Fig. 2
From: Experts fail to reliably detect AI-generated histological data
Experts are more successful in discriminating genuine from artificial images. (a) Overall performance of naive and expert participants (ntrials,naive = 4640, ntrials,expert = 8416, Chi2-Test with Yates’ Continuity correction: p < 0.001). (b) Sensitivity and specificity for naive and expert participants (point estimates and 95% confidence intervals (CI)). (c) Participant performance by image type and prior experience (ntrials,naive = 4640, ntrials,expert = 8416). (d) Results for binomial regression on data shown in (c), displayed as odds ratio (OR) point estimate with 95% CI. “::” indicates statistical interactions. (e) CDF of participant-wise performance by prior experience. Red arrow indicates experts who classified all images correctly. Inset shows the same data as boxplot. nparticipants,expert = 525, nparticipants,naive = 288, Wilcoxon rank sum test p < 0.001.
