Intrinsically motivated curiosity, learning and attention: new biological and computational approach

Abstract

We propose an interdisciplinary project in the areas of neuroscience and developmental robotics. The objective is to devise a novel computational and biological framework for understanding intrinsically motivated curiosity, active learning and attention. This approach will fill large gaps that currently exist in the study of these topics in these different domains. In neuroscience, countless studies have examined learning and attention in highly simplified experimental contexts but few have examined the active control of these processes, i.e. the ways in which agents independently select what they are interested in and what they wish to learn. In developmental robotics, new algorithms have been developed for controlling active exploration in complex environments, but have not yet been validated against biological or psychological observations. We will fill these gaps by testing and extending the computational models that I developed in my doctoral work, based on behavioral and neural observations from rhesus monkeys. Specifically, I will record activity in parietal and frontal attention-related areas and in midbrain dopaminergic neurons, to understand their role in the active control of attention and interest. The project will greatly complement my training in robotics by allowing me to learn an entirely new set of techniques for behavioral and neural recordings in rhesus monkeys. In addition, the results are likely to have a transformative impact on neuroscience and developmental robotics by establishing a first computationally tractable framework for understanding intrinsically motivated curiosity and linking it with systems of attention and intrinsic motivation.