AI-generated deepfake faces are increasingly realistic and difficult to spot until now in a new study that suggests humans have been successfully trained to spot the computer-generated faces from photos of real humans.
The UVic Different Minds Lab team, led by psychology professor Jim Tanaka and post-doctoral fellow Eric Mah, partnered with the Australian National University (ANU) to develop a quick, effective and robust training technique for improving people's detection of deepfake faces, in research published today by scientific journal Proceedings of the National Academy of Sciences.
AI-generated deepfake faces have become so realistic that it is difficult for people to tell them apart from photos of real humans, contributing to increases in AI-related fraud. The researchers trained people to spot AI-generated faces by drawing their attention to six perceptual qualities: distinctiveness, memorability, proportionality, symmetry, attractiveness and expressiveness.
"It was amazing to see the dramatic improvement in people's ability to detect AI faces,'' says Amy Dawel, associate professor and director of the ANU Emotions and Faces Lab. "We've shown our training is effective for some of the most convincing fakes available, StyleGAN faces. Now we need to find out whether that training generalises to other AI-generated faces."
UVic researchers replicated the training method. "The replication showed that the findings weren't a fluke when we trained a new set of people in a different country, we saw them improve just as much," says Mah.
"What surprised us was how quickly people improvedin less than an hour of training, AI face detection accuracy increased by nearly 30 per cent, and participants not only became more accurate, they became faster too," adds Tanaka, who is director of UVic's Different Minds Lab.
"Our results show that AI detection can be trained up like other forms of perceptual expertise."
The researchers point out that AI image-generation technology is improving extremely quickly, and many people underestimate how convincing these faces can be, so the hope is this work can help people navigate increasingly complex online environments.
Learn more about the ANU Emotions and Faces Lab and ways to participate in their studies.
Read the study Training Humans to Detect AI-generated Faces.
