Canada, an Early A.I. Hub, Fights to Stay Relevant

In the late 1980s, Geoffrey Hinton was a few years into teaching at Carnegie Mellon University in Pittsburgh, Pa., when he became increasingly troubled about the state of the nation he had left his home country of England for a decade prior. Hinton took issue with Ronald Reagan’s foreign policy, particularly the mining of harbors in Nicaragua, and the fact that the A.I. research he was pursuing was largely funded by the U.S. Department of Defense. So when he was presented with an opportunity to head North, he jumped at the chance. 

“My wife and I were very fed up with the U.S.,” Hinton told Observer, “and Canada seemed like a good place.” Enticed to Toronto by a strong social system and a generous offer to become a fellow at the Canadian Institute for Advanced Research (CIFAR), a global research organization, Hinton made his way to the country in 1987. He’s largely stayed put ever since, picking up a Nobel Prize for his contribution to A.I. research along the way. 

Hinton wasn’t alone. Decades of sustained funding for curiosity-driven research has brought scores of pioneering A.I. researchers to Canada, where a series of breakthroughs laid the foundations for the A.I. products dominating today’s tech industry. Canada built upon this momentum in 2017 when it became the first country to implement a national A.I. strategy, one that congregated much of its innovative work in three A.I. hubs spread out across Toronto, Montreal and Edmonton. 

Despite the country’s contributions towards the now-booming technology, many say Canada has failed to reap the rewards of its own innovations. It isn’t just ideas that have been exported to the U.S., but much of the nation’s talent. “It’s this historic Canadian challenge of being often the inventors and pioneers of new technology, but not necessarily seeing the commercial success here,” Cam Linke, head of the Alberta Machine Intelligence Institute (Amii), told Observer. 

While attempts to establish competitive A.I. companies in Canada have been largely unsuccessful over the past few decades, a combination of enhanced government funding, bolstered research institutions and changing cultural attitudes is starting to make a gradual impact. The Toronto-based startup Cohere, for example, earlier this year raised $500 million—an unprecedented amount for a Canadian generative A.I. startup—from a mix of Canadian, American and international investors. While conceding that Canada’s A.I. “brain drain” is still an ongoing issue, Nick Frosst, a co-founder of Cohere, told Observer, “I feel the tide is turning.” 

Attracting the best researchers in the game

Long before companies like OpenAI and Anthropic broke out into the scene, Canada was a beacon for those drawn to ambitious A.I. research. The country might have had less national funding than the U.S., but it was an oasis for those pursuing long-term and experimental projects. Due to its social system and funding for basic research, “there were three researchers who were very happy to live in Canada,” said Hinton. They were Rich Sutton, Yoshua Bengio and Hinton himself—the latter two of whom would go on to be donned “Godfathers of A.I.” after winning the 2018 Turing Prize alongside Yann LeCun, now Meta’s chief A.I. scientist.

After Hinton set up shop for himself at the University of Toronto in the late 1980s, Sutton, an American researcher known for his pivotal work in reinforcement learning, headed out west to the University of Alberta as he became disenchanted with U.S. politics. Deep learning pioneer Bengio, meanwhile, returned to his hometown of Montreal to work at the University of Montreal. The presence of the three talents, in combination with Canada’s more lenient immigration politics, drew in even more A.I. researchers, according to Amii’s Linke. “That created this cycle of great people wanting to work with those folks.” 

While they may have been spread out across the country, Hinton, Sutton and Bengio were aligned in their passion for a particular field of A.I. research—one that for decades wasn’t even linked to the term A.I. “There was something called A.I. and there was something called neural nets, and they were in opposition,” according to Hinton, who described the two as “warring camps.” Traditional A.I. emphasized symbolic reasoning, while the neural net worldview was based on mirroring the human brain.

Despite being regarded as a “crazy theory” at the time, according to Hinton, the neural net field was backed by CIFAR. In 2004, for example, it began “Neural Computation and Adaptive Perception,” a program directed by Hinton that Bengio and LeCun also took part in. “It was a while before [neural nets] had practical applications, and so you needed to fund people to work on them without being able to produce any spectacular applied uses of them and so on,” said Hinton. “In the U.S., it was much harder to get that money.”

Researchers involved in the program met annually to present their ideas to each other, according to Ruslan Salakhutdinov, a professor at Carnegie Mellon University. In 2005, Salakhutdinov had strayed away from A.I. and was working in banking when he ran into Hinton, a former teacher of his at the University of Toronto, on the street. Hinton “sort of dragged me into his office,” according to Salakhutdinov, who said Hinton showed him his latest work on deep learning models and successfully urged Salakhutdinov to return to school to pursue a Ph.D. under him.

