Diametrically opposed sex-specific effects of autistic traits on risk-taking in poker.

Publication date: Sep 01, 2026

Autism spectrum disorder (ASD) clinical manifestations diverge significantly by sex, often leading to diagnostic delays in females. While social factors like camouflaging are recognized, the underlying computational mechanisms remain elusive. We investigated whether strategic decision-making under incomplete information reflects these sex-specific phenotypes through a Bayesian framework. Twenty poker-naive adults (9 males, 11 females) with varying ASD traits played 100 hands of Texas Hold’em against a poker program. We employed generalized linear mixed models (GLMMs) incorporating interactions between sex and the Autism Diagnostic Observation Schedule, Second Edition (ADOS-2), while controlling for full-scale intelligence quotient (IQ), anxiety, and depression as covariates. We identified a significant Sex cD7 ADOS-2 interaction. Males with higher ASD traits exhibited “aggressive rigidity”, characterized by increased betting and high-risk bluffing (heavy investment with weak hands), reflecting a prior over-precision that ignores negative feedback. Conversely, high-trait females demonstrated “avoidant rigidity”, marked by extreme risk-aversion and under-betting even with strong hands. While higher IQ facilitated adaptive strategic resets (defensive folding after major losses) in females, this compensatory effect was neutralized in high-trait males. These strategic biases remained independent of general anxiety and depression throughout the session, revealing a covert computational internalizing phenotype. This pilot study points toward a computational dimorphism: high-magnitude aggression in males mirrors visible externalizing symptoms, whereas low-magnitude avoidance in females aligns with covert internalizing symptoms. Identifying extreme risk-aversion as a characteristic feature of the female ASD phenotype provides a novel, computationally informed framework to improve the identification of overlooked individuals.

Semantics

Original Article