Unveiling the Aha! moment: A computational account of insight in active inference
TL;DR
By modeling insight as Bayesian model reduction in a modified Wisconsin Card Sorting Task, it is shown that the model can reproduce a broader computational profile associated with insight, including discontinuous transitions in behavior and confidence, a transient peak in Bayesian surprise, and a phasic increase in expected precision.
Abstract
Insight, also referred to as the “Aha!” moment, is a distinctive cognitive phenomenon in which an idea or solution to a problem suddenly emerges into consciousness. Although the literature describes it extensively, the nature of its distinctive cognitive and affective mechanisms remains poorly understood. In this paper, we formalize a computational approach to insight grounded in the active inference framework. By modeling insight as Bayesian model reduction in a modified Wisconsin Card Sorting Task, we show that the model can reproduce a broader computational profile associated with insight, including discontinuous transitions in behavior and confidence, a transient peak in Bayesian surprise, and a phasic increase in expected precision. In light of those simulations, we interpret restructuring as Bayesian model reduction within the present modeling framework and impasse as a transient state of heightened uncertainty over possible actions. Finally, we propose that, during problem-solving, one regularly engages in phases of (generative) replay to generate and evaluate alternative models for subsequent model reduction.