Yoshua Bengio

Who they are

Yoshua Bengio is a Professor and AI Safety Researcher in Canada — co-founded Element AI (acquired) and launched LawZero in 2025, a non-profit incubated at Mila to develop safe, non-agentic AI.

Person

Yoshua Bengio built his reputation as one of the co-pioneers of deep learning, working alongside Geoffrey Hinton and Yann LeCun — the trio whose foundational work on neural networks reshaped the field. He co-founded Mila, the Quebec AI Institute, which became one of the world's most productive academic AI labs and a launchpad for Canadian AI startups. In 2016 he co-founded Element AI, a commercial venture applying deep learning to industrial and enterprise use cases; the company was later acquired. His angel investments — including Ubenwa Health and a $1.3 million round into Glowstick alongside Element AI co-founders — show continued engagement with applied AI. In 2025 he launched LawZero, a non-profit incubated at Mila and backed by philanthropic funding, focused specifically on building AI systems that predict and understand rather than act autonomously. The through-line is a researcher who moved from foundational science, through commercialisation, and back to safety — each move tracking his deepening concern about where the technology is heading. He is a consistent public voice on AI safety and international governance, speaking at the World Economic Forum and the UN, and his 'Scientist AI' concept — AI that avoids agency and deception — is the intellectual backbone of LawZero.

Network

Bengio's closest cited peers are Geoffrey Hinton and Yann LeCun, his co-pioneers of deep learning — both are frequently named alongside him in public and academic contexts. His network also extends into the Canadian AI startup scene through Mila and his angel investments in companies like Ubenwa Health and Glowstick.

Geoffrey Hinton· Co-pioneer of deep learning
Yann LeCun· Co-pioneer of deep learning

How they likely show up

Long-tenure signal at Mila and University of Montreal → thinks in decade-long research programmes, not product cycles.
Specialist role-type pattern → goes deep on a problem rather than spreading across domains; expects interlocutors to match that depth.
Thought-leader public writing signal and repeated high-profile speaking (WEF, UN, Lex Fridman) → comfortable being a public intellectual, likely responds to substantive intellectual engagement rather than sales framing.
Founded both a commercial company (Element AI) and two non-profits (Mila, LawZero) → distinguishes sharply between commercial incentives and mission-driven work; LawZero's philanthropic backing is a deliberate structural choice.
Scientist AI concept explicitly rejects autonomous agency in AI → he applies the same scepticism to products that claim agentic or autonomous capabilities.

Conversation tips

→ Open with a specific question about the 'Scientist AI' concept or the distinction between predictive and agentic AI — it's his current intellectual project and he'll have a lot to say.
→ Reference LawZero's 2025 launch and its incubation at Mila — showing you know the institutional structure signals you've done real homework.
→ If you have a view on international AI governance, share it — he speaks at the UN and WEF on exactly this and will engage seriously with a prepared opinion.
→ Don't conflate his commercial history at Element AI with his current stance — he has explicitly moved toward safety and non-commercial research; lead with mission alignment, not market opportunity.

Toolbox

Openers

LawZero's 2025 launch and its 'Scientist AI' architecture — an explicit institutional bet that safe AI requires avoiding autonomous agency, not just adding guardrails.
His WEF and UN appearances on AI governance in 2026 — he's actively shaping the policy conversation, not just the research one.
The arc from Element AI (acquired commercial venture) to LawZero (philanthropically funded non-profit) — a deliberate structural statement about how he thinks AI development should be organised.

Discovery questions

How does LawZero's 'Scientist AI' concept translate into concrete safety properties that organisations can evaluate or audit?
Where do you see the biggest gap between what international governance frameworks are proposing and what the research actually says is needed?
When you look at the current wave of agentic AI deployments, what does a responsible pause or constraint look like in practice — and who has the standing to enforce it?

Avoid

Don't pitch agentic AI capabilities or autonomous workflow tools — he has built an entire organisation around the thesis that autonomous agency in AI is the core risk, and leading with those features will immediately frame you as the problem.

Make it yours