Item

Abstract

Objective: A key characteristic of autism is a limited ability to understand others’ mental states. However, what specific neuro-cognitive mechanisms underlie these difficulties in mentalization remains poorly understood, and findings regarding neural correlates are highly mixed. Here, we identify a link between autistic traits and neurocomputational processes that dynamically adapt behavior to changing strategic thoughts of others. Methods: Adult participants (47 autistic and 47 neurotypical matched for age, sex, and education) played repeated Rock-Paper-Scissors games against artificial opponents of varying mentalization depth (of the type “I think that you think that…”). The opponents were carefully calibrated in previous work to mimic human gameplay and were based on a new computational model that was also used to capture participants’ strategizing (Buergi et al. 2024). This model assumes that agents try to infer the strategy of their opponent by mentally simulating what different opponents would play, and comparing these predictions to observed behavior. The resulting belief updates allow adapting to the reasoning process of the opponent and were robustly linked to a multivariate neural action pattern (Buergi et al. 2024). Here, we leverage this pattern as a normative benchmark for comparison between autistic and neurotypical participants. Results: As preregistered, we found that autistic traits are linked to reduced performance in the game (standardized β = -0.26, p = .005). Model-based analyses suggest that this resulted from a decreased sensitivity to the way others reason (β = -0.25, p = .020), leading to reduced belief updates. Analysis of concurrently acquired fMRI data did not reveal altered activation in any single area, but confirmed a reduced expression of the previously-identified neural pattern with higher autistic traits (β = 0.35, p = .003). Both reduced sensitivity and altered neural activation patterns mediated the effect of autistic traits on task performance, corroborating their role in explaining the observed differences. Conclusions: We provide a mechanistic account of mentalizing difficulties in autistic people following a preregistered analysis protocol. Our results characterize a specific neurocomputational process that underlies mentalization ability, is implemented by distributed activity across the brain, and is changed in autism. Our findings inform theories about the neurocomputational basis of mentalization and may aid in the detection and assessment of corresponding behavioral problems.