science at the University of Toronto and a Canada CIFAR AI Chair. “Hiring excellent people and then allowing them to pursue their work was really important.”
During the 1980s, the spotlight shifted to symbolic AI — the use of logic-based systems to model reasoning. “Hector and Ray believed, [along] with many other people, that logical systems were the way to go,” says Hirst. “Vast areas of AI developed that paradigm, and U of T — thanks to Hector and Ray — was a leader in that work.”
In the U.S., the funding often followed the hype, which was decidedly low in the 1980s, but further north, CIFAR and the University of Toronto still supported scholars whose work wasn’t as in vogue — like Hinton’s. And it revealed two things about the culture of research in Toronto at the time. First, that it was collegial at its core. “They all respected each other as reasonable people and outstanding scholars,” explains McIlraith.
But more importantly, their work was not tied to applied research focusing on defense like in the U.S., driven instead by Canada’s funding ethos through the Natural Sciences and Engineering Research Council (NSERC), which was established in 1978 in part “to improve the program of free research.”
“Canada created this place for us to do curiosity-based research,” McIlraith recalls. “That’s one [of the reasons] that we’ve been able to excel.” Decades later, that same ethos would underpin institutions like the Vector Institute, built to turn Toronto’s research strength into applied impact across the economy, but only after years of backing work with no obvious commercial end point.
And while today’s focus is on neural networks and large language models, the pendulum could swing again. “People are recognizing the limitations of large language models in doing correct reasoning,” says Hirst. If and when AI research turns to new frontiers, that curiosity-driven culture leaves Toronto’s labs unusually well prepared to figure out what comes next.
Becoming ‘the centre of the research world’
By the late 1980s, another AI winter set in, and as the economy stalled, so did investment in research. But academics like Hinton, Reiter, Levesque and others continued to train the next generation of AI scholars. One of these graduate students was McIlraith, who completed her PhD at the University of Toronto under Reiter. Her background was in math and statistics, and she ended up studying knowledge representation and reasoning.
Like many in her cohort, McIlraith left Canada for the U.S. in the late 1990s after completing her doctorate, heading to Palo Alto for a joint postdoctoral fellowship at Stanford and the Xerox PARC, a pioneering lab created by the company in 1970. Just 10 months into her two-year postdoc, McIlraith was hired as a research scientist at Stanford, and later was promoted to senior research scientist.
As Canada ramped up computer science and AI research funding in the early 2000s, McIlraith jumped at the chance to return home. She wasn’t alone — several scholars in machine learning made similar moves back to Canada, including David Fleet, now a University of Toronto professor and principal scientist and site lead for Google DeepMind’s Toronto Lab, and Brendan Frey, who returned after work at Caltech and went on to become a leading figure in probabilistic machine learning and co-founder of Deep Genomics as well as Vector.
Roughly a decade later, U of T continued to attract world-class AI researchers. Raquel Urtasun — founder of Waabi and also a co-founder of Vector — and Sanja Fidler, an associate professor at U of T, Vector co-founder and VP of AI Research at NVIDIA, joined during this period, further expanding the institution’s deep talent bench.
This recruitment effort bolstered what Toronto had already built. “We were the centre of the research world,” says Avi Goldfarb, Rotman Chair in Artificial Intelligence and Healthcare, and Professor of Marketing at the Rotman School of Management, University of Toronto. “We were creating the leaders who ended up building tech companies and going to tech companies with the next generation of AI products.”
We were creating the leaders who ended up building tech companies and going to tech companies with the next generation of AI products.– Avi Goldfarb, Rotman Chair in Artificial Intelligence and Healthcare, & Professor of Marketing, Rotman School of Management, University of Toronto
Those behind a 2012 breakthrough in image recognition set the stage for a much larger revolution in AI. Hinton, along with his then-graduate students Alex Krizhevsky and Ilya Sutskever, created a neural network called AlexNet, which was trained on a dataset called
ImageNet, a project started by Stanford professor Fei-Fei Li that categorized hundreds of thousands of images.
AlexNet laid the foundation for the way deep learning is used today in tools like ChatGPT. Earlier this year, it was announced that its source code will be preserved by the Computer History Museum in Mountain View, Calif. The academic paper behind AlexNet “revolutionized the field of computer vision and is one of the most cited papers of all time,” according to Jeff Dean, chief scientist of Google DeepMind and Google Research.
Supercharging start-ups
While Toronto is still nurturing the next generation of talent, says Goldfarb, where it’s made progress since the early aughts is connecting innovative ideas to the market. The same year that AlexNet was released, a key pipeline for commercializing research was built at the Rotman School for Management. Professor of Strategic Marketing Ajay Agrawal saw the dearth of scalable, science-based start-ups in Canada, and created the Creative Destruction Lab (CDL) in 2012 with the mission of supporting seed-stage companies to meet their potential.
Startups that make it through the application process are taken through CDL’s carefully honed nine-month program that includes mentoring by accomplished entrepreneurs, angel investors and scientists. By the 2014-15 academic year, more than half of the strongest applications were AI start-ups, said Goldfarb, crediting Toronto being ahead on the research side for the strong show in the field. That’s when CDL decided to split into two streams for the 2015-2016 year — one for AI, and then another for all other start-ups. That was also the year that CDL went from being of local interest to something Silicon Valley was paying attention to.
The following year, the AI cohort included Xanadu, Tenstorrent, and Ada — which all went on to attain valuations over $1 billion, achieving unicorn status. “There aren’t that many Canadian unicorns, and three of the AI unicorns came through us,” says Goldfarb.
CDL initially set a goal for graduates to cumulatively generate $50 million in equity value within the program’s first five years — today, these companies have generated over $51 billion in equity value. Goldfarb credits Agrawal for its incredible success in becoming a globally-recognized program, one that now operates out of 15 cities, including London, Berlin and Melbourne.
Commercialization got a further boost when the Vector Institute was founded in 2017, with a $135 million investment from various levels of government and 40 companies to create a not-for-profit corporation to advance AI application, adoption and commercialization across Canada. It was born out of the Pan-Canadian AI Strategy, and bolstered by Toronto’s deep-learning leadership, its founders including Frey, Urtasun and Hinton, who became the institute’s chief scientific advisor.
The Vector Institute systematized what had already been happening in Toronto informally: turning research superiority into applied results. Since its founding, Vector has provided millions in scholarships, offered key industry insights, and partnered with heavyweights, including many of Canada’s largest banks and telecoms.
From breakthrough to boom
Canada’s strategy to protect curiosity-driven research, even when it went against prevailing trends, cemented Toronto’s foundational role in the development of AI. Through two AI winters, that commitment sustained a durable ecosystem: the University of Toronto and CIFAR built and maintained the research base, the Vector Institute added structure at the point where that research meets industry, and programs like CDL created the conditions for start-ups to grow into scalable companies.
The result is visible in how AI continues to feed growth in the city. Researchers move into product teams without leaving the ecosystem. Startups grow beside enterprise labs. Banks, hospitals, and manufacturers test, refine, and adopt models with the people who built them. The traffic runs both ways: applied questions shape new research while new techniques are in turn fed back into the field.
This is why Toronto’s legacy matters now. The same culture that kept faith with neural networks during the lean years has made the region effective at applying AI in the real economy — at scale, and with an emphasis on reliability over spectacle.
EXPLORE MORE STORIES
