One study found that AI-generated white faces were perceived as more realistic than real human faces, and there were significant differences in the realism of AI faces for people of color. This trend is believed to be due to bias in AI training, raising concerns about reinforcing racial bias and spreading misinformation. Credit: SciTechDaily.com
A study reveals that AI-generated white faces are more realistic than real human faces, raising concerns about potential racial bias and misinformation in AI technology.
Artificial intelligence (AI) has reached a point where white faces created by AI now appear more real than human faces, according to a study conducted by experts at the Australian National University (ANU).
This study found that more people perceived AI-generated white faces as human compared to real human faces, with a different outcome for images of people of color.
Dr. Amy Dowell, the lead author, explained that the disproportionate training of AI algorithms on white faces contributed to this disparity.
Impact of AI Realism
Dr. Dowell expressed concern about the potential impact of consistently perceiving white AI faces as more realistic, especially in reinforcing racial bias online and its impact on people of color.
This image was generated by AI, specifically Midjourney V5.2. Credit: SciTechDaily.com
Understanding AI “Hyperrealism”
Researchers pointed out the problem of AI’s “hyperrealism,” where people often mistake AI faces for real human faces without realizing it.
The study also identified physical differences between AI and human faces that people tend to misinterpret, highlighting the need for transparency in AI technology.
Potential Consequences
This trend has serious implications for the prevalence of misinformation and identity theft, and the researchers emphasize the importance of increasing transparency around AI technologies and raising public awareness. Source: Psychological Science, Journal of the Psychological Science Association.
Reference: “AI Hyperrealism: Why AI faces are perceived as more realistic than human faces” Elizabeth J. Miller, Ben A. Steward, Zach Witkower, Claire AM Sutherland, Eva G. Kramhuber , by Amy Dowell, November 12, 2023; Psychological Science. DOI: 10.1177/09567976231207095
Source: scitechdaily.com
