Explicit knowledge of task structure is a primary determinant of human model-based action


If you would like to chat with the speaker on the day, please do get in touch with Lauren Burgeno at Lauren.burgeno@dpag.ox.ac.uk.

Human participants in behavioural neuroscience experiments are typically given information about the structure of the task they will perform before they encounter it. The effect of these instructions on behaviour and brain activity are rarely explicitly considered, though where studies have looked for them, instruction effects on behaviour are often profound. Prior work on instruction effects has used simple conditioning or reward guided decision tasks, which do not differentiate between model-based and model-free learning systems, though respectively to generate goal-directed and habitual actions. It is therefore unknown whether or how instructions differentially affect these systems.

To address this, we developed a simplified version of the two-step decision task of Daw et al. (2011), that could be presented to subjects with minimal prior information. Subjects performed 900 trials without instructions, were then told the structure of the task, then performed a further 300 trials to assess how this information changed their behaviour. Initial uninstructed behaviour was model-free, with rewards directly reinforcing preceding actions. Model-based control, which uses predictions of the states resulting from each action to guide choices, emerged with experience in a minority of participants. Providing explicit information about task structure strongly increased model-based control, but also had effects on model-free learning, which we think reflect changes in how the subjects represent the task’s state space.

We also ran the simplified two-step task in individuals with obsessive compulsive disorder (OCD), finding that compared to healthy volunteers they were less able to learn to use model-based control from experience alone, but performed similarly after receiving explicit information about the task structure.