Perceptual decision-making under uncertainty in humans

On a daily basis, humans need to make decisions in a complex uncertain world that requires them to adapt their decision strategies to the environmental volatility. In order to make a decision, humans often need to integrate noisy information and combine it with prior experience they learned through previous similar decision situations. This thesis examines how humans form decisions based on noisy information using electroencephalography (EEG), and how they make use of prior expectation in order to guide their decisions using magnetoencephalography (MEG). For both parts, models that capture the different decision-making processes were used to make predictions of human behaviour and underlying neural activity. To study how humans integrate noisy evidence a novel continuous random dot motion task was developed in which signal periods had to be detected in a stream of noisy evidence which, in contrast to previous studies, is controlled by the experimenter. Simulations with a leaky accumulator model predicted that participants should discount past evidence to solve the task. Participant behaviour suggests that they adapted their evidence strategy to the frequency and length of signal periods in the noise in a manner that is consistent with leaky accumulation. A contingent negative variation (CNV) signal measured with EEG suggests that the brain adapted its evidence accumulation process in line with the behavioural results. In addition, a P300-like signal encoded the difference between previous and current evidence sample, scaling with the size of the difference and trial frequency. To study how humans update their prior belief about decision outcomes a different random dot motion task was used in which participants had to integrate evidence within trials but also had to integrate prior expectation over past trial outcomes to inform choice. A Bayesian model was fit to subjects’ behavioural data to estimate the prior expectation on each trial that was subsequently used to study human choice behaviour and neural activity in the task. Human choice behaviour depended on both, the within-trial evidence, as well as prior belief. Activity in parietal cortex, measured with MEG, at the time of feedback suggested that the feedback type, i.e., whether a choice was correct or incorrect, is encoded in parietal cortex. There was only a very weak influence of prior belief on the neural signal.