A new study from researchers at The Australian National University reveals an alarming trend in artificial intelligence. AI-generated images of white faces now appear more realistic than actual photos of real people.
The study found that participants were more likely to think white faces created by AI algorithms were human, compared to images of real individuals. This effect was not seen for faces of people of color, indicating issues with racial bias in the training data for these AI systems.
According to senior author Dr. Amy Dawel, the lack of diversity in the data used to develop these algorithms leads them to produce hyper-realistic results for white faces, while distorting features of non-white faces. The technology has already been used in problematic ways, like creating AI headshots that alter skin and eye color of people of color to resemble white individuals.
Surprisingly, the research also found that people who incorrectly judged the AI faces as real expressed the most confidence in their judgements. Lead author Elizabeth Miller explains, “People who are fooled by AI imposters don’t realize they’re being tricked.”
The study uncovered why AI faces achieve this misleading level of realism. While subtle physical differences remain between AI and human faces, people often misread these cues, perceiving features like perfect proportions as more natural.
Dr. Dawel warns, “As the technology advances, these differences will likely disappear entirely.” This could enable new forms of misinformation and identity theft if actions are not taken.
The authors argue for greater transparency around the development of AI systems, so problems can be addressed proactively. Raising public awareness is also key, as humans can no longer reliably distinguish AI from real faces.
Dr. Dawel concludes, “Society needs tools that can accurately detect AI impersonators. And people must be educated about the perception of realism in AI faces, so they maintain healthy skepticism about images online.”
The escalating realism of AI faces presents complex challenges at the intersection of technology, ethics, and human perception. A measured, thoughtful approach is required to steer the technology toward benefit rather than harm. Continue reading in Sage Journals (link).