Action predictability in autistic and non-autistic adolescents

I want to share my second first-author publication:

Action predictability is reflected in beta power attenuation and predictive eye movements in adolescents with and without autism

DOI of the official version, which is open access: 10.1016/j.neuropsychologia.2021.107859

This paper appears in the journal Neuropsychologia, which specialises in cognitive neuroscience and has a history of publishing clinical cognitive neuroscience.

The summary of this paper is this:

During observation of increasingly predictable actions, autistic and non-autistic teenagers spontaneously predicted others' actions both behaviourally and neurally, at odds with previous findings and various theories of autism.

Action observation in autism gets a lot of attention, especially the idea that autistic people have a harder time predicting others, and this prediction difficulty leads to social communication differences. Empirical results have been very mixed, however.

Most studies use mu power over sensorimotor cortex to measure motor system activation, and hypothesise this activation during action execution and observation in non-autistic participants, but only (or mostly) during execution in autistic participants.

Because results are so mixed, some authors have suggested the specific type of action matters. Perhaps autistic participants do make predictions for simple actions, but commit less to predicting a specific outcome when there's more uncertainty, e.g. with more complicated actions.

To test this idea, we used the task from Braukmann et al, 2017 (https://doi.org/10.1016/j.neuropsychologia.2017.07.009…) which consists of multi-step actions, where the gist is clear from the start, but as the action unfolds, viewers can build gradually more specific predictions about each step.

In one example, we see a kettle, a cup and a teabag. We can guess tea will be made, but not predict any specific movements yet. Once the actor picks up the teabag, we can predict it'll go in the cup, and may activate our own motor system and look at the cup in anticipation.

The viewer can then build a better prediction about what will happen next, since they know something about how to make tea, and one item has been used and two have not yet been used. In this way, the viewer can make more and more specific predictions as the action steps unfold.

From Braukmann's study, we know that mu power, the usual measure of motor activation, stays constant for all action steps, but beta power and eye movements are modulated by the predictability. So we focussed on these to see how autistic and non-autistic teens build predictions.

Note that the participants were not aware that we were studying predictions or how we might operationalise them until after the study. We told them we wanted to know how they processed daily actions. They were just asked to calmly watch the videos, and didn't have to respond.

Freq and Bayesian ANOVAs show less mu power during action observation than during a baseline period, so participants were recruiting their motor systems as would be expected. The Bayesian test shows not enough evidence to say whether the groups differ but this isn't our qu today.

Freq and Bayesian ANOVAs show that beta power was parametrically modulated by action step, with evidence that the groups do not differ. So autistic and non-autistic teenagers build neural predictions spontaneously and make them more specific as they receive more information.

Multilevel modelling shows modulation of anticipatory eye movements by action step, so viewers don't only use predictability to build models, they use it make behavioural predictions too. Again, the two groups responded in the same manner to the increasing predictability.

We see very little difference between our two groups on any of the measures, and there is evidence from Bayesian stats that the groups truly don't differ. We therefore conclude that autistic teens can and do spontaneously predict others' actions, similar to non-autistic teens.

A few limitations: our participants have quite high IQs (matched between the groups) and this may not reflect the participants from previous studies. We tried to control for decreasing beta power over time regardless of predictability but this didn't go as expected.

So overall, I'm convinced that these autistic teens can and do spontaneously predict. This may not apply to all autistic individuals, but is a problem for any theory that claims they can't, or don't spontaneously.