Can fake faces make AI training more ethical?

Researchers are exploring synthetic faces to ethically train facial recognition AI. While current AI has nearly 99.9% accuracy, it relied on scraping real faces without consent, raising privacy concerns. The new approach generates fake but realistic faces, potentially eliminating privacy issues while maintaining fairness. Models trained on synthetic data show lower accuracy (75% vs 85% with real faces) but more consistent performance across racial groups. Combining synthetic data with consensual real images could provide both privacy protection and high accuracy.