Face the facts.
AI-generated faces appear more “real” than some humans, according to a new study, at least when it comes to white people.
Published in the journal Psychological Sciences This week, researchers found that AI could reliably trick people into thinking that computer-generated faces look more real than a photograph.
“Surprisingly, white AI faces can convincingly pass off as more real than human faces, and people don’t realize they are being fooled,” researchers from the University of Amsterdam team said in the findings. your study.
The researchers said the problem with this type of “hyperrealism” is that people are convinced that their judgment about what is real and what is not is airtight.
“Worryingly, the people who thought the AI faces were real most often were, paradoxically, the most confident that their judgments were correct,” said study co-author Elizabeth Miller, a Ph.D. candidate at the Australian National University he said in a press release.
“This means that people who mistake AI imposters for real people don’t know they are being deceived.”
However, these findings only applied to images of white faces, not to photographs or AI images of people of color.
This is potentially because the AI algorithm used to make AI renderings “is disproportionately trained on white faces,” Amy Dawel, a professor at ANU and lead author of the study, said in the statement.
This can create a whole new type of racial bias online.
“If white AI faces are consistently perceived as more realistic, this technology could have serious implications for people of color by ultimately reinforcing racial biases online,” Dawel said.
“This issue is already evident in current AI technologies used to create professional-looking headshots. When used on people of color, the AI alters their skin and eye color to that of white people.”
The discrepancy between real human faces and what AI represents is counterintuitive to what we might think humans should look like. As Dawel explains, people “tend to misinterpret” natural physical asymmetry as some kind of flaw in AI, thus mistakenly choosing the most perfect digital faces as legitimate shapes. “AI white faces tend to be more proportionate and people mistake this as a sign of humanity,” Dawel said.
“However, we cannot rely on these physical signals for long. “AI technology is advancing so fast that the differences between AI and human faces will probably disappear soon,” he further warned.
Additionally, due to the fact that AI algorithms are trained on mostly white faces, the AI is able to create a more “normal” looking face, making the real and the fake more difficult to distinguish.
This “may lead to white AI faces appearing especially normal and therefore potentially especially realistic,” the study reported.
