Yoshua Bengio
Scientific Director · Turing Award 2018
MILA · Université de Montréal
Turing Award winner and pioneer of deep learning. Leads MILA, the world's largest academic AI lab. Has become increasingly vocal on AI governance, calling for international safety standards and treaty-level coordination.
Core Positions & Ideas
Neural Language Models Can Learn Word Meaning
2003Co-authored the paper introducing neural probabilistic language models — a direct ancestor of modern LLMs. Showed that neural networks could learn distributed representations of words, encoding semantic meaning. This work was largely ignored for a decade before transformers vindicated it.
Deep Architectures Are Necessary for Abstraction
2009In 'Learning Deep Architectures for AI', argued that shallow networks cannot learn the kind of hierarchical, compositional representations needed for intelligence. This provided the theoretical grounding for why depth matters in neural networks.
System 2 Deep Learning — Machines Must Reason, Not Just Perceive
2018Influenced by Kahneman's 'Thinking, Fast and Slow', argued (2018–2020) that current deep learning only implements System 1 (fast, intuitive). AGI requires System 2 (slow, deliberate, causal reasoning). Proposed incorporating causality, variable binding, and attention as steps toward machine reasoning.
AI Safety Is Now the Most Important Research Priority
2023Publicly shifted focus from capability research to safety after 2022. Led the International Scientific Report on the Safety of Advanced AI (presented to 30+ governments, 2024). Argues the field needs treaty-level international cooperation, not voluntary commitments.
We Need 'Cautious AI' That Asks Before Acting
2024Proposed formal frameworks for AI systems that bound their own probability of causing harm and proactively seek human oversight when uncertain. Argues that the right architecture for safe AI is one that knows what it doesn't know — and acts accordingly.
Essential Reading & Watching
A Neural Probabilistic Language Model
The paper that introduced word embeddings and neural language models. A direct ancestor of word2vec, GPT, and all modern LLMs.
Read / Watch →
Representation Learning: A Review and New Perspectives
A comprehensive survey of deep learning and representation learning. One of the most cited papers in ML, accessible to researchers from all backgrounds.
Read / Watch →
The International Scientific Report on the Safety of Advanced AI
Bengio chaired this landmark report presented to 30+ governments. The most authoritative scientific consensus document on AI risk to date.
Read / Watch →
Yoshua Bengio's Blog — AI Safety Essays
A series of in-depth essays on AI existential risk, international governance, and the psychology of researchers confronting catastrophic AI risks.
Read / Watch →
Recent Writing (8)
Introducing LawZero
I am launching a new non-profit AI safety research organization called LawZero, to prioritize safety over commercial imperatives.
June 3, 2025Read original →
Implications of Artificial General Intelligence on National and International Security
As highlighted in the International Scientific Report on the Safety of Advanced AI, the capabilities of general-purpose AI systems have been steadily increasing over the last decade, with a pronounced acceleration in the last few years.
October 30, 2024Read original →
Bounding the probability of harm from an AI to create a guardrail
As we move towards more powerful AI, it becomes urgent to better understand the risks, ideally in a mathematically rigorous and quantifiable way, and use that knowledge to mitigate them.
August 30, 2024Read original →
Reasoning through arguments against taking AI safety seriously
About a year ago, a few months after I publicly took a stand with many other peers to warn the public of the dangers related to the unprecedented capabilities of powerful AI systems, I posted a blog post entitled FAQ on Catastrophic AI Risks as a follow-up to my earlier one about rogue AIs, where I started discussing…
July 9, 2024Read original →
The International Scientific Report on the Safety of Advanced AI
In November 2023, I’ve been given the responsibility to chair the International Scientific Report on the Safety of Advanced AI as part of an international mandate following up from a resolution taken by 30 countries, as well as representatives from the EU and the UN, at the UK AI Safety Summit in Bletchley Park.
June 19, 2024Read original →
Towards a Cautious Scientist AI with Convergent Safety Bounds
How can we design an AI that will be highly capable and will not harm humans? In my opinion, we need to figure out this question - of controlling AI so that it behaves in really safe ways - before we reach human-level AI, aka AGI; and to be successful, we need all hands on deck.
February 26, 2024Read original →
Proposal for a Multilateral Network of Public Good AI Research Labs to Protect Democracy and Humanity
I have been thinking about this question for several months: what if AI continues to progress towards and beyond our abilities in areas where it could become dangerous, and what if our regulations won't be 100% fullproof, opening the door to seriously harmful misuse by bad actors, historically never seen concentration…
September 12, 2023Read original →
Personal and Psychological Dimensions of AI Researchers Confronting AI Catastrophic Risks
On May 31st, 2023, a BBC web page headlined: "AI 'godfather' Yoshua Bengio feels 'lost' over life's work." However, this is a statement I never made, not to the BBC nor to any other media.
August 12, 2023Read original →
