PeopleofSuperintelligence

Researcher

Michael Dennis

Michael Dennis is a research scientist on Google DeepMind's Open-Endedness team who works on automatically generating the environments in which AI agents learn. He spent 2016 to 2023 as a PhD student at UC Berkeley's Center for Human-Compatible AI, advised by Stuart Russell, and in 2020 he was first author of the PAIRED paper that proposed Unsupervised Environment Design. At Google DeepMind, he coauthored Genie, an 11-billion-parameter world model trained from unlabelled Internet videos, which received a Best Paper award at ICML 2024, and he contributed to Genie 2 in 2024. His research interests are open-endedness, unsupervised environment design and AI safety.

Badges

Timeline

  1. 2016–2023

    PhD at Berkeley's CHAI

    From 2016 to 2023, he is a PhD student in artificial intelligence and computational geometry at UC Berkeley, advised by Stuart Russell at the Center for Human-Compatible AI. He comes to the center from theoretical computer science. In November 2022, he gives the dissertation talk "On Problem Specification and Self-Referential Claims", sponsored by Berkeley's Department of Electrical Engineering and Computer Sciences.

  2. 2019

    Adversarial policies against RL agents

    In 2019, he coauthors "Adversarial Policies: Attacking Deep Reinforcement Learning" with Adam Gleave, Cody Wild, Neel Kant, Sergey Levine and Stuart Russell. The paper shows that adversarial policies can win against self-play-trained victims in zero-sum games between simulated humanoid robots, and it is presented at ICLR 2020.

  3. 2020

    PAIRED and Unsupervised Environment Design

    In 2020, he is first author of "Emergent Complexity and Zero-shot Transfer via Unsupervised Environment Design", a NeurIPS paper written with Natasha Jaques, Eugene Vinitsky, Alexandre Bayen, Stuart Russell, Andrew Critch and Sergey Levine. The paper proposes Unsupervised Environment Design and introduces PAIRED, in which an adversary generates environments that maximize the regret between a protagonist and an antagonist agent.

  4. 2021–2022

    Replay-guided design and ACCEL

    In 2021, he coauthors "Replay-Guided Adversarial Environment Design" with Minqi Jiang, Jack Parker-Holder, Jakob Foerster, Edward Grefenstette and Tim Rocktäschel; the paper casts Prioritized Level Replay as Unsupervised Environment Design and names the Dual Curriculum Design class of methods. In 2022, he is a coauthor of "Evolving Curricula with Regret-Based Environment Design", which introduces ACCEL, a method that edits levels to keep them at the frontier of an agent's capabilities.

  5. 2023

    Joining DeepMind's Open-Endedness team

    In 2023, he becomes a research scientist on the Open-Endedness team at Google DeepMind. In December 2023, he is among the organizers of the Agent Learning in Open-Endedness (ALOE) workshop at NeurIPS.

  6. 2024

    Genie and an ICML best paper

    In February 2024, he and colleagues including Jake Bruce, Ashley Edwards and Jack Parker-Holder publish "Genie: Generative Interactive Environments", an 11-billion-parameter foundation world model trained from unlabelled Internet videos that turns text, images and sketches into action-controllable virtual worlds. Genie receives a Best Paper award at ICML 2024.

  7. June–December 2024

    Open-endedness paper and Genie 2

    In June 2024, he coauthors the position paper "Open-Endedness is Essential for Artificial Superhuman Intelligence" with Edward Hughes, Jack Parker-Holder, Tim Rocktäschel and others, which argues that open-endedness is an essential property of any artificial superhuman intelligence. In December 2024, Google DeepMind introduces Genie 2, a world model that generates rich 3D worlds, and names him among its contributors.

  8. As of October 2026

    Research scientist at Google DeepMind

    In February 2025, he is one of the coauthors of the report "Multi-Agent Risks from Advanced AI", which sets out three failure modes of multi-agent systems: miscoordination, conflict and collusion. As of October 2026, he is a research scientist at Google DeepMind on the Open-Endedness team, with research interests in open-endedness, unsupervised environment design and AI safety.

Career

  1. Since 2023Google DeepMindResearch scientist

Education

  1. 2023University of California, BerkeleyPhD

From the biography.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Michael Dennis. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  2. 16% of AI career moves in the 2020s went from a university to a companyIncludes Michael Dennis. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  3. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  4. Google has the highest combined h-index among AI organizationsIncludes Google. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Michael Dennis yet. Send us a free-licensed photo.

  1. Our People – Center for Human-Compatible Artificial Intelligence humancompatible.ai. Checked 21 October 2026.
  2. Michael Dennis | alphaXiv alphaxiv.org. Checked 21 October 2026.
  3. Dissertation Talk: On Problem Specification and Self-Referential Claims | CDSS at UC Berkeley cdss.berkeley.edu. Checked 21 October 2026.
  4. Adversarial Policies: Attacking Deep Reinforcement Learning arxiv.org. Checked 21 October 2026.
  5. Emergent Complexity and Zero-shot Transfer via Unsupervised Environment Design arxiv.org. Checked 21 October 2026.
  6. Emergent Complexity and Zero-Shot Transfer via Unsupervised Environment Design | Institute of Transportation Studies its.berkeley.edu. Checked 21 October 2026.
  7. Replay-Guided Adversarial Environment Design arxiv.org. Checked 21 October 2026.
  8. Evolving Curricula with Regret-Based Environment Design arxiv.org. Checked 21 October 2026.
  9. ALOE 2023 sites.google.com. Checked 21 October 2026.
  10. ICML 2024 Awards icml.cc. Checked 21 October 2026.
  11. Genie: Generative Interactive Environments — Google DeepMind deepmind.google. Checked 21 October 2026.
  12. Open-Endedness is Essential for Artificial Superhuman Intelligence arxiv.org. Checked 21 October 2026.
  13. Genie 2: A large-scale foundation world model — Google DeepMind deepmind.google. Checked 21 October 2026.
  14. Multi-Agent Risks from Advanced AI arxiv.org. Checked 21 October 2026.
  15. Michael Dennis - Google Scholar scholar.google.com. Checked 21 October 2026.