Abstract
Abstract
In this viewpoint, Christoph Kayser, Uta Noppeney, and Tim Rohe highlight the central role of normative Bayesian frameworks, in particular Bayesian causal inference, for explaining how the brain decides whether sensory signals should be integrated or segregated and how they are weighted by their reliabilities. In parallel, the Multisensory Correlation Detector model is emphasized as a mechanistic proposal for how the brain computes spatial and temporal relations across sensory channels. Together, these developments have shifted the field from merely locating multisensory processing in the brain to characterizing the specific computations implemented across cortical hierarchies. The authors outline current trends, such as model-based analyses of neuroimaging and neurophysiological data, increasing use of artificial neural networks to study multisensory processing at larger scales, and emerging work in naturalistic, real-world settings. They discuss ongoing controversies surrounding the influence of attention and tasks on multisensory integration, the limits of Bayes optimality, and the role of neural oscillations, especially alpha rhythms, in shaping multisensory binding. They further advocate for stronger open-science practices and systematic replication of influential findings to consolidate the empirical and theoretical foundations of the field. Finally, they encourage early-career scientists to develop strong methodological skills, cultivate their own conceptual ‘angle’, and actively engage with the multisensory community.