SpaceX

December 27, 2025, 4:06 PM EST. New research shows AI-generated faces can be indistinguishable from real ones to untrained eyes. Generative adversarial networks (GANs) produce convincing synthetic portraits, raising concerns about nefarious uses such as deepfakes and misinformation. A Royal Society Open Science study found people struggle to distinguish human from AI faces, but a five-minute training regimen can boost detection accuracy by teaching viewers to spot common rendering glitches (e.g., extra teeth, odd hairlines, unusual skin textures). In experiments, typical recognizers improved from around 30% to about 60% accuracy after training; even super recognizers benefited. The work suggests practical countermeasures against AI deception online.