The spatial and temporal construction of confidence in the visual scene

Abstract

Human subjects can report many items of a cluttered field a few hundred milliseconds after stimulus presentation. This memory decays rapidly and after a second only 3 or 4 items can be stored in working memory. Here we compared the dynamics of objective performance with a measure of subjective report and we observed that 1) Objective performance beyond explicit subjective reports (blindsight) was significantly more pronounced within a short temporal interval and within specific locations of the visual field which were robust across sessions 2) High confidence errors (false beliefs) were largely confined to a small spatial window neighboring the cue. The size of this window did not change in time 3) Subjective confidence showed a moderate but consistent decrease with time, independent of all other experimental factors. Our study allowed us to asses quantitatively the temporal and spatial access to an objective response and to subjective reports.

Figure 1. Experimental Design and reliability of measurements.

A circular array of twelve letters was presented during 106 ms. Participants fixated in a cross at the center of the array. After a delay (which varied randomly between 0 and 1 s), a small red circle (the cue) was presented in one of the locations of the array indicating the letter that had to be responded. Then, participants had to report with the mouse the confidence level of their response through an ad hoc bar placed at the center of the screen. The response ranged between 0% of Confidence (guessing) and 100% (completely certain). The distribution of confidence reports are shown for every session of each subject at the Top-Right, showing a very high reliability across sessions. Both subjects showed a robust exponential decay of performance with ISI (Bottom-Left).

Figure 2. Correlations and Dissociations between objective and subjective reports.

A) Confidence reports were grouped in four percentile groups of confidence (25, 50, 75 and 100% - blue, orange, yellow and green traces) for each individual session and participant. B) Subjective confidence distribution for incorrect responses (black dotted line). The blue and green traces indicate the low-confidence and high-confidence error trials. C) Distance between the responded letter and the cue when the responded letter was a distractor for high- confidence (green) and low-confidence (blue) errors. The inset shows that this distribution does not change for short and long ISIs. D) Mean performance increased with subject's confidence report. At low confidence, responses were more accurate for short ISI values. E and F) Control experiment in which subjects reported in two subsequent screens the confidence in the position of the seen letter relative to the cue and of the identity of the seen letter for correct (E) and error (F) trials.

Figure 3. Spatial biases in response distributions and subjective confidence.

A) Left panels: Mean performance (p*) across positions for each subject. Center and Rigth panels: Probability of responding to a specific position in the array (independently of the position of the cue). Center: Each column corresponds to a location in the array (Nth column refers to N o'clock position. Each line corresponds to a different session. Right: Data collapsed across all sessions. B) Top Panel: Probability of responding to the location (i, in columns) given that the cue was in location (j, in lines), P(i,j). The matrix P approaches the diagonal (correct responses) for increasing levels of subjective confidence. Bottom panel: P in a polar representation for the lowest and highest subjective confidence. Mutual information of the distributions of the position of the cue and of responded letters increases with confidence but does not saturate even for the highest confidence values.

Figure 4. The temporal evolution of confidence.

Time course of subjective confidence responses after separating responses in correct (blue), near (red) and far (purple) errors (in which the distance between the target and the cue was smaller/larger than three), and errors in which the responded letter was neither the target nor the distractor (black). The subjective confidence decreases with time when spatial factors are factored out.