Abstract
Recent research has demonstrated that the prefrontal cortex (PFC) has a critical role in metacognitive judgments of confidence. Given that PFC receives inputs from across the visual cortex, one counter-intuitive prediction is that signals that are represented in different visual areas may interact with each other at the metacognitive level. We tested this prediction by asking subjects to judge the dominant color and shape in a set of 40 letters. In Experiment 1 we found that confidence “leaked” between the two tasks (i.e., high confidence on the color task was associated with higher confidence on the shape task on that same trial) even though no such effect was found for perceptual accuracy. To ensure that the results were not due to response priming, in Experiment 2 confidence was given on a continuous scale for one task, and through the “opt-out” paradigm (in which subjects can choose to withhold their response and earn a certain small reward) on the other. Despite that difference in responding procedures, confidence leak between the two tasks was just as strong manifesting in 42 of the 44 subjects across the two experiments. Experiments 3 and 4 showed that confidence leak was also present for simple Gabor stimuli and that manipulations of signal quality through contrast and attention did not change the strength of confidence interactions. Finally, in Experiment 5 we disrupted PFC function with theta-burst TMS and observed a decrease in confidence leak compared to TMS to control regions. The results across all experiments are well described by a single-parameter model based on Bayesian inference, in which observers are constantly predicting the overall strength of signals in the environment in order to respond to them optimality. The model thus provides a normative explanation for the ubiquitous and novel phenomenon of confidence leak between different tasks.
Meeting abstract presented at VSS 2015