Abstract:Our brains predict future sensory input based on their current beliefs about the world around us, but interpreting prediction errors can be challenging in a volatile environment because they can be caused by stochastic noise or by outdated predictions. Noisy signals should be integrated with prior beliefs to improve precision, but the two should be segregated when environmental changes render prior beliefs irrelevant. Bayesian inference provides a statistically optimal solution to deal with situations in which there is uncertainty about the cause of the prediction errors. However, the method quickly becomes memory intensive and computationally intractable when applied consecutively. Here, we systematically evaluate the predictive performance of Bayesian causal inference for human perceptual decisions in a spatial prediction task based on noisy audiovisual sequences with occasional changepoints. We elucidate the simplifying assumptions of a previously proposed reduced Bayesian observer model, and we compare it to an extensive set of models based on alternative simplification strategies. Model-free analyses revealed the hallmarks of Bayesian causal inference: participants seem to have integrated sensory evidence with prior beliefs to improve accuracy when prediction errors were small, and prior influence gradually decreased as prediction errors grew larger, signalling probable irrelevance of the prior due to a changepoint. Model comparison results further supported the hypotheses that participants computed probability-weighted averages over the causal options (noise or changepoint) and that they iteratively summarized their beliefs while accounting for causal uncertainty akin to the reduced Bayesian observer model. However, we also found that participants' reliance on prior beliefs was systematically smaller than predicted by the model, and this was best explained by individually fitting lower-than-optimal parameters for the a-priori probability of prior relevance. We conclude that perceptual belief updating in volatile environments with stochastic noise is well described by a simplified model of consecutive Bayesian causal inference. Observers utilize priors flexibly to the extent that they are deemed relevant, though also conservatively with a lower tendency to bind than an ideal Bayesian observer.

Causal inference predicts the transition from integration to segmentation in motion perception

Hierarchical motion perception as causal inference

To integrate or not to integrate: Temporal dynamics of hierarchical Bayesian causal inference

Seeing "what" Through "why": Evidence from Probing the Causal Structure of Hierarchical Motion.

How relevant is the prior? Bayesian causal inference for dynamic perception in volatile environments

Causal inference in perception

Multisensory Causal Inference in the Brain

Integrated Perceptual Decisions Rely on Parallel Evidence Accumulation

Causal inference during closed-loop navigation: parsing of self- and object-motion

Learning to segment self-generated from externally caused optic flow through sensorimotor mismatch circuits

The use of chromatic information for motion segmentation: differences between psychophysical and eye-movement measures

Visual-Based Spatial Coordinate Dominates Probabilistic Multisensory Inference in Macaque MST-d Disparity Encoding

Post-decision biases reveal a self-consistency principle in perceptual inference

Transsaccadic integration of visual information is predictive, attention-based, and spatially precise

Neural dynamics of causal inference in the macaque frontoparietal circuit

Visual Motion Integration for Perception and Pursuit

Solving the causal inference problem

The interpretation of visual motion: evidence for surface segmentation mechanisms

Probabilistic Model of Visual Segmentation

A Normative Theory for Causal Inference and Bayes Factor Computation in Neural Circuits

Illusory Movement of Stationary Stimuli in the Visual Periphery: Evidence for a Strong Centrifugal Prior in Motion Processing.