Bill Thompson
wdt@berkeley.eduGoogle Scholar
I am an Assistant Professor at the University of California, Berkeley and director of the Computational Cognitive Science Laboratory. I am Faculty in the Department of Psychology, the Program in Cognitive Science, and Berkeley's new Masters in Computational Social Science.
My research examines human cognition from a computational perspective, combining experimental psychology with machine learning and cognitive modeling to address basic questions about human learning and reasoning. I am particularly interested in the computational problems that arise in multi-agent intelligence and the solutions that human minds and societies have discovered: language, social cognition, and cultural evolution. I am also interested in the application of cognitive science to related problems in artificial intelligence.
Prior to UC Berkeley, I was a postdoc in the Department of Computer Science at Princeton University, working with Tom Griffiths. I did my PhD at the University of Edinburgh, working with Simon Kirby and Kenny Smith.
Representative Papers
You can read a recent overview of my research program and general perspective in this short essay, or watch this video:
An ever-evolving mind
Thompson, B. D. (2022). Science, 378(6620), 610–611.
My lab has specialized in large-scale experimental studies of distributed intelligence. We have examined the discovery and transmission of algorithmic knowledge in domains such as problem-solving and joint-action:
Complex cognitive algorithms preserved by selective social learning in experimental populations
Thompson, B. D., van Opheusden, B., Sumers, T., & Griffiths, T. L. (2022). Science, 376(6588), 95–98.
Preservation and loss of algorithmic structure in experimental diffusion of joint-action norms
Yu, D., & Thompson, B. D. (2026). arXiv:2610.24150.
These projects have repeatedly highlighted the limitations of traditional methods for induction of algorithmic structure from behavioral data, leading us to develop new AI-driven methods for scalable cognitive modeling of algorithmic behaviors, combining LLMs with probabilistic programing:
Hypothesis-guided discovery of cognitive algorithms via program refinement
Yang, H. A., Ho, M. K., & Thompson, B. D. (2026). arXiv:2610.02523.
I am interested in applying this mindset to problems that arise in modern social technologies, leveraging computational theories of human social intelligence to gain a deeper understanding of how people learn and reason in the context of e.g. social media environments and conversational AI systems.
Resampling reduces bias amplification in experimental social networks
Hardy, M. D., Thompson, B. D., Krafft, P. M., & Griffiths, T. L. (2023). Nature Human Behaviour, 7(12), 2084–2098.
I particulrly interested in the cultural evolution of conceptual systems. We have utilized advances in language modeling to shed light on this process and on human reasoning about culturally-transmitted concepts:
Cultural influences on word meanings revealed through large-scale semantic alignment
Thompson, B. D., Roberts, S. G., & Lupyan, G. (2020). Nature Human Behaviour, 4(10), 1029–1038.
Conceptual similarity as aggregation over feature sets in geometric spaces
Kaushik, K., & Thompson, B. D. (2026). Cognition, 266, Article 106302.
Words with more diverse semantic networks are more readily extended to novel meanings
Liu, Q., Lupyan, G., Srinivasan, M., & Thompson, B. D. (2025). In Proceedings of the Annual Meeting of the Cognitive Science Society.
I try to contribute to the mathematical literature on cultural evolution:
A resource-rational analysis of when (not) to trust your elders
Kaushik, K., Alister, M., & Thompson, B. D. (2026). In Proceedings of the Annual Meeting of the Cognitive Science Society.
Culture shapes the evolution of cognition
Thompson, B. D., Kirby, S., & Smith, K. (2016). Proceedings of the National Academy of Sciences, 113(16), 4530–4535.
And I have argued that human intelligence makes most sense when viewed in light of cultural evolution, as the consequence of distributed computations that arise at the population level, and that we now have better tools than ever to understand these processes:
How do humans overcome individual computational limitations by working together?
Vélez, N., Christian, B., Hardy, M., Thompson, B. D., & Griffiths, T. L. (2023). Cognitive Science, 47(1), Article e13232.
Using LLMs to advance the cognitive science of collectives
Sucholutsky, I., Collins, K. M., Jacoby, N., Thompson, B. D., & Hawkins, R. D. (2025). Nature Computational Science, 1–4.