Bayesian Account of Perceptual Decision-Making

By making predictions, learning from mistakes, and updating memories to include new information, the brain enables adaptive behaviour in daily activities. For instance, in perceptual decision-making tasks, it is critical to rapidly select the best behaviours based on current sensory inputs, that are frequently ambiguous or masked by noise. Using random dot motion (RDM) tasks, previous research on perceptual decision-making emphasised the role of sensory information in directing behaviour by varying simply the stimulus coherence and analysed the data using models that more or less explicitly presuppose bottom-up processing (e.g., drift-diffusion models). However, accumulating evidence (e.g., Bayesian models and the Free Energy Principle applications) suggests that the brain approximates optimal Bayesian inference rather than simply being a passive information filter. As a result, we need to shed light on the computations involved in goal-directed decision-making, with a focus on the predictive mechanisms at work in volatile experimental contexts. Here we used a probabilistic Random Dot Kinematogram (pRDK) in which the probability of witnessing a rightward/leftward motion changes throughout the task. Furthermore, to operationalise the predictions of the left and right dot motion in each trial based on previous information, an Ideal Bayesian Observer was used. This allowed us to study top-down predictions' impact on decision-making. The behavioural analyses revealed a substantial impact on behaviour from both coherence levels and probabilistic contexts. Specifically, a significant interaction between the probability of motion and direction was found, indicating faster responses when predictions matched what was presented.