Do you think you can easily recognize AI generated faces? A new study suggests you may be overestimating yourself.
Researchers from the University of New South Wales (UNSW) and the Australian National University (ANU) tested 125 participants on their ability to distinguish real human faces from AI-generated ones. Most volunteers were confident in their judgment, but the results told a different story. Even people with exceptional facial recognition skills, so-called super-recognizers, performed only slightly above average.
The study used AI-generated faces that eliminated obvious visual defects, making them extremely convincing. While early specimens often had telltale defects such as distorted teeth or mismatched glasses, modern imaging systems produce symmetrical, well-proportioned images that are much harder to see.
“We saw that people with average facial recognition ability performed only slightly better than chance,” says Dr. James Dunn from UNSW. “And while super-recognizers performed better than other participants, it was only just. What was consistent was people’s confidence in their ability to recognize an AI-generated face – even if that confidence didn’t match their actual performance.”
How overconfidence can endanger you
This overconfidence has real-world implications, making people and companies more vulnerable to scams, fake social media accounts, and fraudulent profiles when they rely solely on instinct to judge authenticity. Research shows that gut feeling is no longer enough to distinguish real from artificial intelligence.
The study shows that even people with above-average facial recognition skills can have difficulty identifying AI-generated faces, and there is still no clear evidence that this skill can be reliably trained or predicted.
If you thought it was easy to spot a fake face, this study shows it’s now a lot harder than you think.
