Contextual priors do not modulate action prediction in children with autism

Area of Science:

• Neuroscience
• Developmental Psychology
• Cognitive Science

Background:

• Autism Spectrum Disorder (ASD) is increasingly understood through Bayesian computational models.
• These models propose that ASD may involve difficulties in estimating the probability of future events, potentially due to altered prior beliefs.

Purpose of the Study:

• To investigate the hypothesis that autistic individuals have an impaired ability to use contextual priors for action prediction.
• To examine how autistic children utilize probabilistic information from action-context associations when faced with perceptual ambiguity.

Main Methods:

• A behavioral paradigm involving familiarization and testing phases was used with children with and without autism.
• During familiarization, action-context associations were implicitly biased by their probability of co-occurrence.
• During testing, shortened action videos (ambiguous kinematics) required inference based on learned priors.

Main Results:

• Typically developing children showed a probabilistic effect, with responses biased by contextual priors to resolve kinematic ambiguity.
• Autistic children did not exhibit this modulation, indicating a failure to use contextual priors effectively.
• This suggests a specific impairment in leveraging prior expectations for predicting actions in uncertain situations.

Conclusions:

• The findings support Bayesian accounts of autism, highlighting a potential deficit in using prior probabilities for action prediction.
• Autistic children appear to have difficulty integrating contextual information to predict ongoing actions when sensory input is ambiguous.
• This impairment may contribute to social interaction challenges in autism spectrum disorder.