Tuned inhibition in perceptual decision-making circuits can explain seemingly suboptimal confidence behavior

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Modeling

Background:

• Dominant views link perceptual confidence to decision accuracy probability.
• Recent studies show confidence and accuracy can be dissociated.
• An alternative hypothesis proposes confidence reflects evidence magnitude, less sensitive to opposing evidence.

Purpose of the Study:

• To biologically instantiate an alternative hypothesis of perceptual confidence.
• To develop a neural network model explaining confidence-accuracy dissociation.
• To investigate the role of tuned inhibition in confidence computation.

Main Methods:

• Developed a neural network model with tuned inhibition, a property of sensory neurons.
• Modeled how different levels of tuned inhibition influence decision and confidence computations.
• Compared model fits to behavioral data including accuracy, reaction time, and metacognitive sensitivity.

Main Results:

• The model successfully accounts for findings where confidence and accuracy dissociate.
• Confidence computed from units without tuned inhibition best explains behavioral data.
• Tuned inhibition levels differentially contribute to perceptual decisions and confidence judgments.

Conclusions:

• The brain may use a heuristic for perceptual confidence, relying on diverse neural resources.
• Neural network models with tuned inhibition offer a biologically plausible explanation for confidence computation.
• Findings suggest distinct neural mechanisms underlie perceptual decisions and confidence reporting.