Examining the effects of age and personal characteristics on deepfake detection: an eye-tracking study

Each experiment included a set of real and deepfake images and videos. Experiment 1 examined deepfake discrimination in younger and older adults, using StyleGAN-2 generated face images; Experiment 2 examined deepfake discrimination in younger adults, using Stable Diffusion generated face images.

The videos in both datasets were generated using LTX studio. Real face images in both experiments were selected from the FFHQ database; real face videos in both experiments were selected from the deepfake detection challenge (DFDC) and FaceForensics databases. </p><p dir="ltr">Both datasets include: </p><p dir="ltr">1) behavioural data (accuracy and confidence ratings) in a deepfake detection task</p><p dir="ltr">2) individual differences data (AI Literacy scale; Trust in Automation scale; deepfake exposure OFCOM questionnaire; face recognition ability - PI20; autistic traits - AQ-10; impulsivity - BIS-11; a bespoke deepfake detection strategy questionnaire for images and videos)</p><p dir="ltr">3) eye movement data (fixations and dwell time to pre-defined areas of interest; individual fixation location and duration data)</p>