A recent study by researchers from The Australian National University (ANU) and the University of Aberdeen uncovered that AI-generated White faces are frequently mistaken for real human faces, raising concerns about racial biases in AI technology and its potential implications for misinformation and identity theft.
AI Faces Perceived as More Real than Human Faces: A Study Overview
On November 14, 2023, researchers from The Australian National University (ANU) and the University of Aberdeen released a study revealing that AI-generated White faces are often perceived as more realistic than actual human faces. The study found that participants were more likely to mistake AI-generated White faces as human than those of real people, a trend not observed for faces of people of color.
Dr. Amy Dawel, the study’s senior author, highlighted that AI algorithms are predominantly trained on White faces, leading to this discrepancy. This bias raises concerns about reinforcing racial biases through AI technology, especially as it’s being used to create professional headshots that can alter the appearance of people of color.
Elizabeth Miller, a PhD candidate at ANU and co-author of the study, noted that individuals who mistakenly identified AI faces as real were paradoxically confident in their judgments. Dr. Dawel added that while physical differences currently exist between AI and human faces, these differences are diminishing rapidly due to advances in AI technology.
The study, published in Psychological Science, underscores the potential implications for misinformation and identity theft. Dr. Clare Sutherland from the University of Aberdeen, a co-author, emphasized the necessity of inclusive AI development to prevent disadvantaging any group based on ethnicity, gender, or age. The researchers advocate for greater transparency in AI technologies and public education to mitigate risks associated with hyper-realistic AI faces.
