Tracking Perceptual Uncertainty in Rapid Serial Visual Presentations

JOAQUIN NAVAJAS; HARPAL PANESAR; URI HERTZ; MEHDI KERAMATI; BAHADOR BAHRAMI

Liverpool

Conferencia; European Conference of Visual Perception; 2015

Optimal multisensory integration (Ernst & Banks, 2002) and dyadic interaction (Bahrami et al., 2010) rely on computing the uncertainty of visual information. But this is at odds with decision-making models disregarding evidence reliability (Ratcliff & McKoon, 2009) and with data showing that variance is severely underweighted in confidence judgments (Zylberberg, Roelfsema & Sigman, 2014). Here, we asked whether subjects can track the reliability of serially presented visual information and how it guides choice and confidence. Participants (N=20) observed a sequence of 30 tilted Gabor patches in rapid serial visual presentation at 4 Hz. The orientation of the patches was drawn from uniform distributions with different variance. Subjects were instructed to report the grand-average tilt of the patches compared to the vertical meridian (clockwise or counter-clockwise) and to state their confidence (from 1 to 6). We observed that objective and subjective performance decreased with increasing variance. We explain these results using a learning model that updates both mean and variance but with different rates. We propose rules for combining these quantities in order to reproduce the observed patterns of choice and confidence. Overall, our results provide novel insights on how uncertainty is tracked by the visual system and communicated as confidence.