_individuals with ASD make less use of prior knowledge, so that they see things as they are rather than as how they expect them to be_
“Bayesian models of ASD similarly argue that “top-down” processing distorts how individuals with ASD integrate sensory information (Pellicano & Burr, 2012). For example, Pellicano and Burr (2012) proposed that individuals with ASD make less use of prior knowledge, so that they see things as they are rather than as how they expect them to be. Similarly, inflexibility in predictive processes (e.g., predictive coding) are said to disrupt perceptual inferences required to recognize objects and events (Sinha et al., 2014; van Boxtel & Lu, 2013; Van de Cruys et al., 2014), possibly by giving greater weight to sensations than to contextual cues and learned interpretations (Karvelis, Seitz, Lawrie, & Series, 2018; Lawson, Rees, & Friston, 2014; Palmer, Lawson, & Hohwy, 2017).
Bayesian models would seem to predict that children with ASD should show advantages at learning new perceptual categories, especially when no feedback is available, and that learning abstract categories should be more difficult. In contrast, HF children with ASD who show no apparent deficits in academic coursework sometimes show profound deficits in learning to categorize novel, abstract shapes (Church et al., 2010; Church et al., 2015). Additionally, theories that suggest children with ASD rely less on learned Bayesian priors (top-down processing) than TD children provide no clear explanation for inter- or intra-individual heterogeneity in perceptual category learning by children with ASD.”