The researchers emphasised that surprisingly caucasian portraits generated by artificial intelligence “can convincingly pass as more real than human faces – and people do not realise they are being fooled”.
A team comprised of scientists from Australia, the UK, and the Netherlands, stressed the significant real-world implications of their findings, particularly in areas such as identity theft, where individuals could unwittingly fall prey to digital impostors.
The researchers warned the results may not apply to images of people of colour, potentially due to the predominantly white-centric training of the AI face-generation algorithm.
Dr Zak Witkower, a co-author of the study from the University of Amsterdam, underscored the potential ramifications of this bias, noting its impact on various domains from online therapy to robotics.
He said: “It’s going to produce more realistic situations for white faces than other race faces.”
The scientists’ study, published in the Psychological Science journal, details two experiments conducted by the team.
In one, white adults assessed a mix of 100 AI-generated white faces and 100 real white faces.
The participants were tasked with determining whether each face was artificially generated or real, rating their confidence on a 100-point scale.
Astonishingly, the results from 124 participants reveal that 66 per cent of AI images were deemed human, compared to 51 per cent of actual images.
The researchers also revisited data from a previous study, and concluded people were consistently more likely to perceive white AI faces as human than real white faces.
Notably, the trend did not hold for people of colour, where around 51 per cent of both AI and real faces were judged as human.
The team asserted participant race did not influence those results.