Can Large Language Models Simulate Human Cognition Beyond Behavioral Imitation?
For AI researchers, this work provides a first-stage empirical study on LLMs' cognitive simulation capabilities, though the findings are incremental due to the benchmark's limited scope and low performance.
This paper investigates whether LLMs can simulate human cognition beyond behavioral imitation, introducing a benchmark based on 217 researchers' publication trajectories. Results show current LLMs achieve limited cognitive alignment, with best models reaching only 0.35 alignment score, indicating significant room for improvement.
An essential problem in artificial intelligence is whether LLMs can simulate human cognition or merely imitate surface-level behaviors, while existing datasets suffer from either synthetic reasoning traces or population-level aggregation, failing to capture authentic individual cognitive patterns. We introduce a benchmark grounded in the longitudinal research trajectories of 217 researchers across diverse domains of artificial intelligence, where each author's scientific publications serve as an externalized representation of their cognitive processes. To distinguish whether LLMs transfer cognitive patterns or merely imitate behaviors, our benchmark deliberately employs a cross-domain, temporal-shift generalization setting. A multidimensional cognitive alignment metric is further proposed to assess individual-level cognitive consistency. Through systematic evaluation of state-of-the-art LLMs and various enhancement techniques, we provide a first-stage empirical study on the questions: (1) How well do current LLMs simulate human cognition? and (2) How far can existing techniques enhance these capabilities?