Artificial intelligence (AI) is getting so good, it’s sometimes hard to tell what’s real – and what’s not.
After all, who among us hasn’t been nearly caught out by the odd deepfake?
New research has confirmed our suspicions – but there’s a catch.
Advert
The study from the Australian National University (ANU) found that AI can generate Caucasian faces that seem more realistic than actual human faces – but the same did not apply to people of color.
According to experts, this is down to AI algorithms being trained disproportionately on white faces.
“If white AI faces are consistently perceived as more realistic, this technology could have profound implications for people of color by ultimately reinforcing racial biases online,” said senior author of the paper, Dr Amy Dawel.
Advert
“This problem is already apparent in current AI technologies used to create professional-looking headshots.
“When used for people of color, the AI is altering their skin and eye color to those of white people.”
The study also found that with hyper-realistic faces, most of us don’t even realise when we’re being fooled.
Apparently study participants who were most confident they were looking at a human, were actually picking out AI faces. Awkward.
Advert
If you did want a cheat code, the best way to suss out if someone is AI or human is to look at the proportions of the face – the more in-proportion it is, the more likely AI has generated it.
But AI is getting smarter at a dizzying pace, and Dawel said: “We can’t rely on these physical cues for long.
“AI technology is advancing so quickly that the differences between AI and human faces will probably disappear soon.”