Supplemental Material for Domain-General Object Recognition Predicts Human Ability to Tell Real From AI-Generated Faces

Supplemental information for "Domain-general object recognition predicts human ability to tell real from AI-generated faces"To investigate the effect of careless responses on our hypotheses, we compared models fitted to the data for the first 250 participants regardless of attention check errors to models fitted to the data for the 250 participants who committed no ACE (250NoACE).In study 1, these two samples overlap in that 218 of the participants are the same, but the 250NoACE sample replaced 32 participants who made 1 or more ACE, out of a total of 12 attention checks (22 participants made 1, 4 made 2, 2 made 3, 2 made 4, 1 made 6 and 1 made 9).Study 2 had 11 attention checks and 35 participants were replaced (22 participants made 1 ACE, 9 made 2, 1 made 3, 1 made 4 and 2 made 5).The data and code we make available at https://osf.io/82krc/(Study 1) and https://osf.io/cgbf5/(Study 2) provide all the details about models.Here, we provide summary tables about the models in the main article (the 250NoAce models) compared to the corresponding First250 model in each study.Note that we do not provide a First250 comparison for the M2 model in Study 1, because it was inadmissible, with some eigenvalues smaller than zero and negative variances.