Inside insight: decoding how insight emerges from competing world models

Abstract

When and how does insight emerge? We conceptualize insight as a sudden realization arising from restructuring a world model: an internal interpretation linking actions to outcomes. Yet these latent dynamics remain difficult to access, even with behavior and verbal report. Here we developed inside insight dynamics (IID), a machine-learning framework that estimates latent world-model dynamics from behavioral data. Using IID, we analyzed mouse behavior in indirect- and direct-rule tasks, each requiring a shift from an initial world model to a rule-consistent representation. IID inferred the "when" of insight-like shifts by estimating the timing of transitions between competing world models, and examined the "how" by comparing alternative learning processes underlying them. This analysis revealed distinct mechanisms of world-model learning: the indirect- and direct-rule tasks were better explained by gated learning and parallel learning, respectively. Thus, IID opens a route to quantifying latent insight dynamics from observable behavior alone.