Implicit causal inference in audiovisual spatial representations

The causal inference problem in multisensory perception poses a fundamental challenge to our brains in a multisensory environment: how to decide whether sensory stimuli originate from a common source and should be integrated, or from distinct sources and should be segregated. The brain addresses this problem by inferring causal structure from the spatiotemporal disparity of multisensory stimuli. However, it remains unclear whether the brain handles causal inference implicitly, or whether it requires effortful and explicit cognitive processing. This study investigated how human observers (N = 47) implicitly infer causal structure when judging the auditory distance of two sequential audiovisual stimuli. In this distance task, we combined representational similarity analysis and multidimensional scaling to retrieve participants' auditory spatial representations. We then compared visual biases on auditory representations (i.e., the ventriloquist effect) to visual biases in three classical auditory localisation and causal judgment tasks. We found that visual biases in the distance task were less influenced by the spatial disparity of the audiovisual stimuli compared to the classical tasks. This pattern was best fitted by a computational stochastic-fusion model. Only in the joint localisation and causal task, small spatial disparity increased the visual bias as predicted by a computational Bayesian causal inference model. Our results suggest that causal inference requires explicit cognitive processing that observers only apply if the causal structure of stimuli is directly relevant to the task. Otherwise, the brain relies on simpler automatic decision strategies such as stochastic fusion.