Reason to Play: Frontier LLM Reasoning Models Align with Human Brain Patterns During Game Learning

arXiv·medium signal

Compares behavioral and neural alignment between large reasoning models (LRMs) and human subjects learning novel games, finding that frontier LRMs exhibit learning curves and strategy patterns that correlate with fMRI-measured brain activation in human game learners. First quantified evidence that LRM reasoning trajectories parallel human cognitive processes during novel task acquisition. Implications for cognitive science and for understanding what 'reasoning' means in these models.