Reduced prediction updating shapes serial dependence in autistic traits.

Researchers used a virtual reality soccer game to study how people with autistic traits make predictions. Participants had to guess where a ball would land after it disappeared mid-flight. People with higher autistic traits were less accurate at predicting where the ball would go and had trouble adjusting their predictions when given feedback. They stuck rigidly to their initial guesses even when they were clearly wrong, while others learned to be more flexible.

This virtual reality study examined how individuals with varying autistic traits update their predictions during a penalty-kick task where the ball's trajectory was artificially altered. Participants with higher autistic traits showed less accurate forward predictions and demonstrated reduced flexibility in updating their reliance on prior predictions when faced with feedback. Unlike those with lower autistic traits, they maintained rigid reliance on priors even when distortions were maximal. Classical stimulus and response biases remained unaffected, suggesting a specific impairment in prediction updating rather than general perceptual differences.

The findings indicate that autistic traits are associated with both mis-calibrated predictions and inflexible deployment of those predictions.

Findings suggest therapeutic approaches should target prediction flexibility rather than general perceptual training. May inform development of interventions focusing on adaptive updating of internal models. Could guide occupational therapy strategies for motor planning difficulties in autism.

Sample size not reported, limiting assessment of statistical power. Study design unclear from abstract. Virtual reality task may not generalize to real-world prediction scenarios. Cross-sectional design prevents causal inferences about autistic traits and prediction updating.

Serial dependence, the influence of prior experience on current perception or decision, has typically been studied in static, perceptual contexts. Here, we investigate whether serial dependence reflects not just passive carryover but feedback-based updating of internal models, and how this process varies with autistic traits. In an immersive virtual reality penalty-kick task, participants kicked a ball that disappeared mid-flight and estimated its landing position. By laterally displacing the ball upon reappearance, we introduced trial-by-trial prediction errors.

We found that individuals with higher autistic traits showed larger prediction deviations, indicating mis-calibrated forward predictions. At the same time, their responses were more strongly shaped by those priors, and unlike lower autistic traits individuals, they did not down-weight reliance when distortions were maximal. This pattern suggests reduced flexibility in updating prediction use: priors were both less accurate and more rigidly applied. Classical stimulus and response history biases were unaffected by autistic traits, highlighting a specific impairment in prediction updating.

Football experts, by contrast, combined low directional updating with near-zero prediction consistency, suggesting robust mappings that resist transient perturbations. These findings suggest that serial dependence in dynamic tasks reflects not only prediction formation but the flexible (or rigid) deployment of those predictions in the face of changing feedback. Our results highlight a distinctive rigidity in prediction weighting, rather than a general perceptual bias, in individuals with elevated autistic traits, and reveal contrasting stabilization strategies in domain experts.