This paper investigates how human biases affect the detection of deepfakes, a form of diverse misinformation. An observational survey (N=2016) exposed participants to videos, assessing their ability to identify deepfakes. Results show accuracy varies by demographics, with participants better at classifying videos matching their own demographics. A mathematical model explores population-level impacts, suggesting diverse social groups may offer "diverse correction," where friends protect each other from misinformation.

Juniper Lovato, Jonathan St-Onge, Randall Harp, Gabriela Salazar Lopez, Sean P. Rogers, Ijaz UI Haq, Laurent Hébert-Dufresne, Jeremiah Onaolapo