Conference on Cognitive Computational Neuroscience

When we make perceptual decisions, confidence usually co-varies with decisional accuracy. However, sometimes this correspondence breaks down, e.g. in atypical environments or clinical populations. This raises an important question: what are the neural computations of perceptual metacognition if their output can diverge from perceptual decisions themselves? In a recent paper, we argued that tuned inhibition (TI)—i.e., the degree to which a neuron is inhibited by neighboring neurons with opposing tuning preferences, which varies from neuron to neuron—is a crucial part of the underlying mechanism. Here we explore how we might validate the TI model using fMRI data, by simulating the activity of 'voxels' of different compositions in the presence of evidence for and against a perceptual decision in a decision+confidence task. We show that we can quantify how a voxel's TI level dictates its predictive power for confidence judgments, providing support for use of these stimuli and analyses in fMRI data to validate the TI model of perceptual metacognition.