Figure 2. Our two hypotheses of how the binding tendency may change: general predictions, and specific applications to Bayesian causal inference.

(A) Hypothesis #1, general predictions: priors are updated to reflect statistics of stimuli. Within our first hypothesis, all types of sensory experience are informative of relationships between auditory and visual signals in the environment, and therefore should cause a change in audiovisual binding. For instance, exposure to audiovisual stimuli that are congruent (either in space or in time) should lead to an increase in audiovisual binding, as the brain learns the association between the stimuli and updates its model of the world. Similarly, exposure to incongruent audiovisual stimuli should lead to a decrease in audiovisual binding. (B) Hypothesis #2, general predictions: mismatch between internal model and sensory experience requires modification. Within our second hypothesis, mismatches between the current model of the world and sensory experience are most informative regarding novel relationships between auditory and visual signals that the brain must take into account. Thus, whenever sensory experience indicates that congruent audiovisual stimuli actually originate from independent causes, this should change the brain’s internal model to decrease audiovisual binding. Similarly, if experience indicates that incongruent audiovisual stimuli actually originate from a common cause, the brain should increase binding of audiovisual information. (C) Hypothesis #1 and Bayesian causal inference. In the Bayesian causal inference framework, according to the first hypothesis, each audiovisual exposure serves to update the prior for binding (i.e., binding tendency). Evidence of a common cause by each exposure to congruent signals accumulates in this prior, increasing the tendency to bind, and evidence of independent causes by each exposure to incongruent signals decreases the tendency to bind. (D) Hypothesis #2 and Bayesian causal inference. According to our second hypothesis, the brain may only update its internal model of the world when there is a mismatch between sensory evidence and the current model. In our experiments, only conditions which present repeated evidence (i.e., strong temporal relationships between audiovisual stimuli) that signals thought to originally have independent causes actually originate from a common cause produce an error signal, and thus require an update of the internal model of the world. As evidence indicates that spatial relationships are not as strong of a cue as temporal relationships in updating integration tendencies (Wallace & Stein, 2007), it is unlikely that repeated presentation of temporally unrelated and spatially related stimuli would be potent enough to modify the brain’s tendency to bind.