Model-Based Learning and Enhanced Systemising Ability in High Functioning Autism and Asperger's Syndrome

Background:

Restricted and repetitive interests/behaviours and need for sameness represent an important source of impairment for individuals with autism-spectrum-conditions (ASC) yet their relationship to altered decision-making processes have rarely been investigated. This is surprising given that impaired decision-making processes are central to another neurodevelopmental disorder e.g. impaired delayed discounting in ADHD(Barkley et al., 2001; Ellen Demurie, 2010) .

Here we used a hierarchical decision making task(Daw et al., 2011) to investigate whether individuals with ASC tend to adopt simple model free or alternate more sophisticated model-based strategy that require generation of an internal model of the task structure. We predicted that unlike individuals with other forms of compulsive behaviour e.g. OCD who are predicted to use simple model-free approaches, individuals with ASC who show greater systematising are more likely to use model-based strategies.

Objectives:  

Determine whether the use of sophisticated model-based strategies is affected by ASC.

Methods:  

11 ASD (7 male, mean age 42 ± 14) and 11 control (3 male, mean age 24 ± 4) participants were recruited. Each performed the task 2 times, each consisting of 67 iterations, following a comprehensive tutorial. AQ, EQ, NART, BDI and STAI questionnaires were also administer to measure mood and IQ.

In the task there were two stages: a transition stage where choices govern the probability of being presented with different pairs of stimuli in the next stage, namely the reward stage which determines the probability of being rewarded. The task was computerised, and controlled using the keyboard. I used standard RL equations to model participants’ estimates of reward and transition probabilities. The model was a hybrid of model-based and model-free RL, the ratio of use between these two sub-models being determined by a parameter.

Results:  

Participants with autism showed greater use of model based RL (T₍₁₁₎=2.4,p<0.05) . AQ correlated with use of model based RL (R²=0.30, p<0.05), while IQ did not (R²=0.035, p>0.05). Recruitment is ongoing and full results will be presented at the conference. Effects remained significant after controlling for age.

Conclusions:  

Our provisional results suggest an increased tendency to model-based decision-making processes in ASC. This is interesting as it is in contrast to predictions that individuals with other types of repetitive behaviour e.g. OCD or compulsive drug are more likely to use model-free strategies. This insight may help explain why restrictive/repetitive behaviours are resistant to conventional CBT approaches developed for OCD and may help inform refinements of these. Performance on this task may also suggest an alternate way of indexing tendency to systematise in ASC.