Avinash Ranjan¹ (), Jongmin Moon¹, Robbe L. T. Goris¹; ¹The University of Texas at Austin

Perceptual interpretations of the state of the environment are subject to noise and therefore inherently uncertain. Humans know this and typically report more confidence in certain than in uncertain interpretations. The computational processes that underlie this form of metacognition have mostly been studied using binary decision-making tasks in which subjects assign sensory stimuli to one of two categories. Yet many real-world tasks involve continuous estimation rather than binary decision-making. We wondered whether observers possess similar introspective insight in the reliability of continuous perceptual estimates. To investigate this, we conducted a behavioural experiment in which subjects were presented with a visual stimulus and then provided both a continuous estimate of stimulus orientation and their confidence in this estimate (using a binary response scale). We manipulated task difficulty by varying stimulus contrast, dispersion, and duration. As expected, mean estimation error depended on each of these factors. Interestingly, ‘confident’ estimates were systematically associated with lower estimation error than ‘not confident’ estimates. This was true across and within experimental conditions, revealing a robust link between perceptual confidence and estimation quality. We then asked whether standard models of decision confidence can capture these results. Surprisingly, the answer is no. When sensory noise is modelled as arising from a single stochastic process, the downstream confidence computation cannot reveal variations in estimation quality within the same experimental conditions. To capture such introspective insight, it is necessary to assume that (1) sensory measurements are corrupted by a doubly stochastic noise process, and (2) subjects have some knowledge of the second layer of variability. Our findings thus both enrich our understanding of the noise that limits the quality of perception, and of the metacognitive mechanisms that enable us to accomplish goals in spite of the resulting uncertainty.

Acknowledgements: UO1NS136338