Autistic Individuals Are Slow in Updating Their Internal Representations and Predictions

Background:

In the last 15 years our lab studied dyslexia, asking what neural and cognitive mechanisms underlie their poorer performance in a broad range of simple serial discrimination tasks. We found that dyslexics' performance reflects impaired inference of stimuli statistics rather than a noisier sensory system. To assess dyslexics' sensitivity to context we studied a phenomenon termed "contraction bias" – the pulling of perception towards the "prior" – the mean of similar, previously encountered, stimuli. This pulling occurs shortly after stimulus presentation, and can be understood within a Bayesian framework of optimizing perception based on previous knowledge, given noisy sampling or memory. Dyslexics' contraction bias is smaller than controls', reflecting inefficient use of environmental statistics. Using behavioral, ERP and computational methods developed in our lab, we found that this inefficiency stems from faster decay of their implicit memory of previous stimuli.

Objectives:

We now asked whether the pattern of inference of high functioning autistic individuals, with no language difficulties, is similar to dyslexics'. Recent studies found reduced efficiency of statistical inference in Autism. However, Autistics' difficulties are quite different than those of dyslexics.

Methods:

We recruited a population of 58 adults, 28 autistic and 30 typically developing (TD) individuals, matched for age and reasoning skills, and replicated experiments of simple serial frequency discrimination, and paced finger tapping, which we previously administered to adult dyslexics.

Results:

Classical thresholds did not differ between the groups. However autistics' contraction bias was substantially smaller than that of TD participants. Analyzing the impact of previous trials on autistics' performance we found that they tend to under weigh recent trials. While both dyslexics and TD participants weigh recent trials more than earlier ones when implicitly calculating the mean of previous trials, autistic participants do not. However, their accumulation of stimuli statistics based on earlier trials is similar to that of TDs. This observations suggests that while they may accumulate adequate statistics in stable environments, they are less flexible in adjusting their representations, and hence predictions, to novel ones where statistics changes. To test this interpretation we administered a simple synchronize to metronome tapping task, where performance is determined only by very recent intervals. While dyslexics' performance was similar to TDs', autistics could not reliably synchronize to the external beats. Particularly challenging situations were those when metronome' beat changed and autistic participants were slower in adapting to new intervals, further supporting the slow-update interpretation.

Conclusions:

Autistics and dyslexics show inefficient statistical learning, but due to different underlying mechanisms. While dyslexics are fast adapters but show fast decay of their memory traces, autistics adapt slowly but show adequate retention. This difference is consistent with the broader behavioral profiles - dyslexics' fast adaptation but impoverished categorical representations, and autistics' slow adaptation but adequate categorical representations. A broad range of daily behavioral characteristics can also be explained as a consequence of these implicit learning patterns.