PeopleofSuperintelligence

Researcher

Jonathan Richens

Also known as Jon Richens

Jonathan Richens is a senior research scientist at Google DeepMind whose theoretical work asks what an AI agent must learn about its environment to generalize and stay robust under distributional shifts. Trained in quantum information theory at Imperial College London, he moved into machine learning at Babylon Health, where a 2020 Nature Communications paper reformulated medical diagnosis as a counterfactual inference task. His later papers include Counterfactual Harm at NeurIPS 2022, Discovering Agents with Tom Everitt and others, and Robust agents learn causal world models, which received an Honorable Mention at ICLR 2024. In 2025, he was first author of General agents need world models, with David Abel, Alexis Bellot and Tom Everitt, which showed that any agent capable of generalizing to multi-step goal-directed tasks must have learned a predictive model of its environment. He works on agent foundations, causality and their application to alignment and safety.

Timeline

  1. By 2020

    Quantum information at Imperial

    By 2020, he has completed a PhD in quantum information theory at Imperial College London and moved into machine learning research.

  2. 2020

    Counterfactual diagnosis at Babylon

    As a senior researcher at Babylon Health, he is first author, with Ciarán M. Lee and Saurabh Johri, of a 2020 Nature Communications paper that reformulates diagnosis as a counterfactual inference task. On a set of clinical vignettes, the counterfactual algorithm places in the top 25% of 44 doctors and the standard associative algorithm in the top 48%.

  3. 2022

    Counterfactual Harm at NeurIPS

    He is first author, with Rory Beard and Daniel H. Thompson, of Counterfactual Harm, accepted at NeurIPS 2022. The paper proposes a formal definition of harm and benefit using causal models and shows that standard machine learning algorithms that cannot perform counterfactual reasoning are guaranteed to pursue harmful policies following distributional shifts.

  4. 2022–2023

    A causal definition of agents

    In 2022, he coauthors Discovering Agents with Zachary Kenton, Ramana Kumar, Sebastian Farquhar, Matt MacDermott and Tom Everitt. The paper proposes a formal causal definition of agents and a causal discovery algorithm for finding agents in empirical data, and appears in the journal Artificial Intelligence in 2023.

  5. 2024

    An ICLR honorable mention

    With Tom Everitt, he shows in Robust agents learn causal world models that any agent satisfying a regret bound under a large set of distributional shifts must have learned an approximate causal model of the data-generating process. The paper receives an Honorable Mention outstanding paper award at ICLR 2024.

  6. 2025

    General agents and world models

    In 2025, he is first author of General agents need world models, published at ICML with David Abel, Alexis Bellot and Tom Everitt. It shows that any agent capable of generalizing to multi-step goal-directed tasks must have learned a predictive model of its environment. He also coauthors The Limits of Predicting Agents from Behaviour with Alexis Bellot and Tom Everitt, which derives bounds on an agent's behaviour in unseen deployment environments.

  7. As of October 2026

    Senior research scientist at DeepMind

    He is a senior research scientist at Google DeepMind, working on agent foundations, causality and their application to alignment and safety. For the Fall 2026 round of SPAR, he mentors a project on what can be learned about agents with causal identifiability.

Career

  1. NowGoogle DeepMindSenior research scientist
  2. From 2020Babylon HealthSenior researcher

Education

  1. Imperial College LondonPhD in quantum information theory

From the biography.

  1. 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.
  2. 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.
  3. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Google. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.
  4. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Google. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.

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 Jonathan Richens yet. Send us a free-licensed photo.

  1. Improving the accuracy of medical diagnosis with causal machine learning | Events | Imperial College London imperial.ac.uk. Checked 21 October 2026.
  2. AIMI Journal Club: Improving the Accuracy of Medical Diagnosis With Causal Machine Learning - Jonathan Richens, PhD | Center for Artificial Intelligence in Medicine & Imaging aimi.stanford.edu. Checked 21 October 2026.
  3. Improving the accuracy of medical diagnosis with causal machine learning pmc.ncbi.nlm.nih.gov. Checked 21 October 2026.
  4. [2204.12993] Counterfactual harm arxiv.org. Checked 21 October 2026.
  5. Introduction | Causal Incentives Working Group causalincentives.com. Checked 21 October 2026.
  6. [2208.08345] Discovering Agents arxiv.org. Checked 21 October 2026.
  7. ICLR 2024 Awards iclr.cc. Checked 21 October 2026.
  8. [2402.10877] Robust agents learn causal world models arxiv.org. Checked 21 October 2026.
  9. [2506.01622] General agents contain world models arxiv.org. Checked 21 October 2026.
  10. [2506.02923] The Limits of Predicting Agents from Behaviour arxiv.org. Checked 21 October 2026.
  11. Formalizing what can and cannot be learned about agents with (causal) identifiability - SPAR Project sparai.org. Checked 21 October 2026.