Abstract
Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that remains poorly understood. Recent predictive coding theories suggest that atypical perceptual learning could play a central role in ASD. Here, we used two behavioral experiments to investigate prior learning and adjustment in ASD. In Experiment 1, 54 adults (31 neurotypical [NT] and 23 ASD) participated in a two-alternative forced-choice task where they had to compare the size of two discs. This task was designed to elicit a time-order effect, where perceptual decisions are biased toward the prior (i.e. mean disc size). The disc sizes followed either a narrow or broad distribution to elicit a strong or moderate prior bias, respectively. As expected, the NT group showed a stronger time-order effect in the narrow condition than in the broad condition. However, the ASD group did not show such difference between conditions, suggesting that ASD participants did not adapt the prior weight depending on the context. In Experiment 2, 51 adults (29 NT and 22 ASD) participated in an associative learning task. After hearing a high or low tone, they first had to predict whether they would see a clockwise or counterclockwise tilt, and then to report what they perceived. The cue and outcome were congruent in 62.5% of the trials and incongruent in 25% of the trials (unambiguous trials). In another 12.5% of the trials, there was no actual tilt (ambiguous trial). On average, both groups were able to predict the outcome above chance level, but contrary to NT, ASD participants did not adapt their predictions after a change in contingency. Finally, the two groups tended to perceive the ambiguous trials according to the current contingency. Altogether, these results suggest that individuals with ASD can learn priors, but are less flexible than NT to adjust their priors according to the context.