Artificial intelligence makes faces look more real

Hours after the European Union announced its first law governing artificial intelligence (AI), Australian scientists have published a study involving more than 600 volunteers that says AI-generated white faces now look more real than human faces. .

They used face space theory and qualitative reports from 610 participants to identify key facial attributes that differentiate AI from human faces. They explain that these volunteers misinterpreted “what led to AI hyperrealism. However, these attributes allowed for high accuracy through machine learning.”

This paper from the Australian National University also states that psychological theory “can contribute to understanding the results of AI and provide guidance on how to eliminate bias in algorithms, promoting the ethical use of this technology.”

Published by Psychological Science, in this study The same cannot be said for depictions of people of color.

The reason for the discrepancy, emphasizes Dr. Amy Dowell, lead author, is that AI algorithms are disproportionately trained on white faces. “If white AI faces are perceived as more realistic, this technology,” he concludes, “could have serious consequences for people of color, ultimately reinforcing racial prejudice.”

This clinical psychologist stresses the need for more transparency around AI, “so that researchers and civil society can identify problems before they become major problems.”