← All research spotlights

Research spotlight · Unit 16

Yoshua Bengio

Université de Montréal and Mila

Research field
Deep learning, AI safety, the scientific assessment of AI risks

Yoshua Bengio shared the 2018 Turing Award for the work on deep learning that the Foundation units of this masterclass presented, and since 2023 he has worked mainly on the safety of the most capable systems. With 24 co-authors, among them Geoffrey Hinton, Pieter Abbeel, and Atılım Güneş Baydin, he published a consensus paper in Science in 2024 that argues from the published record that safety research lags behind capability research and recommends that companies and funders direct at least one third of their AI research budgets to safety, that governments gain insight into frontier development through registration of frontier systems, incident reporting, and audits, and that developers make a safety case before deploying the most capable systems. He chaired the International AI Safety Report of January 2025 and its second edition of February 2026, which review the evidence without recommending policy. The unit presents the paper after its policy section and keeps the two documents apart.

In the masterclass

Section 6, after the policy record and the Asilomar excursion, presents the paper's recommendations and distinguishes them from the report he chaired.

The research is presented in Unit 16: Interpretability and Alignment, in a research spotlight box that explains the result at the level of the unit and links it to what you have just learned.

The paper

Bengio, Y., Hinton, G., Yao, A., Song, D., Abbeel, P., Darrell, T., et al. (2024). Managing extreme AI risks amid rapid progress. Science, 384(6698), 842‑845. https://doi.org/10.1126/science.adn0117

The scientists named here are not contributors to the AI Masterclass, are not affiliated with it, and have not endorsed it. The program presents and explains their published research, with references to the original papers.