Study: People are overconfident they can tell AI-made faces from real


Gone are the give-away glitches like strange hair and 12 fingers. The AI-generated photos of faces, increasingly seen on social media, are better than ever and even fooling experts. — Elisa Schu/dpa

SYDNEY: As artificial intelligence-fuelled scams and fraud become increasingly prevalent, most people remain far too confident of their ability to spot a fake face, going by Australian research that asked people to say if photos were real or AI-generated.

Carried out by the University of New South Wales in Sydney and Canberra's Australian National University, the study found "out of date" assumptions about AI among the 125 participants, some of whom were selected for their "exceptional face-recognition ability."

In previous years, give-away glitches such as strange hair, overly smooth features and other so-called "artefacts" in AI-generated photos of faces would give them away, however advances with AI have led to increasingly realistic looking images of people.

So-called "super recognisers" now prove to be no more than a hair’s breadth more competent than the average person, the researchers say.

"What was consistent was people’s confidence in their ability to spot an AI-generated face – even when that confidence wasn’t matched by their actual performance," said UNSW’s James Dunn.

"The faces created by the most advanced face-generation systems aren’t so easily detectable anymore," Dunn explained.

With AI-fuelled con artistry a growing problem, the researchers warned that the mix of hard-to-catch fakes and "misplaced confidence" will leave people and companies more vulnerable to scammers, fraudsters and bad actors.

Rather than looking out for Cyclops foreheads or triple nostrils, people should instead be vigilant for facial imagery that is almost too good to be true.

"The most advanced AI faces aren’t given away by what’s wrong with them, but by what’s too right," said Amy Dawel of ANU.

"Rather than obvious glitches, they tend to be unusually average – highly symmetrical, well-proportioned and statistically typical," Dawel explained.

"As AI-generated identities become more prevalent in daily life, understanding who can detect them and why will be vital for both cognitive science and societal resilience," the researchers concluded, in a paper published by the British Journal of Psychology in February. – dpa

Follow us on our official WhatsApp channel for breaking news alerts and key updates!

Next In Tech News

OpenAI reaches deal to deploy AI models on U.S. Department of War classified network
Are password managers safe? Not as much as you think, research shows
Most AI chatbots have murky safety provisions, researchers find
Anthropic says it will challenge Pentagon's supply chain risk designation in court
Canadian minister to meet with OpenAI's Altman to discuss safety measures after shooting
Trump says he is directing federal agencies to cease use of Anthropic technology
Exclusive-QIA, Visa and ADIA set to anchor SoftBank's PayPay IPO, sources say
Goldman bucks private credit redemption trend as AI disruption fears mount
Brookfield's new AI unit Radiant valued at $1.3 billion after merger with UK startup, sources say
Trump Media considers spinning off Truth Social into public company, reports wider annual loss

Others Also Read