Visual judgement alone is no longer reliable in detecting Al faces – impacting social media, online dating, recruitment and other areas where profile pictures are used.
Even people with exceptional face-recognition abilities struggle to distinguish real from fake Al faces, according to a study in the British Journal of Psychology by University of New South Wales and Australian National University researchers.
This is because most people rely on outdated visual cues that no longer apply to Al faces.
“With Al-generated faces now almost impossible to distinguish from real ones, this misplaced confidence could make individuals and organisations more vulnerable to scammers, fraudsters and bad actors,” warn the researchers led by UNSW School of Psychology’s Dr James Dunn.
“Up until now, people have been confident of their ability to spot a fake face,” Dr Dunn said in a statement.
“But the faces created by the most advanced face-generation systems aren’t so easily detectable anymore.”
There were 125 participants in the study, including 36 people with exceptional face-recognition ability, and 89 control participants, who completed an online test in which they were shown a series of faces and asked to judge whether each image was real or Al-generated. Obvious visual flaws were screened out beforehand.
What interested the researchers was how readily those with exceptional face-recognition abilities were fooled.
While this group performed modestly better on average, their accuracy remained far below what they typically achieved when recognising real human faces.
There was also a substantial overlap between groups, with some that had exceptional abilities outperforming those with average skills.
“What was consistent was people’s confidence in their ability to spot an Al-generated face – even when that confidence wasn’t matched by their actual performance,” Dr Dunn said.
Much of that confidence stemmed from visual cues that were shaped by earlier, less sophisticated systems. Early Al-generated faces gave obvious clues like distorted teeth, glasses that merged into faces and ears that did not attach properly,
Now the most realistic pictures no longer show obvious flaws – leaving faces that look convincing and far harder to judge using once-familiar cues.
ANU psychologist Associate Professor Amy Dawell said ironically Al faces were not revealed by looking at what was wrong with them – but by what was too right.
“Rather than obvious glitches, they tend to be unusually average – highly symmetrical, well-proportioned, and statistically typical,” Associate Prof Dawell said.
Qualities such as symmetry and average proportions usually signal attractiveness and familiarity. But in the current study, they become a red flag for artificiality.
“It’s almost as if they’re too good to be true,” Associate Prof Dawell said.
“As face-generation technology continues to improve, the gap between what looks plausible and what is real may widen – and recognising the limits of our own judgement will become increasingly important.”
