Yoshua Bengio

If you have come across his name attached to a talk or an interview about where artificial intelligence is heading, the two threads to hold onto are representation learning and ethical AI. Those are the areas his name now stands for, and in his case the ethical questions were not a late addition to the technical work. Regulatory and ethical debate was treated as part of the design brief itself, a constraint to build around rather than a complaint to answer afterwards.

He was born in Montreal, Quebec, in March 1964, and works as an artificial intelligence researcher and professor. The early pattern in his work was translation: taking what was technically possible and finding the operational form it needed to take. Scale, from the beginning, was something he treated as a design problem rather than something that simply happens to a system once it succeeds. Progress came through small improvements that compounded, and through a willingness to run the risk of iteration.

By mid-career he was leading teams that took prototypes and turned them into platforms, making decisions about architecture, culture and where the money went. Failures were written up as experiments with something to learn from rather than buried. Much of his public output — memos, interviews, keynotes — has functioned less as promotion than as an explanation of method, which is why it travels well outside his own field.

Later work has leaned towards mentorship and building institutions. The archive of talks and design notes shows the amount of rigour sitting underneath what can look, from the outside, like intuition. Others have carried the approach into different domains, and his influence is now visible less in particular products than in the habits and organisations that outlasted them.