AI Study Reveals Realistic Faces Can Mislead People
A new study conducted by specialists from the Australian National University (ANU) has discovered that artificial intelligence (AI) can create more realistic faces of white people compared to real human faces. This finding has raised concerns about racial prejudices and the potential dissemination of misinformation through AI technologies.
During the study, participants overwhelmingly perceived white faces generated by AI as more realistic compared to faces of individuals with different skin colors. Dr. Amy Dowell, the senior author of the study, suggests that this bias might stem from the fact that AI algorithms have primarily been trained on datasets with predominantly white individuals.
The phenomenon of “hyperrealism” in AI-generated faces can have serious consequences, particularly for people with different skin colors, as it can amplify racial biases on the internet. The issue is already evident in current AI technologies used for creating professional portraits, as AI often alters the skin and eye color of individuals with dark skin to appear lighter.
Researchers have also discovered that individuals often fail to recognize that they are being deceived by AI-generated faces. People who perceive these faces as real are often excessively confident in their judgments, unaware that they are mistaken. The study revealed that there are discernible physical differences between AI-generated faces and real human faces, but people interpret these differences incorrectly. For instance, the AI-generated faces tend to have more proportional features, which is falsely perceived as a sign of their authenticity.
Researchers caution that this trend could contribute to the proliferation of misinformation and identity theft. They emphasize the need for greater transparency in AI technologies to enable researchers and the general public to identify potential problems at an early stage. Increasing public awareness and educating people about the realism of AI-generated faces can help society cultivate a critical approach to the images encountered online.
The study’s findings have been published in the journal Psychological Science, which is owned by the Association for Psychological Science.