Bio

Yoshua Bengio is a French-born Canadian computer scientist, born March 5, 1964 in Paris, France. He is a full professor at the Université de Montréal in the Department of Computer Science and Operations Research, the founder and former scientific director of Mila (the Quebec Artificial Intelligence Institute), and a co-recipient of the 2018 ACM Turing Award (with Geoffrey Hinton and Yann LeCun) for foundational work on deep learning. He stepped down as scientific director of Mila in 2025 to launch LawZero, a non-profit AI-safety research organization, where he serves as co-president and scientific director. Bengio is one of the most-cited researchers in computer science globally and a senior public voice on AI-safety policy.

At a glance

• Education: Bachelor of Engineering in electrical engineering, McGill University (1986); Master of Science in computer science, McGill University (1988); PhD in computer science, McGill University (1991), advised by Renato De Mori. Postdoctoral fellow at MIT under Michael I. Jordan (1991 to 1992) and at AT&T Bell Laboratories (1992 to 1993).
• Current roles: Full professor at the Université de Montréal (since 1993); founder and senior research advisor at Mila; co-president and scientific director of LawZero (since June 2025); co-director of CIFAR's Learning in Machines and Brains program (since 2014).
• Key contributions: the 2003 neural probabilistic language model; the 2014 attention-based encoder-decoder for neural machine translation with Dzmitry Bahdanau and Kyunghyun Cho; the 2014 generative adversarial networks paper as a co-author with Ian Goodfellow; the 2016 textbook Deep Learning with Goodfellow and Aaron Courville; co-founder of Element AI (2016, sold to ServiceNow in 2020); founder of LawZero (2025).
• Awards: 2018 ACM Turing Award; Killam Prize (2019); Marie Victorin Prize of Quebec (2017); Officer of the Order of Canada (2017) and Companion (2022); Fellow of the Royal Society of London (2020); VinFuture Grand Prize (2024); Queen Elizabeth Prize for Engineering (2025); Princess of Asturias Award (2022).
• Wikipedia: Yoshua Bengio
• Personal site: yoshuabengio.org
• Mila profile: mila.quebec/en/directory/yoshua-bengio

Origins

Bengio was born on March 5, 1964 in Paris to Sephardic Jewish parents from Morocco who had moved to France. His family emigrated to Canada when he was a child, and he grew up in Montreal. His brother Samy Bengio is also a senior machine-learning researcher, formerly at Google Brain and now at Apple.

He studied at McGill University in Montreal, completing a Bachelor of Engineering in electrical engineering in 1986, a Master of Science in computer science in 1988, and a PhD in computer science in 1991, advised by Renato De Mori. The doctoral thesis, "Learning a Synaptic Learning Rule", combined neural networks with hidden Markov models and applied the resulting hybrid systems to speech recognition.

After completing his PhD, Bengio took a postdoctoral fellowship at MIT under Michael I. Jordan from 1991 to 1992, working on machine-learning theory, and a subsequent postdoctoral position at AT&T Bell Laboratories in Holmdel, New Jersey, in 1992 to 1993, where he overlapped with Yann LeCun, Léon Bottou, and Patrick Haffner. The Bell Labs period contributed to the 1998 LeNet-5 paper, "Gradient-Based Learning Applied to Document Recognition", on which Bengio is a co-author.

Career

Bengio joined the Université de Montréal as a faculty member in the Department of Computer Science and Operations Research in 1993 and has remained at the university for the entirety of his subsequent academic career. He held the Canada Research Chair in Statistical Learning Algorithms from 2000 onward and served as scientific director of the Montreal Institute for Learning Algorithms (the early acronym MILA, subsequently rebranded as Mila) from its founding in 1993, leading the institute through its 2017 reorganization into a multi-university Quebec collaboration spanning Université de Montréal, McGill University, Polytechnique Montréal, HEC Montréal, and other partner institutions.

The 2000 paper "A Neural Probabilistic Language Model", with Réjean Ducharme and Pascal Vincent, addressed the curse of dimensionality in language modeling and is widely cited as a foundational neural-network approach to language. The September 2014 paper "Neural Machine Translation by Jointly Learning to Align and Translate", with Dzmitry Bahdanau and Kyunghyun Cho, introduced soft-attention as a learned alignment mechanism between encoder and decoder. The attention mechanism, generalized in subsequent work, is the conceptual ancestor of the self-attention used in the 2017 transformer paper. The June 2014 paper "Generative Adversarial Nets" with Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, and Aaron Courville introduced the GAN framework that became one of the dominant generative-modeling approaches across image, video, and other modalities through the late 2010s.

