If the behaviours described by Jaswal & Akhtar (J&A) are “adaptive responses to a particular situation” (sect. 2.5, para. 2), then the crucial question is this: Why are the adaptive responses to the environment different in autism than in a neurotypical population? Or, if many of these behaviours are used by the neurotypical population, then why is the frequency of their use different in autism? Given the same environment, what is different about autistic individuals that makes their behaviours distinct, yet still adaptive?

In evolutionary ecology, adaptive behaviour consists of responses to the demands of the environment that promote survival and reproductive success. While originally related to phenotypic strategies of whole populations, it has been extended to individual differences (Buss & Greiling Reference Buss and Greiling1999; Wilson Reference Wilson1998) and co-opted by clinical psychology to refer to abilities that conform to social expectations for age-appropriate independent living (Coulter & Morrow Reference Coulter and Morrow1978; cf. Sohn Reference Sohn1976). J&A repeatedly state that characteristic autistic behaviours are adaptive (10 occurrences). This should be taken to mean that the behaviours have cognitive utility (or constitute a cognitive phenotype with evolutionary success; Montague et al. Reference Montague, Dolan, Friston and Dayan2012). We should agree with J&A that many distinctively autistic behaviours are adaptive in this way. This observation is, however, best framed in terms of predictive processing theories of autism.

Predictive processing accounts of autism are promising in that they explicitly account for differences in adaptive strategy and thereby are able to address the question I posed for J&A at the outset (Brock Reference Brock2012; Lawson et al. Reference Lawson, Rees and Friston2014; Reference Lawson, Mathys and Rees2017; Palmer et al. Reference Palmer, Lawson and Hohwy2017; Pellicano & Burr Reference Pellicano and Burr2012; Van de Cruys et al. Reference Van de Cruys, Evers, Van der Hallen, Van Eylen, Boets, de-Wit and Wagemans2014). Predictive processing is a general and unifying explanation of brain function with growing application to psychiatry (Friston et al. Reference Friston, Stephan, Montague and Dolan2014; Reference Friston2017). These accounts argue that, as the brain seeks to model current and future states of the world, incoming sensory information is weighted differently in autism than in the neurotypical case. Action and perception become tools for inference about the causal origins of sensory inputs, and these theories can thereby explain differences in both domains in autism. The purported difference in general processing in autism generates different responses from neurotypicals because superficially identical environments are mentally represented differently. For example, an adaptive response as an autistic person may be to exploit highly predictable affordances (Constant et al. Reference Constant, Bervoets, Hens and Cruys2018), whereas for neurotypical individuals, it may be to engage in more exploration. Note that our actions shape our environment, and so this challenges the purported equality of the environments experienced by individuals in these two groups, further giving reason for why the adaptive response to it might differ.

J&A are correct to say that insofar as the social motivation theory is meant to be a unified explanation of autistic cognition and behaviour, it fails to explain all the available evidence (sect. 3 introduction). This includes not just the (very important) firsthand testimony, but also other findings not discussed by J&A. Predictive processing theories account for the tendency for autistic individuals to perceive small elements of the sensed world particularly precisely, therefore accounting for differences in sensitivity to sensory information (Ben-Sasson et al. Reference Ben-Sasson, Hen, Fluss, Cermak, Engel-Yeger and Gal2009), as evidenced by superior performance in visual search. Weaker prior expectations for stimulus qualities (Pellicano & Burr Reference Pellicano and Burr2012), higher sensory precision (Brock Reference Brock2012; Lawson et al. Reference Lawson, Rees and Friston2014), or inflexibly high sensitivity to the differences between expectations and outcomes (prediction error; Van de Cruys et al. Reference Van de Cruys, Evers, Van der Hallen, Van Eylen, Boets, de-Wit and Wagemans2014) are potential specifications of this learning rate difference in autism (Palmer et al. Reference Palmer, Lawson and Hohwy2017). Increased interest in highly regular domains due to the tendency to construct a prediction-satisfying environment (Constant et al. Reference Constant, Bervoets, Hens and Cruys2018) may also account for autistic savant skills (Meilleur et al. Reference Meilleur, Jelenic and Mottron2015).

Furthermore, predictive processing accounts of autism offer plausible explanations of the four key pieces of behavioural evidence discussed by J&A.

Predictive processing explains why it may be necessary for autistic people to engage in calming, self-regulatory behaviour in social situations, such as avoiding eye contact. Social situations involve some of the most complicated interacting causes in our environment, and so learning from social stimuli (and thereby participating in successful interaction) requires integrating information over many instances to learn what actions and stimuli might yield the clearest social signal. It is hard to predict another person's behaviour, partly because each social interaction is, in many ways, completely novel, and partly because social interactions are interpreted against a rich tapestry of background information. Reduced eye contact during highly demanding social contexts may be related to decreased precision of social cues (from failing to learn these over many instances), which thereby decreases the ability to reduce uncertainty overall (Palmer et al. Reference Palmer, Lawson and Hohwy2017). Predictive processing accounts of autism also explain repetitive motor stereotypies as active ways of making incoming sensory information more precise (Palmer et al. Reference Palmer, Lawson and Hohwy2017).

A similarly complex social action is pointing. One must learn to use actions like pointing to reduce uncertainty by controlling and predicting the flow of an interaction based on one's social history. Reduction in pointing may be explained by a weaker understanding of what states in the interlocutor are influenced by the autistic person’s actions and how to achieve desired states.

Echolalia too can be understood as an adaptive behaviour in that it reduces prediction error. Oral participation in conversation is made more predictable by reusing heard utterances to communicate similar meanings. This plausibly makes the interlocutor's response more predictable, as the same situation is repeated over multiple events. Predictive processing theories are also compatible with firsthand accounts that social situations are not less appealing, but potentially less accessible to autistic individuals due to the many inferred interacting causes which must be modelled.

Predictive processing accounts of autism suggest that differences in updating mental representations of the self and the environment lead to differences in strategies of inference. This includes perception and action selection which may account for differences in adaptive behaviours between neurotypical individuals and autistic individuals.