Thomas Icard

Timeline

1. 2006–2007

   Middlebury and a Fulbright

   He graduates from Middlebury College in 2006 after majoring in Russian and philosophy. A 2006–2007 Fulbright award supports his study of formal semantics at the University of Amsterdam's Institute for Logic, Language and Computation.

2. August 2008

   Master of Logic in Amsterdam

   In August 2008, he defends his Master of Logic thesis, Topological Interpretations of the Polymodal Provability Logic, at the University of Amsterdam's Institute for Logic, Language and Computation. His supervisors are Lev Beklemishev and Dick de Jongh.

3. By 2014

   Doctorate and Carnegie Mellon fellowship

   By 2014, he completes a doctorate in Stanford's Department of Philosophy under Johan van Benthem, with the dissertation The Algorithmic Mind. In 2014, he is a Herbert Simon Postdoctoral Fellow in Carnegie Mellon University's Department of Philosophy.

4. September 2014

   Assistant professor at Stanford

   He joins Stanford University as an assistant professor in September 2014, with interests in logic, probability, and the foundations of cognitive science, especially language and reasoning.

5. 2020

   Logic for causal reasoning

   In 2020, he and Duligur Ibeling publish Probabilistic Reasoning Across the Causal Hierarchy at AAAI. They formalize association, intervention, and counterfactual reasoning as probabilistic logical languages.

6. 2022

   Causal distillation for language models

   In 2022, he coauthors Causal Distillation for Language Models at NAACL. The work trains a smaller language model to imitate the causal dynamics of a larger teacher model, with improvements on language understanding and question answering benchmarks.

7. 2024

   Moore scholar at Caltech

   In 2024, he serves as a Moore Distinguished Scholar at the California Institute of Technology.

8. 2024

   Distributed alignment search

   In 2024, he coauthors Finding Alignments Between Interpretable Causal Variables and Distributed Neural Representations at the Third Conference on Causal Learning and Reasoning. The paper introduces distributed alignment search, which uses gradient descent to align interpretable causal models with distributed neural representations.

9. 2025

   Causal abstraction for interpretability

   In 2025, he coauthors Causal Abstraction: A Theoretical Foundation for Mechanistic Interpretability in the Journal of Machine Learning Research. The work gives a common mathematical framework for methods that probe and change neural networks, including activation patching, sparse autoencoders, and steering.

10. As of October 2026

    Stanford professor and Goodfire researcher

    As of October 2026, he is Stanford's Clarence Irving Lewis Professor of Philosophy and Professor, by courtesy, of Computer Science. He spends 2026–2027 at Goodfire working full-time on the science of interpretability, and holds faculty affiliations with the Stanford AI Lab, Center for Automated Reasoning, and Symbolic Systems Program.