Yoshua Bengio

Who they are

Yoshua Bengio is a Professor at Université de Montréal and co-founder of Mila — he won the Turing Award for deep learning research and in 2025 founded LawZero, a nonprofit AI safety organization incubated at Mila, where he now serves as Co-President and Scientific Director.

Person

Yoshua Bengio joined Université de Montréal in 1993, the same year he founded what would become Mila — Quebec's AI research institute — when deep learning was a fringe academic bet. He took his B.Sc. in Electrical Engineering and his M.Sc. and Ph.D. in Computer Science all from McGill, then spent decades building the Montreal school of deep learning alongside Yann LeCun and Geoffrey Hinton, the three of whom now define the canonical origin story of the field. He co-authored the MIT Press textbook 'Deep Learning' (2016) with Ian Goodfellow and Aaron Courville, which became the field's standard reference, and co-founded Element AI in 2016 to commercialise deep learning in industry — that company was later acquired. His public stance shifted markedly: he's said publicly he regrets not thinking earlier about the potentially harmful consequences of AI, and has since reoriented his platform around AI safety and governance. In 2025 he founded LawZero, a nonprofit AI safety research organization incubated at Mila, where he is Co-President and Scientific Director, with a board that includes Yuval Noah Harari and Sir John Rose. He writes at yoshuabengio.org on deep learning, AI safety, ethical AI, AI governance, and his 'Scientist AI' concept — AI systems designed to predict and understand rather than act autonomously. The through-line across 30+ years: build the capability, then reckon publicly with its consequences.

Company

Université de Montréal recently had the Eastern Canada Pandemic Preparedness Hub it leads receive nearly $100 million in federal government funding for four innovative projects — a signal of its continued pull on large-scale government research mandates. The university is a leading research institution in Canada with recognized strength in medicinal chemistry, pharmacology, and life sciences, and is embedded in Montreal's biotech ecosystem. It also participates in major international research collaborations, including the CFI-funded ANDES instrument for the European Extremely Large Telescope. For Bengio specifically, the university is the institutional home base from which he runs both his professorship and, through Mila, his broader AI research and safety agenda.

Market

Université de Montréal competes for faculty, funding, and graduate talent against McGill University, Université de Sherbrooke, Laval University, and Memorial University of Newfoundland, among others. Montreal's broader research and biotech ecosystem is supported by strong government grants, venture funding, and academia-industry partnerships, with many startups seeking Series A/B rounds to scale — Mila itself has been a feeder for that ecosystem. International research collaboration and federal funding competition are the two primary axes on which Canadian research universities differentiate.

Network

Bengio's closest intellectual peers are Yann LeCun and Geoffrey Hinton — the three co-pioneers of deep learning who collectively defined the field. His LawZero/LoiZéro board includes Yuval Noah Harari, Sir John Rose, Valérie Pisano, Maria Eitel, and Mariano-Florentino Cuéllar — a deliberately cross-disciplinary group spanning policy, history, and philanthropy. He also advises Valence Discovery and Recursion Pharmaceuticals as a scientific advisor, and has angel-invested in Ubenwa Health and Glowstick alongside Element AI co-founders.

Yann LeCun· Co-pioneer of deep learning; co-director of CIFAR program
Geoffrey Hinton· Co-pioneer of deep learning
Ian Goodfellow· Co-author, Deep Learning (MIT Press, 2016)
Aaron Courville· Co-author, Deep Learning (MIT Press, 2016)
Yuval Noah Harari· Co-board member, LoiZéro
Sir John Rose· Co-board member, LoiZéro
Valérie Pisano· Co-board member, LoiZéro

How they likely show up

Multi-decade tenure at Université de Montréal (joined 1993) → thinks in generational arcs, not product cycles; patience with slow-compounding bets is baked in.
Hybrid role pattern — professor, institute founder, nonprofit co-president, angel investor, scientific advisor — → operates across lanes simultaneously and likely expects conversations to move between the technical and the institutional without losing altitude.
Public writing at yoshuabengio.org covering deep learning, AI safety, governance, and the 'Scientist AI' concept → comfortable being a public intellectual, not just a researcher; expects interlocutors to have read the material.
Founded LawZero in 2025 and assembled a board including Yuval Noah Harari — a historian — alongside policy and philanthropy figures → deliberately builds coalitions outside the ML community; values interdisciplinary legitimacy over in-group signaling.
Publicly stated regret about not thinking earlier about AI's harmful consequences → self-critical in a disciplined way; will likely engage seriously with safety critiques rather than deflect them.
Angel investor in Ubenwa Health and Glowstick → still tracks the applied/commercial layer even while focusing on safety research; not purely academic in orientation.

Conversation tips

→ Come in having read something specific from yoshuabengio.org — the 'Scientist AI' framing or the LawZero founding post — he'll register immediately whether you've done the work.
→ He operates at the intersection of technical depth and policy advocacy; don't force him into just one lane — the most interesting exchange will cross both.
→ He has said publicly he regrets not acting earlier on AI risk — don't treat this as a throwaway admission; it's a load-bearing part of how he frames his current work and he'll engage seriously if you take it seriously.
→ Reference the LawZero board composition (Harari, Rose, Cuéllar) as a genuine question, not flattery — why those people, and what does that coalition unlock that a purely technical board wouldn't?

Toolbox

Openers

Open on the 'Scientist AI' concept — he's developed a specific architectural philosophy where AI systems should predict and understand rather than act autonomously, and LawZero is the institutional vehicle for that bet. It's a pointed technical-and-moral stance, not just a safety slogan.
Reference the LawZero board announcement from January 2026 — assembling Yuval Noah Harari, Sir John Rose, and Mariano-Florentino Cuéllar signals a deliberate move to legitimize AI safety arguments in policy and historical terms, not just ML ones.
Bring up the Element AI arc — he co-founded it in 2016 to industrialize deep learning, it was acquired, and now he's moved in the opposite direction with a nonprofit focused on restraining AI agency. That pivot is the most revealing data point about how his thinking has shifted.

Discovery questions

The 'Scientist AI' framing draws a hard line against autonomous agency — how do you think about where that line sits practically, given that most deployed AI systems already act on the world in some sense?
LawZero's board is unusually cross-disciplinary — a historian, a former diplomat, a philanthropist alongside AI researchers. What does that composition let you do that a technically-homogeneous board couldn't?
You've said publicly you regret not thinking earlier about AI's harmful consequences — at what point in the deep learning research arc do you think it became possible to see those consequences clearly, and what would it have taken to act on them sooner?

Avoid

Don't position AI safety as a fringe or speculative concern — he has built an institution around it and assembled a serious international board; treating it as a niche worry will read as not having engaged with his last several years of work.

Make it yours