Perceptual learning without feedback is accompanied with systematic changes in confidence processing.

Perceptual learning can occur without external feedback, raising questions about what internal signals might guide this learning. Here, we test the hypothesis that confidence, namely the observer's subjective sense of decision accuracy, is this internal teaching signal. In the present study, participants performed a visual motion discrimination task across 5 days without feedback. After every two trials, they indicated which decision they were more confident in, using a confidence forced-choice paradigm. In addition, following previous literature, we attempted to manipulate confidence independently of accuracy by varying dot density while holding signal-to-noise ratio constant (overall-strength manipulation), however, this manipulation did not consistently separate confidence from performance. In our study without feedback, we found that some participants improved their performance in the visual task, but other participants failed to become better. Those participants who improved their perceptual thresholds over time ("learners") also exhibited systematic changes in their confidence computations: not only sensory noise decreased, but confidence efficiency declined due to increases in confidence-specific noise. Learners also showed an increased reliance on additional evidence not used in the perceptual decision itself (confidence boost). Non-learners showed no comparable changes. These results suggest that confidence judgments evolve during perceptual learning and may serve as an index of how internal monitoring adapts during feedback-free training.