Latent Cause Inference As a Novel Explanation for the Neurocomputational Mechanisms of Compulsion

Abstract

Compulsivity is an important transdiagnostic symptom in several psychological conditions like OCD and substance use disorders. Prominent literature on compulsivity supports a habitual control theory, where compulsivity is a result of a deficiency in goal-directed behavior. Current research suggests that the brain is over-relying on a ‘model-free’ learning mechanism, which is insensitive to goal value change and leads to the formation of inflexible habits. However, habitual control fails to adequately explain the seemingly goal-directed behavior present in observational accounts of compulsivity disorders, such as cravings, obsessions, and ruminations. This study proposes a novel explanation of compulsivity derived from the latent cause framework. The latent cause model theorizes that people learn associations between a stimulus and a response within a specific context (‘cause’). Failure to generalize information between causes can also lead to behavioral inflexibility. This may be a better neurocomputational explanation for compulsivity, and apparent model-free behavior may actually represent aberrant latent cause inference. We modified the Daw et al. (2011) two-step task, which has been used in the past to study putative model-free behavior and its relationship to compulsivity, to explicitly induce latent cause inference and found increased model-free behavior as well as increased susceptibility to state-splitting. Additionally, we found a weaker association between compulsivity and reduced goal-oriented behavior in the modified task, suggesting that the relationship is modulated by a different type of learning. We contribute novel insights on the neural computations of decision-making behavior and the relationship with certain symptom dimensions in the hopes that the discovery of a different mechanism of compulsion can enrich the literature of these psychiatric disorders and improve current therapeutic interventions.