The multisensory nature of perceptual consciousness

In this chapter, David J. Bennett, Julia Trommershaeuser, and Loes C. J. van Dam present an overview of Bayesian principles used in modeling and analyzing perceptual capacities. On a Bayesian formulation ‘prior knowledge’ (e.g. about the distribution of slants), is combined with assumptions about how sensory information is generated (e.g., given viewing geometry in sensing a textured surface and given sensory system noise). When multiple sensory cues are available, cue combination is often well modeled as a weighted average, with the optimal weights inversely proportional to the variability of the sources of the estimates. With optimal weights, such ‘linear cue combination’ will maximize the precision of the upshot estimates. We show how this is a special case of a more general Bayesian cue combination formulation. The chapter furthermore discusses when to combine sensory information, the potential loss of knowledge of individual sensory estimates when such estimates are fused, and relevant links to the philosophy of perception (for example, concerning perceptual binding and concerning Molyneux’s question).