Training humans to detect AI-generated faces
Proceedings of the National Academy of Sciences · 07.07.2026 · 123 (27)
DOI: 10.1073/pnas.2602122123 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the National Academy of Sciences, online erschienen am 29.06.2026. Verfasst von Amy Dawel und 6 weiteren Autor:innen. Beteiligte Einrichtung: School of Medicine and Psychology, The Australian National University, Canberra, ACT 2600, Australia. Hauptthemen laut MEDLINE: Face, Artificial Intelligence, Facial Recognition. Die Arbeit wurde bisher 4-mal zitiert (Crossref) und nennt 16 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Department of Education and Training | Australian Research Council (ARC).
Abstract (Originalsprache)
As AI-generated faces become indistinguishable from real ones, deepfake technology poses escalating threats to information integrity and security. While algorithms can detect deepfakes, they suffer from opacity and critical vulnerabilities—and training humans to identify specific visual artifacts has proven largely ineffective. Here, we introduce a fundamentally different approach that harnesses people’s global impressions of faces. Building on findings that AI and human faces evoke systematically different perceptual impressions (Miller et al., 2023), we trained participants to attend to these distinguishing qualities without explicit instruction on how to use them. Using a rigorous pre–post design with untrained test faces, we demonstrate that all participants ( N = 45) improved, with mean accuracy nearly doubling. High performers achieved near-perfect detection, and participants developed metacognitive insight, showing appropriate confidence calibration only after training. A test–retest control study ruled out practice effects as an explanation for training gains, and an online replication demonstrates the scalability of our approach. As the technology advances, targeting systematic biases that are inherent to generative AI, and which manifest in global features, may offer a more durable defense than approaches reliant on image artifacts alone.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- School of Medicine and Psychology, The Australian National University, Canberra, ACT 2600, Australia.
- Department of Psychology, University of Victoria, Victoria, BC V8P5C2, Canada.
- School of Psychology, University of New South Wales Sydney, Kensington, NSW 2052, Australia.
- School of Psychology, University of Aberdeen, King's College, Aberdeen AB243FX, Scotland.
- School of Psychological Science, University of Western Australia, Crawley, WA 6009, Australia.
ORCID-Profile: Amy Dawel, Eric Y Mah, James D Dunn, Clare A M Sutherland, Nick Argument, James W Tanaka
Förderung
- Department of Education and Training | Australian Research Council (ARC)
Zitieren
APA
Dawel, A., George, T., Mah, E. Y., Dunn, J. D., Sutherland, C. A. M., Argument, N., & Tanaka, J. W. (2026). Training humans to detect AI-generated faces. Proceedings of the National Academy of Sciences, 123(27). https://doi.org/10.1073/pnas.2602122123
Vancouver
Dawel A, George T, Mah EY, Dunn JD, Sutherland CAM, Argument N, et al. Training humans to detect AI-generated faces. Proceedings of the National Academy of Sciences. 2026;123(27). doi:10.1073/pnas.2602122123
BibTeX
@article{Dawel2026,
author = {Dawel, Amy and George, Tanya and Mah, Eric Y and Dunn, James D and Sutherland, Clare A M and Argument, Nick and Tanaka, James W},
title = {Training humans to detect AI-generated faces},
journal = {Proceedings of the National Academy of Sciences},
year = {2026},
volume = {123},
number = {27},
doi = {10.1073/pnas.2602122123}
}
EUVIMED