Bayesian Causal Inference Accounts for Multisensory Filling-In at the Blind Spot

Area of Science:

• Neuroscience
• Cognitive Science
• Psychophysics

Background:

• Multisensory perception relies on integrating information from different senses.
• The physiological blind spot lacks direct retinal input, posing a challenge for visual processing.
• Understanding how the brain processes information in the absence of direct sensory input is crucial.

Purpose of the Study:

• To investigate if audiovisual integration occurs without bottom-up visual input across the blind spot.
• To determine if the brain adjusts sensory uncertainties and priors in the blind spot.
• To assess if causal-inference computations are preserved in the blind spot.

Main Methods:

• Participants judged audiovisual illusions presented across the blind spot and a control location.
• Bayesian Causal Inference (BCI) models were used to analyze responses.
• Sensory noise, priors, and causal-inference strategies were estimated and compared.

Main Results:

• Audiovisual illusions were robust at both the blind spot and control locations, indicating preserved integration.
• The Bayesian Causal Inference model revealed higher visual uncertainty and broader prior expectations at the blind spot.
• Auditory precision and the causal prior remained stable, suggesting flexible adaptation.

Conclusions:

• Perceptual inference remains intact even in the absence of retinal input.
• The brain flexibly adapts internal sensory uncertainty rather than altering core multisensory computations.
• The computational architecture for causal inference is preserved across sensory-deprived regions.