Test deferred causal attribution in human or biological inference

Determine whether a common initial response that defers causal attribution, followed by subsequent differentiation of learning rate, occurs in human or biological inference when changepoints and outliers are interleaved within the same sequence and their event-time observation distributions are controlled.

Background

The paper proposes Bayesian and inverse Bayesian (BIB) inference as a mechanism for handling ambiguous prediction errors. Rather than immediately attributing a large error to either a persistent latent-state change (a changepoint) or a transient observation outlier, BIB initially applies a common response and delays the application of its nullification candidate until a subsequent observation is available. Later observations then determine whether the candidate is maintained, producing differentiated learning-rate responses for changepoints and outliers.

The reported evidence for this principle comes entirely from simulations. The authors explicitly state that it remains unknown whether the same sequence—an initially common response that defers causal attribution followed by subsequent learning-rate differentiation—occurs in human or biological inference. They propose experimental testing with behavioral paradigms that interleave changepoints and outliers while controlling the observation distributions at event times.