# Sparks of In Silico Cognitive Science: Theories from Simulated Data Can Generalize to Humans

arXiv: https://arxiv.org/abs/2609.08003 · HTML: https://mlsanity.com/paper/2609.08003/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 8 of its 15 references that are in the map.
- First posted on arXiv 7 Sep 2026.

## Builds on (most cited first, 8 of 8)

- [AlphaEvolve: A coding agent for scientific and algorithmic discovery](https://mlsanity.com/paper/2506.13131/) (2025)
- [Survey of multifidelity methods in uncertainty propagation, inference, and optimization](https://mlsanity.com/paper/1806.10761/) (2018)
- [Generating Computational Cognitive Models using Large Language Models](https://mlsanity.com/paper/2502.00879/) (2025)
- [Closing the Loop to Discover Psychological Theories with an Automated Cognitive Scientist](https://mlsanity.com/paper/2606.26448/) (2026)
- [Not Yet AlphaFold for the Mind: Evaluating Centaur as a Synthetic Participant](https://mlsanity.com/paper/2508.07887/) (2025)
- [Toward an automated science of the mind](https://mlsanity.com/paper/2603.20988/) (2026)
- [auto-psych: Automating the science of mind using agent-driven theory discovery and experimentation](https://mlsanity.com/paper/2606.26460/) (2026)
- [Position: RL Researchers Need to Distinguish Between Solving Simulators and Using Simulators as a Proxy](https://mlsanity.com/paper/2606.28433/) (2026)

## Abstract (arXiv)

Behavioral foundation models have been proposed as stand-ins for human participants across settings, but it is unclear whether theories discovered on them generalize to humans or merely characterize the simulator. We ran the Automated Cognitive Scientist (textscAutoCog), a closed-loop discovery system in which LLM agents design theory-discriminating experiments, collect responses, arbitrate between competing theories, and synthesize successors, entirely on behavior simulated by Centaur, a foundation model of human behavior. In a multi-attribute decision-making setting, the theories textscAutoCog found on Centaur generalized to human data: they outperformed canonical theories on ten held-out experiments and were rivaled only by theories found by running the same loop on people. We argue that this succeeds despite the simulator's inevitable imperfections because a discovery loop that arbitrates between competing theories demands less of its simulator than estimation does. The simulator only needs to capture the regularities that distinguish the theories, and not necessarily reproduce behavior precisely. Imperfect simulators can therefore widen the search over theories, with human data then testing whether the surfaced theories generalize.