While the field itself was still struggling to gain recognition, a palpable sense of excitement was building throughout Canada’s neural net community—and by the 2010s, their hard work finally began to pay off. First came breakthroughs in neural networks displaying superior speech recognition abilities compared to other existing technology. Then, in 2012, Hinton and two of his students at the University of Toronto—Alex Krizhevsky and Ilya Sutskever, the latter of whom went on to co-found OpenAI—made waves when they won an object recognition competition known as ImageNet with neural networks and nearly halved the challenge’s previous error rate. The trio subsequently created a startup called DNNresearch, which was acquired by Google (GOOGL) for $44 million in 2013.

In the following years, the language generation capabilities of neural nets exploded and became mainstream knowledge through the release of chatbots like OpenAI’s ChatGPT. Hinton, meanwhile, was particularly struck by A.I. ‘s advances when he realized that language models like Google’s PaLM could explain why a joke was funny. “At that point, we all felt totally vindicated,” said Hinton.

“We all basically went to the U.S.”

As the promise of neural nets came true, the money started flowing and the offers came rushing. “All of a sudden, it’s like, ‘Hey, this actually has some legs—deep learning is something that has commercial applicability,’” Cameron Schuler, chief commercialization officer at the Vector Institute, told Observer. Sutton, the reinforcement learning pioneer, helped establish a DeepMind lab in Alberta, while Bengio co-founded startup Element AI and LeCun was snapped up by Meta (META). 

Hinton took up a position at Google while still advising graduates at the University of Toronto. The first to strike up this kind of dual appointment between Google and academia, he didn’t manage it without a fight. “You weren’t allowed to supervise graduate students, because you might leak Google IP to them,” said Hinton, who subsequently spent $400,000 on lawyers to negotiate with Google the right to supervise graduate students. “It wasn’t easy, but that sort of set the road for it.” 

As their professors attempted to balance industry with university-level research, young researchers headed to the U.S. in troves as the likes of Google, DeepMind and Apple (AAPL) snapped them up. The select group of neural net researchers, who had faced dire job prospects just a few years prior, was now getting offers above the $500,000 range. “It was just ridiculous,” said Salakhutdinov, who ended up at Apple for a few years and now works on generative A.I. research at Meta. “We all basically went to the U.S.” 

Canada had spent the past few decades pouring funds into the efforts of these researchers. “There’s always the risk that, when we do that without the commercial capacity, those people end up leaving, and that’s what actually started to happen,” said the Vector Institute’s Schuler. To combat Canada’s growing A.I. brain drain and to avoid massive losses on its investments, its government needed to take action.

In 2017, the Canadian government introduced the Pan-Canadian A.I. Strategy in a bid to regain its loosened grip on A.I. The country has since poured more than 2 billion Canadian dollars ($1.4 billion) into A.I. research and plans to invest an additional 2.4 billion Canadian dollars ($1.7 billion) to help bolster the sector. “It really was the recognition that if we didn’t do something to retain research, they would just leave,” said Schuler. Such figures still pale in comparison to spending from U.S. companies like Google, Microsoft (MSFT), Meta and Amazon (AMZN), which are on track to spend more than $200 billion on A.I.-related efforts in 2024 alone.

One of the A.I. Strategy’s first moves was to establish three Canadian hubs for the emerging technology. Bengio was firmly in charge of the Montreal one, known as Mila, while Sutton was an advisor to Alberta’s Amii, and Hinton joined the newly-created Vector Institute in Toronto. 

Despite Canada’s bolstered research scene, the domestic tech industry wasn’t as quick to embrace the new technology. In 2011, for example, after one of Hinton’s students made key advances in speech recognition via neural nets, the professor approached the Waterloo-based BlackBerry (then known as Research In Motion) to see if it was interested in developing the new system with the help of an intern. The company “had no interest in exploiting the first really useful industrial application of neural nets—we were going to give it to them for free, and they weren’t interested,” said Hinton. “Part of the reason that research in Canada isn’t as exploited as it might be, is that the big Canadian companies are far too conservative,” he added. 

Bengio’s A.I. startup Element was taken over by a U.S. company in 2020 when it was sold to Santa Clara, Calif.-based ServiceNow. “They didn’t really succeed in getting a product and were spending a lot of money quickly,” Simon Lacoste-Julien, an associate professor at Mila, told Observer.  

