Yoshua Bengio is a French-Canadian computer scientist and one of the three researchers, alongside Geoffrey Hinton and Yann LeCun, commonly credited with the foundational work that made Deep learning practical. He is a professor at Université de Montréal and founder of Mila, one of the world's largest academic deep learning research institutes.
Research contributions
Born in 1964 in Paris and raised largely in Canada, Bengio earned a PhD at McGill University before joining Université de Montréal in 1993. Through the 1990s and 2000s, while neural networks were unfashionable relative to methods like support vector machines, his lab kept working on gradient-based learning, contributing early neural probabilistic language models that anticipated the word- and sequence-embedding techniques used throughout modern Natural language processing. His students and collaborators made major contributions to sequence modeling, and his group trained Ian Goodfellow, who invented the Generative adversarial network as a Bengio-lab student in 2014.
Bengio shared the 2018 Turing Award with Hinton and LeCun "for conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing." Unlike LeCun, Bengio has generally been more aligned with Hinton in treating advanced AI capability as a serious source of risk rather than dismissing it.
Turn toward AI safety
Since 2023, Bengio has devoted much of his public profile to AI safety and governance. He was a signatory of the March 2023 "Pause Giant AI Experiments" letter (the Pause Giant AI Experiments letter) calling for a temporary halt on training the most powerful models, and he chaired the International Scientific Report on the Safety of Advanced AI, an initiative that grew out of the November 2023 AI Safety Summit at Bletchley Park summit and was commissioned by multiple governments to synthesize expert opinion on frontier AI risk.
Bengio has argued that the same scaling trends driving rapid capability gains in systems built on the Transformer (architecture) architecture also increase the plausibility of loss-of-control scenarios, and has called for stronger international coordination, mandatory safety testing and possibly binding regulation, positioning him among the most policy-engaged of the deep learning pioneers on questions of Existential risk from AI and near-term harms alike, an area also studied within the broader field of AI alignment research. He has also proposed technical directions of his own, including work on "AI scientists" designed to be trustworthy, non-agentic reasoning tools rather than autonomous actors, reflecting a broader shift in his research agenda from raw capability toward the safe deployment of increasingly powerful systems.