Unsophisticated AI coaching does not improve performance on symptom and performance validity tests in a Romanian sample of experimental feigners

The present study aimed to investigate the ability of symptom and performance validity indicators to detect experimental feigning with symptom-coaching vs. AI assistance. Fifty-nine undergraduates and community members (45 females, M_Age=22.6, SD_Age=2.1; M_Education=14.5, SD_Education=1.4) were randomized into three experimental groups: controls (n = 20), symptom-coached feigners (n = 20), and AI-coached feigners (n = 19). The two feigning groups were instructed to feign a mild traumatic brain injury (TBI) for evading work responsibilities. All participants were assessed online with the Inventory of Problems-29 and its memory module (IOP-29-M) and the Memory of Objects and Digits and Examination of Memory Malingering Test (MODEMM). Results indicated that AI-coached participants exaggerated symptomatology and underperformed significantly more compared to the symptom-coached group. All validity indicators of the IOP-29-M and MODEMM discriminated between experimental feigners and control participants at previously reported cutoffs with high sensitivities (.69-1.00) and excellent specificities (.95-1.00). Our findings contribute to nascent research on feigning with AI assistance, also providing proof of concept for the accuracy of the IOP-29-M and MODEMM in detecting symptom-coached and AI-coached feigning.