Uninformative visual experience establishes long term perceptual bias.

Area of Science:

• Visual perception
• Computational neuroscience
• Bayesian inference

Background:

• Visual appearance requires resolving ambiguities between 2D retinal images and 3D scenes.
• This process can be modeled as Bayesian perceptual inference, combining sensory data with prior beliefs.
• Efficient visual systems should adapt priors based on environmental statistics.

Purpose of the Study:

• To investigate how perceptual priors are modified by exposure to visual stimuli.
• To test the conjecture that priors are updated only when sensory data disambiguate perception.
• To determine whether ambiguous or unambiguous stimuli are more effective in modifying priors.

Main Methods:

• Utilized a perceptually bistable stimulus (rotating wire-frame cube) to assess changes in the prior for 3D rotation direction.
• Compared the effects of ambiguous versus disambiguated stimuli on perceptual bias.
• Employed three groups of observers with matched perceptual experiences.

Main Results:

• Modification of the prior, indicated by a bias resistant to reversal learning, was significantly greater for ambiguous stimuli.
• Contrary to the initial conjecture, priors were modified preferentially when observers actively resolved ambiguity.
• Disambiguated stimuli had less impact on prior modification compared to ambiguous ones.

Conclusions:

• Active resolution of visual ambiguity, rather than passive exposure to environmental statistics, is a key mechanism for updating perceptual priors.
• Resolving difficult-to-interpret stimuli may be a primary method for establishing adaptive perceptual biases in natural environments.
• The findings challenge existing models of perceptual learning and Bayesian inference.