According to new research from the Australian National University, AI-generated faces with white skin appear more realistic than human faces, raising concerns about the perpetuation of racial biases online. The study, published in the journal Psychological Science, reveals that most participants recognized white-skinned faces generated by AI as human, compared to the faces of real people.
The discrepancy between AI and human faces is due to the fact that AI algorithms are primarily trained on white-skinned faces. This has serious implications for people of color, as it can lead to AI technologies altering their skin color and eyes to make them resemble those of people of Caucasian descent. As a result, professional-looking photos created by AI reinforce racial biases.
To address these issues, researchers recommend greater transparency around AI and public awareness about its limitations. They argue that educating people about the perceived realism of AI faces could be especially helpful in counteracting biases and promoting diversity and inclusion in technology.