Fig. 3
From: Experts fail to reliably detect AI-generated histological data
Response times are shorter for correct classifications. (a) Empirical cumulative density functions (CDFs) of response times for correctly vs. incorrectly classified images (ncorrect = 8364, nincorrect = 4474, Scheirer-Ray-Hare test: pcorrectness < 0.001). Inset displays full distributions. (b) CDFs of response times for naive vs. expert participants (nnaive = 4526, nexpert = 8312, Scheirer-Ray-Hare test: pexperience < 0.001). Inset displays full distributions. (c) CDFs for group wise comparisons (nnaive,correct = 2531, nnaive,incorrect = 1995, nexpert,correct = 5833, nexpert,incorrect = 2479, post-hoc Wilcoxon rank sum test with Bonferroni correction: pn,c-n,i = 0.0015, pn,c-e,c < 0.001, pn,c-e,i < 0.001, pn,i-e,c = 1, pn,i-e,i < 0.001, pe,c-e,i < 0.001). (d) CDFs of within-participant response times for correct and incorrect responses (nparticipants = 803, Wilcoxon signed rank test p < 0.001). Inset shows the same data as boxplots. Data beyond 30 s not shown for display purposes.