In October 2016, Bengio co-founded Element AI, a Montreal-based AI consultancy and incubator, with Jean-François Gagné and others. The company raised more than CAD 200 million from institutional investors and was acquired by ServiceNow in November 2020. The 2016 textbook Deep Learning, written with Ian Goodfellow and Aaron Courville and published by MIT Press, became one of the most widely used graduate-level reference texts in modern machine learning.

Through the late 2010s and early 2020s, Bengio's research focus expanded from foundational deep-learning research to include AI safety, AI alignment, and AI for socially beneficial applications including climate, health, and education. He served as a co-director of CIFAR's Learning in Machines and Brains program with Yann LeCun from 2014 onward, and chaired the production of the inaugural International AI Safety Report commissioned at the November 2023 Bletchley Park summit and published in January 2025.

In June 2025, Bengio launched LawZero, a non-profit AI-safety research organization headquartered in Montreal, where he serves as co-president and scientific director. The organization's stated mission is to develop "Scientist AI", a non-agentic, trustworthy AI system designed to understand and explain phenomena rather than act as an autonomous agent, and to study guardrails that detect and block harmful behavior in agentic systems. Funders include the Future of Life Institute and Schmidt Sciences. He stepped down as scientific director of Mila in 2025 to focus on LawZero, transitioning the Mila scientific-director role to Hugo Larochelle while remaining as founder and senior advisor.

Affiliations

• MIT: Postdoctoral fellow under Michael I. Jordan, 1991 to 1992.
• AT&T Bell Laboratories: Postdoctoral researcher, 1992 to 1993.
• Université de Montréal: Faculty, 1993 to present (full professor; Canada Research Chair in Statistical Learning Algorithms 2000 to present).
• Mila: Founder, 1993; scientific director through 2025; founder and senior advisor since 2025.
• CIFAR: Co-director, Learning in Machines and Brains program, since 2014.
• Element AI: Co-founder, October 2016; acquired by ServiceNow in November 2020.
• LawZero: Co-president and scientific director, since June 2025.

Notable contributions

Bengio's contributions span foundational deep-learning research, generative modeling, language modeling, and AI safety policy.

• Neural probabilistic language model (2003). "A Neural Probabilistic Language Model" with Réjean Ducharme and Pascal Vincent. Foundational neural approach to statistical language modeling, addressing the curse of dimensionality through learned distributed word representations.
• Stacked denoising autoencoders and unsupervised pre-training (2007 to 2010). A research thread with Pascal Vincent and others on unsupervised representation learning, which alongside Hinton's deep belief networks contributed to the late-2000s revival of deep neural networks.
• Attention mechanism for neural machine translation (September 2014). "Neural Machine Translation by Jointly Learning to Align and Translate" with Dzmitry Bahdanau and Kyunghyun Cho introduced soft-attention as a learned alignment between encoder and decoder. The mechanism is a conceptual ancestor of the self-attention in the 2017 transformer paper.
• Generative adversarial networks (June 2014). Co-author of "Generative Adversarial Nets" with Ian Goodfellow and others. The GAN framework became one of the dominant generative-modeling approaches in computer vision and adjacent domains.
• Curriculum learning, sequence-to-sequence translation, word embeddings, denoising autoencoders. Sustained research output across multiple foundational deep-learning research threads.
• 2016 textbook Deep Learning with Ian Goodfellow and Aaron Courville, MIT Press. One of the most widely used graduate-level deep-learning reference texts.
• Element AI co-founding (October 2016). Co-founder of the Montreal-based AI consultancy and incubator, acquired by ServiceNow in November 2020.
• Mila institutional leadership (1993 to 2025). Founder and long-term scientific director of the Quebec Artificial Intelligence Institute, which became one of the largest concentrations of academic AI research talent globally.
• 2018 ACM Turing Award, shared with Geoffrey Hinton and Yann LeCun "for conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing."
• AI-safety advocacy and policy (2023 onward). Bengio's signature on the March 2023 Future of Life Institute open letter calling for a six-month pause on advanced AI training, the May 2023 Statement on AI Risk of Extinction signed alongside leading AI researchers, and the chairing of the International AI Safety Report commissioned at Bletchley.
• LawZero founding (June 2025). Co-president and scientific director of the AI-safety research nonprofit, focused on Scientist AI and safety guardrails for autonomous systems.

