August 2023
Volume 23, Issue 9
Open Access
Vision Sciences Society Annual Meeting Abstract  |   August 2023
Active inference slows reversal learning in uncertain environments
Author Affiliations
  • Jet Lageman
    Vrije Universiteit Amsterdam
  • Johannes J. Fahrenfort
    Vrije Universiteit Amsterdam
  • Heleen A. Slagter
    Vrije Universiteit Amsterdam
Journal of Vision August 2023, Vol.23, 4897. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Jet Lageman, Johannes J. Fahrenfort, Heleen A. Slagter; Active inference slows reversal learning in uncertain environments. Journal of Vision 2023;23(9):4897.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

Perception is often framed as the inference of hidden states from noisy sensory input. Depending on whether sensory observations are passively sampled or actively generated, prior beliefs guiding the inference process can be informed by probabilistic sensory cues, or by knowledge about action-outcome relationships. Recent studies suggest that humans may hold stronger prior beliefs about the expected outcomes of actions than about sensorily cued observations. However, it remains unclear to what extent this effect interacts with perceptual features of the environment, such as sensory uncertainty. Here, we compared the outcomes of inference for perceptual judgments or goal-directed actions under different conditions of uncertainty. In a probabilistic reversal learning task, participants were either asked to infer a hidden state from computer-sampled observations, or to sample observations determined by a hidden state, while we manipulated the uncertainty of sensory observations. Both tasks required learning of the probabilistic relationship between hidden states and observations, and keeping track of sudden reversals in the hidden state. Critically, participants received identical sequences of evidence for the current hidden state under each instruction. Intermediate results indicate that reversal learning is slower when observations are actively sampled rather than passively monitored, especially when these observations are uncertain. In addition, Bayesian computational modelling demonstrates that participants perceived the environment as less volatile in the active task than in the passive task, regardless of sensory uncertainty. These findings replicate past work showing that humans are slower to update their beliefs on the basis of self-generated observations compared to passively perceived observations. The current study adds that this effect may increase under circumstances of high sensory uncertainty. This could signify a natural outcome of active inference: that by acting, we shape our perception of the inherently uncertain world.


This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.