What is probabilistic learning? How might it be related to emotion recognition?
Probabilistic learning is a term we use to describe the learning of relationships between cues and outcomes that are not reliable. That is, sometimes the cue may signal an outcome and sometimes it may not.
We assume that to recognise an emotion correctly, one needs to make correct inferences based on certain cues, some of which are more reliable than others. For example, if we see someone with tears running down their face, we may infer that they are sad since our past experiences have taught us that people who are sad may cry. However, it is also possible that those are tears of joy and that in fact, that person is actually happy, though our experiences tell us that tears are a less reliable cue for happy than for sad.
Is probabilistic learning the underlying difficulty autistic individuals have with emotion recognition?
We think that autistic individuals may have more difficulty recognising emotions than neurotypicals partly because autistic individuals have atypical integration of probabilistic cues. That is, autistic individuals may find it more challenging to learn when the unreliable cue does vs. does not signal an outcome compared to neurotypicals. If this is indeed true, it is possible that probabilistic learning may explain other autistic experiences such as insistence of sameness and difficulties coping with uncertainty.
What did we find so far?
Across two studies and various tasks, we examined whether there are any group differences among autistic vs. non-autistic individuals in probabilistic learning. We found that group difference was only evident when the association between the cue and outcome is relatively weak (approximately 60% predictive). Moreover, computational modelling suggests that, compared to non-autistic individuals, autistic individuals were more likely to incorporate feedback — even those that should be ignored because it is ‘noise’– to their subsequent decision, which may lead to differences in their performance on the weak cue-outcome associations. Though we didn’t examine this relationship directly, this may also explain the challenges autistic individuals face in social interactions, given that such interactions are typically governed by complex and very weak cue-outcome associations.
Publications
Ong, J. H. and Liu, F. (2022) Probabilistic learning of cue-outcome associations is not influenced by autistic traits. Journal of Autism and Developmental Disorders. ISSN 0162-3257 doi: https://doi.org/10.1007/s10803-022-05690-0
Paper in preparation.