Schools, too, presented barriers to entrepreneurial efforts. At the University of Toronto, one of the most significant academic institutions for A.I. in the country, students using university resources to spin their research into startups had to give the school larger chunks of company ownership than their American counterparts at schools like Stanford and Carnegie Mellon, claimed Carnegie Mellon’s Salakhutdinov. Such equity “is typically negotiated as a minority shareholding on a case-by-case basis,” said the University of Toronto in a statement to Observer, adding that the school’s share typically ranges from single digit to low double digits.

Not to mention the nation’s sheer lack of compute infrastructure compared to the U.S., which Hinton described as “the one big drawback” for young researchers. Hinton recalled an instance where one of his former students, Jimmy Ba, went on to work at the Vector Institute but was unable to access enough of the graphics processing units (GPUs) needed to train large language models. He now works for Elon Musk’s A.I. startup xAI. “I don’t think you can expect [Canada] to be a world leader in anything other than the basic research—which it was, for a while—because it just doesn’t have the resources,” said Hinton. 

“A big company is not seen as a success.”

Despite a seemingly mass talent exodus, not all Canadian researchers departed the country—in some cases, foreign companies met them where they were and established outposts or research labs throughout the nation. Lacoste-Julien, for example, runs a Samsung (SSNLF) lab at Mila and notes that the expanded presence of international offices means more and more students are able to stay in Canada after graduating. “I don’t think it’s fully solving it because I don’t think it’s solvable, necessarily,” he said of the country’s brain drain issue. “But it’s day and night.” 

The principles ingrained across Canadian culture—the same ones that first attracted the likes of Hinton and Sutton to the country—might mean that even if the nation had the resources to compete, its A.I. aspirations simply won’t ever reach the heights of its foreign rivals. In provinces like Quebec, for example, “a big company is not seen as a success,” according to Lacoste-Julien, who has previously pushed back on requesting what he views as unnecessarily large amounts of funding for Mila. “What is a good success is a good life, good equality,” he added. 

Some emerging startups, however, are going against the grain. Founded by a group of former Google DeepMind researchers, the Edmonton, Alberta-based Artificial Agency, which aims to use generative A.I. to enhance gaming, emerged from stealth earlier this year with $16 million in funding. “We received a lot of compliments from the venture capital community when we went out to raise for having a very non-Canadian attitude about the ambition that we were trying to build,” Brian Tanner, one of the startup’s co-founders, told Observer. The company made sure to not shy away from describing its “grand vision” of developing a system that could control any character in a video game while attempting to court potential backers, a strategy “that really seemed to resonate in a way that some Canadian startups have struggled,” he said.

Artificial Agency isn’t the only company making a name for itself in Canada. After years of lackluster growth and support, a growing sense of momentum is seemingly underway across the country’s startup scene, especially in cities like Toronto. In 2022, the Canadian A.I. sector received $8.6 billion in venture capital—placing it behind only the U.S. and U.K. among G7 countries with the highest per-capita venture capital investment in A.I. Besides the rapid growth of OpenAI rival Cohere, promising Canada-based companies include the likes of Waabi, a startup focused on autonomous vehicles that raised $200 million in June. Artificial Agency, Cohere and Waabi are all backed by Radical Ventures, a Toronto-based venture capital firm that has emerged as one of the most powerful supporters of local A.I. initiatives and recently raised $800 million for a new fund focused on the new technology. 

Further bolstering such startup growth is integration between businesses and Canada’s research hubs. At Vector, for example, 94 percent of last year’s more than 1,000 graduates of Vector-recognized A.I. programs are still in Canada, with 91 percent in Ontario. “You’re seeing these Canadian companies, that historically would have seen them leave the country, instead really decide to build and grow in Canada,” said Amii’s Linke. “The opportunity to be able to tap into this talent that they desperately need is a big, really big, reason that they want to stay.” 

Artificial Agency, too, has noticed an increased synergy between the startup and academia scene. “We’re having to start competing for graduate students,” said Tanner, who described the nation’s A.I. hubs like Amii as “a collector for talent.” The presence of nascent, promising startups like his own, meanwhile, has established a new path forward for young researchers. Throughout the past few decades, “the pipeline was always to move away,” Mike Johanson, another Artificial Agency co-founder, told Observer. But with the arrival of a local A.I. ecosystem rich with examples for graduates to look at, he’s hoping a new pattern will be set. “Now, there’s a tradition to follow for the startups to come after us.”