Fig. 1: Experiment 1.

A Experimental procedure. Participants fixated on a cross superimposed on scrambled frame images. An embedded face stimulus was presented for three consecutive 200 ms frames at a random time starting 600 ms to 1800 ms after trial onset. Faces are replaced by a white placeholder in the figure to satisfy Nature’s policy about AI-generated images. When participants pressed the button between 400 ms and 1500 ms after face stimulus onset, the trial was considered a hit (trials with response times <400 ms or >1500 ms were excluded, see C and methods). When participants did not press, the trial was considered a miss. B Average and individual detection rates for all participants (n = 30), both in catch trials (black) where no target was presented and in threshold trials (blue) where the target was presented at participants’ detection threshold. C Mean percentage of responses as a function of reaction times (100 ms bins) across participants (n = 30). Dashed vertical lines show the 400 ms and 1500 ms limits for considering the trial as a hit (blue bars). Early responses (black bars) were not analyzed. D Analyzed channels normalized to an MNI template, with channels in the ventral visual cortex (VVC, dark blue), superior parietal cortex (SPC, light blue), inferior frontal cortex (IFC, green) and dorsolateral prefrontal cortex (DLPFC, yellow) and in other brain regions (black). E Total number of channels in the four ROIs. The box with black contours shows the proportion of responsive channels, defined as channels showing a significant positive difference between hit and miss trials. The red line indicates the maximum number of responsive channels expected by chance (α = 0.05). F Cross-validated average area under the receiving operator characteristic (AUROC) for the hit vs. miss decoder in the four ROIs (bar plot: average; dots: individual participants). Red lines indicate group-level chance-level performance, as assessed by permuting labels across trials (VVC: n = 18; SPC: n = 5; IFC: n = 19; DLPFC: n = 9). Red dot surrounds indicate participants who displayed above chance-level performance after false discovery rate correction. All error bars indicate 95% confidence interval.