Perceptual development as optimisation of inference
Marko Nardini, Durham University
To survive, organisms must deal with many kinds of uncertainty, such as recognising objects given partial or uncertain information, or planning an action (e.g. reaching for a cup) with an uncertain outcome. Recent evidence suggests that the adult nervous system meets these challenges by implementing or approximating principles of Bayesian Decision Theory (BDT), which provides optimal solutions to problems of perception and action under uncertainty. This raises the interesting possibility that the long developmental trajectory for some perceptual skills in childhood can be understood as a process of optimisation of inference. I will present results from recent studies in support of this idea, showing that key elements of BDT are not in place until remarkably late in childhood – in perceptual tasks, motor tasks, and brain circuits. Even in quite simple tasks such as judging the layout of 3D surfaces, we see that sub-optimal computation (as distinct from noise) makes a major contribution to children’s relatively low performance. These results provide a starting point for investigating the processes of development and learning by which the nervous system optimises its perception and action abilities. Progress on this problem has important future applications to atypical development, sensory / motor rehabilitation, and the design of intelligent agents who can learn from their environment.