As Aaron Courville steps into leadership at IVADO, he urges the research community to harness AI’s collaborative power for real-world solutions—insisting optimism and oversight, not fear, should shape the future of artificial intelligence.
Aaron CourvilleCredit: Maryse Boyce
Aaron Courville doesn't fear artificial intelligence. He embraces it.
"There have always been ethical concerns about these kinds of automated systems, and if we were to do nothing about it that would be concerning, but on the whole, I'm very optimistic about what AI can offer," said the 50-year-old computer science professor and incoming scientific director of IVADO (Institut de valorisation des données), the AI research consortium partly affiliated with Université de Montréal.
"AI can be part of the solution for many things," Courville said during a career-spanning interview in his office at Mila - Quebec AI Institute, which he co-founded. "The same challenges that we face with multi-agent systems" – teams of algorithms that combine expertise to solve problems – "can be transformed to help us work together to help alleviate climate change, for example."
Or take the realm of public policy, he added: "Imagine we can harness AI to do computational policy design, creating more effective policies that we can use to govern our societies. There are many upsides to a lot of these mechanisms – and they're upsides that people don't talk enough about. We really should be making this clear to everyone, not just ourselves."
In this regard, Courville distinguishes himself from his friend and colleague Yoshua Bengio, the UdeM computer science professor he's now replacing at IVADO. It was Bengio who, two decades ago, recruited Courville to the university as a young research fellow, wrote the reference textbook Deep Learning with him, and is now a leading global voice warning of the dangers of unfettered AI.
"Yoshua is very concerned about the existential risk of AI: like, the end of humanity," said Courville.
"It's a pessimistic scenario that imagines some kind of AI that's all-powerful, but that's not the future I see. The future I see will have many AIs, each with their roles and responsibilities, and they will check the power and influence of each other. No monolithic AI will get the keys to the kingdom.
"For the most part, this is how we have organized our societies against tyranny and injustice and I see no reason why we would not continue with this strategy into the age of AIs."
Courville comes by his optimism naturally: as a scientist he has dedicated his life, as he puts it, to figuring out how things work. His epiphany came in the late 1990s when, after completing his undergraduate studies in applied science at the University of Toronto, he spent one of his summers as a master's student in biomedical engineering in the laboratory of UofT professor Berj Bardakian.
"There was something there in that lab, in the interface of biology and engineering, that really caught my imagination," Courville recalled. "Biology is a great inspiration for elegant design, and I was inspired by that quite heavily. It's what got me interested in neural networks, taking what happens in the biology of the brain and finding similar methods to teach computers to process data."
He pursued his interest south of the border. As a PhD student in the Robotics Institute of Carnegie Mellon University, in Pittsburgh, "I dug through the vaults, reading up on 100-year-old experiments on animal learning and figuring out how to model examples of Pavlovian conditioning using Bayesian methods of machine learning. My thesis was that animals are causal Bayesian learners: what they think of the world is only as complex as the data they're exposed to."
Courville's own world, growing up, was unsophisticated, though it did have its dualities. There was the French family name, for one, a legacy of a Franco-Ontarian grandfather. Born in Regina, son of a SaskTel administrator and a nurse, brother to an older sister, young Aaron spent his early years there and in Saskatoon, where he was in French immersion at school. At age 8, the family moved to Toronto and at age 11 to Cornwall, where Aaron's father changed professions and started lawyering in general practice.
"As a boy, I was always interested in science." Courville recalled.
"I played a lot of Lego, had a model rocket set, that sort of thing. I thought I was going to be an inventor, actually. I didn't know what a researcher was, but if I had, you could have added that to my list."
At Carnegie Mellon, Courville had his second epiphany: he met his intellectual and romantic match. Joëlle Pineau was also doing her PhD in computer science, was a bilingual Canadian (her mother tongue is French and her hometown Ottawa), and planned to find work back in Canada after graduating. First came co-habitation, then a first child (the first of four), and in 2004 the couple moved to Montreal.
Pineau took up a professorship at McGill University and would go on to head up AI research at Meta, the parent company of Facebook (eight years in, she'll be stepping down from the job later this month). Courville finished up his PhD thesis in coffee shops and a borrowed office at McGill and wondered what to do next. "I wasn't going to work on animal experiments for the rest of my life. Machine learning was where it was at."
He approached Bengio, they met and chatted, he gave a talk to his mentor's students and soon joined the team as one of his postdocs. And then this happened: "I walked by his office one day and Yoshua came out, very excited, saying he'd just got off a call with Geoff Hinton (of UofT) and Yann LeCun (of New York University). 'We're starting this thing called 'deep learning' is what he said. It's amazing I was there for that moment."
In 2012 Courville joined the computer science department at UdeM as an assistant professor, four years later published the AI textbook with Bengio and Google's Ian Goodfellow, and in 2023 was made full professor. Like Bengio, Courville steered clear of accepting jobs with the tech giants that benefit from his research, though he continues to take grants from the likes of Google, Microsoft, Samsung, Sony and Hitachi.
"I've never felt a need to be something other than I am, and that includes working for the tech companies," he said. "Of course I collaborate with them all the time – Meta, Microsoft, Google, they're all here in Montreal – but they're not my employer, UdeM is.
"I'm still very much in the trenches," he explained. "I'm a researcher. This is what I do. I like my job, I like Mila, and I like working with students." He now has 20, mostly PhDs but also a few master's, offering them classes in English and French. His alumni include Open AI's Max Schwartzer, Cohere's Sara Hooker and Florian Strub, and UdeM assistant professor David Scott Krueger.
Now comes the opportunity to do more with IVADO. With his research interests lying in generative models, large language models and reinforcement learning, Courville had advised for the consortium as a member of its executive for several years before he was approached this year to replace Bengio, who was stepping down to concentrate on his AI Safety projects, warning of AI's risks to humanity.
At IVADO, Courville will bring experts together around 10 "research regroupements," or "thrusts", ranging from neuroscience to the environment to supply chains. "There's a chance for a lot of cross-fertilization, always with the idea of 'How can we push AI into a wider set of fields across academia?'," he said. "There's so much more we can do together as our discipline develops."
What does the immediate future hold for Aaron Courville?
The lanky academic leaned his 6'2" frame back in his chair, nudged his Ray-Ban prescription glasses up the bridge of his nose, gave a little toss of his longish pepper-and-salt hair, and laughed. "I'll either be doing what I do now ... or be retired and fishing somewhere," he said. He held up his phone to show a picture of a big striped bass his family caught on a recent expedition to New Brunswick's Miramachi region.
As for the future of AI, one thing's for sure: Courville is keeping an open mind.
"It is a time of change, no question," he said. "There's lots still to negotiate. But the field is not going to stop advancing and I think we, as researchers and as academics, need to participate in shaping its future. It's either that or be left behind."