Racial Bias in AI Face Generation Enables Misinformation Risks
-
AI-generated white faces are perceived as more realistic than actual human faces in research studies. This effect is not seen with non-white faces.
-
The reason is that AI algorithms are disproportionately trained on white faces, leading to racial bias.
-
People are easily fooled by AI fake faces and confident in their (incorrect) judgements. This could enable misinformation and identity theft.
-
Physical differences still exist between AI and human faces, but people misinterpret these as signs of realism.
-
Researchers argue for greater AI transparency, public awareness campaigns, and tools to identify AI-generated faces to reduce risks.