David Kristjanson Duvenaud (born 1983) is a Canadian computer scientist at the University of Toronto specializing in probabilistic machine learning, generative AI, and AI safety. He is a CIFAR AI chair since 2021 and a Schwartz Reisman Chair in Technology and Society since 2024. His research spans deep generative models, neural ordinary differential equations, and more recently the societal implications of advanced artificial intelligence.
Education
Duvenaud was born in suburban Winnipeg, Manitoba. He received a B.Sc. in computer science from the University of Manitoba in 2006 and an M.Sc. in computer science from the University of British Columbia in 2010, supervised by Kevin P. Murphy. He earned a PhD from the University of Cambridge in 2014 under the supervision of Carl Rasmussen and Zoubin Ghahramani. After a postdoctoral position at Harvard University in the group of Ryan P. Adams, he joined the faculty of the University of Toronto in 2016.
Career and Research
Duvenaud is an associate professor in computer science at the University of Toronto and a founding faculty member at the Vector Institute, a leading Canadian institute for artificial intelligence research. His early research focused on deep generative models, including generative models of molecules and neural ordinary differential equations. His group introduced a type of continuous-time normalizing flow, which received the NeurIPS best paper award in 2018. This work contributed to the development of neural networks that can model continuous-time dynamics, with applications in machine learning and deep learning.
In 2023, responding to the rapid adoption of large language models, Duvenaud shifted his research focus to AI safety. He took a sabbatical at Anthropic, where he led the Alignment Evaluations team. There, he worked on identifying jailbreaks and developing evaluations for models' ability to surreptitiously conduct sabotage. His recent work addresses the concept of "gradual disempowerment": as AI becomes more capable, large-scale societal systems such as economies, cultures, and states may progressively cease to serve human interests. This is because the alignment of these systems with human needs has historically depended on their reliance on human participation, which AI increasingly renders unnecessary.
Contributions to AI Safety
Duvenaud's pivot to AI safety reflects a broader concern within the generative AI community about the long-term risks of advanced systems. His work at Anthropic contributed to the development of evaluation methods for detecting undesirable behaviors in transformer-based models. He has also written about the potential for AI to gradually erode human agency, a theme that connects to ongoing debates about alignment and RLHF in the field.
Recognition and Affiliations
Duvenaud holds the CIFAR AI Chair since 2021 and the Schwartz Reisman Chair in Technology and Society since 2024. He is affiliated with the University of Toronto and the Vector Institute. His research has been recognized with the 2018 NeurIPS best paper award. He is a frequent contributor to conferences such as NeurIPS, ICML, and ICLR, and his work is widely cited in the machine learning community.
Selected Works and Impact
Duvenaud's early work on neural ordinary differential equations has influenced subsequent research in continuous-time models and residual networks. His papers on generative models of molecules have applications in drug discovery and materials science. More recently, his AI safety research has informed discussions on how to evaluate and mitigate risks from advanced AI systems, including those deployed in OpenAI and other organizations. As of 2025, he continues to teach and supervise students at the University of Toronto, focusing on the intersection of probabilistic modeling and societal impact.