Abstract
Introspective agents can recognize the extent to which their internal perceptual experiences deviate from the actual states of the external world. This ability, also known as insight, is critically required for reality testing and is impaired in psychosis, yet very little is known about its cognitive underpinnings. We developed a Bayesian modeling framework and a novel psychophysics paradigm to quantitatively characterize this type of insight. To induce strong perceptual distortions, we created a task based on a variant of the motion after-effect (MAE) illusion. To measure insight in a quantitative manner, we quantified participants’ ability to report beliefs in which they demonstrate compensation for their perceptual distortions. Healthy participants compensated for the MAE illusion in a condition in which they were asked to infer the actual direction of motion (‘Believe’ condition), in addition to a condition when they reported perceived motion (‘See ’ condition). Across two experiments with 44 participants total, we found that participants could compensate for the illusion when judging the actual direction of a motion stimulus in the ‘Believe’ condition. In a second experiment (N = 22 observers), a parametric choice of the test stimuli allowed us to fit Bayesian model variants jointly to the participants’ category responses and confidence reports. When compared with a response-bias model variant and a category prior variant, the insight model provided the best fit of the data. Furthermore, reaction-time and pupil-dilation data showed signatures consistent with inferential adjustments in the Bayesian insight model. Our results suggest that people can question the veracity of what they see and make insightful inferences that incorporate introspective knowledge about internal distortions.