Deepfake faces generated by artificial intelligence (AI) are now so realistic they can be nearly impossible to distinguish from photographs of real humans, fuelling misinformation and AI-related fraud.

Researchers at the Australian National University’s (ANU) Emotions and Faces Lab have shown that people can be trained to spot AI-generated faces by drawing their attention to 6 key qualities.

The research is published in the journal PNAS.

Amy Dawel, a clinical and cognitive psychologist at ANU and lead author of the study, says it is important to improve human AI-detection abilities because AI cannot be relied upon to solve the problem alone.

"While algorithms offer one solution to detecting deepfake faces, their decision-making processes remain opaque and recent benchmarking reveals serious weaknesses. We need approaches that are ethical and explainable – for which keeping humans in the loop is key.

“Training on visual artifacts, like looking for a sixth finger or odd earrings, has had limited success, partly because the AI is getting too good and fraudsters may avoid using pictures with obvious flaws anyway.

“Our training directs people’s attention to global qualities that differ between AI and human faces. AI faces tend to be more symmetrical, proportional and attractive, but without training we often think these are markers of being human.”

Training also draws people’s attention to the fact that AI-generated faces as less distinctive, memorable and expressive.

The researchers explain that these “global impressions” arise because generative AI algorithms are “inherently biased toward the mathematical average of the tens of thousands of faces on which they are trained”.

Participants discovered these patterns through a guided experience in which they rated AI-generated and human faces on the 6 qualities: distinctiveness, memorability, proportionality, symmetry, attractiveness and expressiveness.

“We found that even relatively short training sessions helped participants improve their accuracy in detecting AI-generated faces, highlighting the potential for practical education tools in this area,’’ says ANU honours student and study co-author Tanya George.

Some particularly high performing participants achieved near-perfect accuracy.

“AI image-generation technology is improving extremely quickly, and many people underestimate how convincing these faces can be. Research like this can help people navigate increasingly complex online environments,” says George.

The training was replicated by researchers in Canada and shown to be just as effective.

“The replication shows 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 co-author Eric Mah of the University of Victoria’s department of psychology.

“Online training was effective, so our training program could easily be implemented at scale for little cost.”

Because the training targets biases in AI-generated content, rather than specific visual artifacts, the approach can adapt as generation techniques change over time by updating the training images used.

The authors say that future research should establish optimal training protocols, assess if AI-detection skills are retained in the long term and determine whether training generalises to other formats such as audio and video deepfakes.

Registrations are open to participate in the ANU AI Faces Study.