Decision-making with hierarchical representations in humans

We can make good decisions by capturing and exploiting the structure of the natural world. It is thought that the formation of hierarchical representations allows humans to encode some of this structure. This thesis aims to investigate the behavioural and neural mechanisms underlying hierarchical representations. For this, I make use of diverse behavioural tasks, each with an intrinsic hierarchical structure, to ask if and how hierarchical representations are manifested in behaviour. I further use functional resonance magnetic imaging (fMRI) to pinpoint the brain regions associated with such hierarchical representations. In the first set of experiments, I studied hierarchical representations in the context of learning the composition of multi-feature visual stimuli that can be recursively grouped by mutual association. In a behavioural experiment, I found that participants learned a set of hierarchical relationships faster if they had previously learned stimuli with the same structure (compared to a baseline, non-hierarchical structure). A subsequent fMRI study revealed the neural concomitants of features associated with each hierarchical level. In a second imaging study, I studied the role of hierarchical representations during navigation in a virtual subway environment where stations are grouped into lines. I found behavioural costs (reaction time) that increased with the hierarchical description length of the plan to the goal. This was accompanied by neural costs (univariate activity) in a region of interest where I also could decode the current hierarchical context using a multivariate analysis. In a third imaging study, I modelled behaviour and imaging data in a rule discovery task, where object categorisation was determined by an unknown verbalisable rule. This allowed me to reveal the learning rates associated to different conditions, as well as neural signals related to the uncertainty of the rule. Finally, I discuss limitations of the studies, and the generality and specificity of the neural mechanisms underlying hierarchical representations.