University of Victoria study improves public detection of AI-generated faces

Researchers at the University of Victoria’s Different Minds Lab, led by psychology professor Jim Tanaka and post‑doctoral fellow Eric Mah, partnered with the Australian National University to develop a rapid training program for spotting AI‑generated deepfake faces. The study, published in the Proceedings of the National Academy of Sciences, taught participants to focus on six perceptual qualities—distinctiveness, memorability, proportionality, symmetry, attractiveness and expressiveness.

In trials, participants’ accuracy at identifying StyleGAN‑generated faces rose by nearly 30 % after less than an hour of training, and they also responded more quickly. “It was amazing to see the dramatic improvement in people’s ability to detect AI faces,” noted ANU associate professor Amy Dawel. Tanaka added, “Our results show that AI detection can be trained up like other forms of perceptual expertise.” The researchers hope the technique can help the public counter increasingly convincing AI‑fraud.

The work highlights a practical education tool as AI image‑generation technology continues to improve, offering a potential safeguard against deepfake‑related scams.