Hugo Larochelle is the newly appointed scientific director of Mila, the Quebec Artificial Intelligence Institute, in Montreal.Boris R. Thebia/The Globe and Mail
When Hugo Larochelle started his PhD in computer science, he wondered whether the artificial-intelligence community was really the one for him. He studied under people who are recognized today as influential – namely, Yoshua Bengio and Geoffrey Hinton – but the work they were doing 20 years ago was not always seen as important.
The first paper Dr. Larochelle worked on was met with disinterest because it involved artificial neural nets, a concept inspired by the way humans think that had yet to gain much traction. “It was a bit of a shock, because that was the only kind of AI I knew,” he recalled.
The situation is very different today, and the research work done in Canada on neural nets and deep learning help form the basis of the current wave of generative AI technologies. In that sense, Dr. Larochelle is well positioned for the role he assumes Tuesday, as the new scientific director at Mila, the Quebec Artificial Intelligence Institute, in Montreal.
He succeeds Dr. Bengio, the Turing Award winner who founded Mila. (Laurent Charlin, an associate professor at HEC Montreal, had been serving in the interim since March.) Dr. Bengio is now running LawZero, a non-profit dedicated to AI safety research.
Opinion: Ottawa’s AI push must translate into savings
Dr. Larochelle, an adjunct professor at University of Montreal who most recently worked at Google DeepMind, has been appointed at a critical time for AI in Canada. The federal government has announced $2-billion to improve access to the computing power necessary for building and running AI models, while Prime Minister Mark Carney wants to spur adoption in the public service. He’s also created a ministerial role dedicated to AI.
The rest of the world is just as determined to capitalize on AI, and big tech companies are spending huge sums of money to secure talent. For Mila, attracting and retaining the brightest minds could become more difficult.
But Dr. Larochelle said he is proof of the institute’s appeal, having left DeepMind’s Montreal office earlier this year to join the non-profit. “I view the mission of the scientific method applied to AI as being very important, and that’s a little bit less of a priority for some of these players,” he said. “I’m probably not alone.”
Mila is one of three national research outfits, including the Vector Institute in Toronto and the Alberta Machine Intelligence Institute in Edmonton, that form the backbone of a federal AI strategy launched in 2017.
‘I hope I’m wrong’: Why some experts see doom in AI
“It’s hard to follow in Yoshua’s footsteps, but I think Hugo is a good choice,” Dr. Hinton said, describing his former student as an energetic researcher and manager. “A lot of very good researchers aren’t also very good managers, but Hugo is,” he said.
Dr. Larochelle completed a computer science PhD in 2009 under Dr. Bengio at the University of Montreal, and finished a post-doc at the University of Toronto with Dr. Hinton. Both labs pursued deep learning, an approach to AI that others in the field didn’t always believe held much promise.
He recalled attending AI conferences where the ideas and concepts under discussion were far removed from what he and his colleagues were talking about. “I’m kind of wondering, am I in the wrong place?” he said.
In 2012, a paper he co-wrote about using machine learning algorithms to hone other algorithms did prove influential, however. So much so that he and his colleagues founded a company, Whetlab, based on the research. Twitter, now X, acquired it in 2015, and Dr. Larochelle moved to Boston.
Opinion: AI is winning hearts and minds in the classroom – but at what cost to our cognitive future?
He left less than two years later and returned to Montreal to head up Google’s AI outpost in the city, where the mix of projects were eclectic: AI coding tools, classifiers to identify bird calls and AI agents to control balloons that beamed internet connectivity to remote areas. He left this past April after more than eight years when the opportunity at Mila opened up.
Now at the institute, he hopes to boost AI adoption by working with companies, while helping to spin off startups and advancing the science. In particular, he sees opportunities to use AI to protect biodiversity, stemming from his work at DeepMind. The tools developed there have been used by ecologists to analyze the health of coral reefs, identify bird calls in Hawaii to improve monitoring of endangered species and model bird occupancy on Christmas Island, according to a paper he contributed to earlier this year.
Even though AI is hypercompetitive, Dr. Larochelle said Mila has advantages for attracting talent. AI is becoming a more closed field, for one thing, with some companies less interested in engaging in open science for competitive reasons – the opposite of academia. That’s partly what motivated him to take the gig at Mila. “It seemed like a natural path for me to go to an institute where scientific discovery and contributing are the primary mission,” he said.
He expects that philosophy to draw others. “If you’re seeing less of this open collaboration in the industry, that’s one thing we can contribute,” he said.