Honors

• Officer of the Order of Canada (2017) and Companion of the Order of Canada (2022).
• Fellow of the Royal Society of Canada (2017) and Fellow of the Royal Society of London (2020).
• Marie Victorin Prize of Quebec (2017); Killam Prize in Engineering (2019); 2018 ACM Turing Award, shared with Geoffrey Hinton and Yann LeCun.
• Princess of Asturias Award for Technical and Scientific Research (2022), shared with Yann LeCun, Geoffrey Hinton, and Demis Hassabis.
• VinFuture Grand Prize (2024), shared with Yann LeCun, Geoffrey Hinton, Jen-Hsun Huang, and Fei-Fei Li.
• Queen Elizabeth Prize for Engineering (2025), shared with Geoffrey Hinton, Yann LeCun, John Hopfield, Bill Dally, Jen-Hsun Huang, and Fei-Fei Li.
• Most-cited computer scientist globally by total citations and h-index per Google Scholar (with citation counts exceeding 700,000 across his career).

Position in the field

Bengio occupies a distinctive position among the senior figures of modern deep learning. The combination of a forty-year academic career at a single institution, the founding of Mila, the most-cited record in computer science, the 2018 Turing Award, and the post-2023 leadership of the International AI Safety Report and LawZero is structurally distinct from the operator-executive trajectory taken by Yann LeCun at Meta and the Google-and-emeritus trajectory taken by Geoffrey Hinton at University of Toronto.

The Mila lineage is itself a substantial element of his influence. The institute hosts more than 140 affiliated professors and has produced senior researchers across Google DeepMind, Anthropic, OpenAI, Meta AI, Cohere, and academic positions globally. The 2014 attention paper with Bahdanau and Cho is cited in the lineage of the 2017 transformer paper, and Bengio's PhD students and postdoctoral collaborators include Ian Goodfellow (GAN inventor; subsequently at OpenAI, Google Brain, Apple, and Google DeepMind), Aaron Courville, Pascal Vincent, Dzmitry Bahdanau, Kyunghyun Cho, Hugo Larochelle (Mila scientific director and former Google Brain Montreal lead), and Joelle Pineau (former Meta AI FAIR senior leadership; subsequently Cohere).

The post-2023 AI-safety positioning has placed Bengio as one of the most prominent senior research voices in favor of strict regulatory and technical safeguards on advanced AI, alongside Geoffrey Hinton and Stuart Russell, and in contrast to the more skeptical-of-doom posture of Yann LeCun. The launch of LawZero formalizes a working position rather than purely academic commentary on AI risk.

Outlook

Open questions and watchable signals over the next 6 to 18 months:

• LawZero research output. Whether Bengio's nonprofit produces concrete technical artifacts, papers, and prototypes that validate the Scientist AI thesis and the autonomous-agent guardrail approach.
• Mila trajectory under Hugo Larochelle. The continued evolution of the institute under the post-Bengio scientific direction, with sustained Quebec government funding and senior-faculty recruiting.
• AI-safety policy engagement. Bengio's continued role in the international AI safety summit process and any subsequent commissioned reports succeeding the January 2025 International AI Safety Report.
• Public commentary trajectory. Bengio's commentary on AI capability progression, particularly any updates on the probability assessments of catastrophic AI outcomes, and the comparison with peer commentary from Geoffrey Hinton, Stuart Russell, and the contrarian voice from Yann LeCun.
• Student-and-collaborator trajectories. The continued progression of senior researchers from the Mila lineage, particularly those at Anthropic, Google DeepMind, and Cohere where Mila alumni hold senior research-leadership roles.

Sources

• Yoshua Bengio. Wikipedia biographical entry covering family, education, career, and publications.
• Yoshua Bengio personal site. Yoshua Bengio's official site with publication list, blog posts, and current activities.
• Yoshua Bengio Mila profile. Mila faculty profile and research overview.
• Yoshua Bengio ACM Turing Award page. ACM citation and biographical detail for the 2018 Turing Award.
• A Neural Probabilistic Language Model. The 2003 JMLR paper with Réjean Ducharme and Pascal Vincent on neural language modeling.
• Neural Machine Translation by Jointly Learning to Align and Translate. The September 2014 paper with Dzmitry Bahdanau and Kyunghyun Cho introducing the attention mechanism for neural machine translation.
• Generative Adversarial Nets. The June 2014 NeurIPS paper with Ian Goodfellow and others introducing GANs.
• Deep Learning textbook. The 2016 MIT Press textbook with Ian Goodfellow and Aaron Courville.
• Introducing LawZero. The June 2025 announcement of the LawZero AI-safety nonprofit.
• Yoshua Bengio launches LawZero. LawZero's launch announcement and mission statement.
• International AI Safety Report 2025. The January 2025 report chaired by Bengio and commissioned at Bletchley.
• Element AI. Wikipedia entry on the Bengio-co-founded AI company acquired by ServiceNow in November 2020.
• Turing Award honours CIFAR's pioneers of AI. CIFAR's record of the Hinton-Bengio-LeCun lineage and the Learning in Machines and Brains program.