Interpersonal Predictive Coding Across the Autistic Spectrum

Background:

Action perception is not simply a reflection of what happens, but a projection of what will happen next. In this talk, the notion of interpersonal predictive coding will be introduced and reviewed in light of evidence from behavioral studies, which investigate the ability to use social information to learn from and predict others' actions by using point-light displays that depict communicative as compared to non-communicative actions between two agents. Furthermore, computational modeling was used in a second paradigm, which required individuals to learn about the probabilities of social and non-social cues and to integrate them in order to gain points in a card selection task. Objectives:

These studies aimed at investigating whether - in spite of intact performance on action recognition tasks - individuals with autism might show impairments of the ability to predict subsequent actions during an observed dyadic encounter. An additional aim was to characterize and quantify the processes that allow individuals to integrate social and non-social cues during decision-making.

Methods:

The first study used point-light displays that depict communicative as compared to individual actions between two agents and were taken from an established database. Agent A demonstrated either a communicative or an individual action while agent B either showed a communicative action or was not present. As part of this two-alternative forced choice paradigm, participants had to indicate whether they thought agent B had been present in one of two trials. Eyetracking measurements were also obtained.

In a second study the ability to integrate social and non-social information during decision-making was studied by making use of a probabilistic learning task in conjunction with Bayesian modeling to investigate autistic trait-related differences in basing one's decisions on social rather than non-social information. Using computational modeling a weighting parameter was calculated that indicated the extent to which participants were influenced by the social cues about which they had received no information during the instructions.

Results:

Using a two-alternative forced choice paradigm, it was shown that participants' sensitivity for detecting a second agent masked by noise signals was significantly higher when a first agent showed a communicative action. This measure of sensitivity also showed a negative and significant correlation with increasing autistic traits across the entire spectrum. Results from the second study show that autistic traits are negatively correlated with the extent to which decisions are based on social rather than non-social information. These autistic trait-related differences in cue integration also explained performance differences across different groups of participants.

Conclusions: Taken together, these findings highlight that autistic traits are related to differences in the integration, anticipation and automatic responding to social cues, rather than a general inability to register and learn from social cues. Importantly, such differences may only manifest themselves in sufficiently complex situations.