Dr. Clare Sutherland of the University of Aberdeen is currently exploring whether individuals can be taught to identify synthetic human faces created by artificial intelligence. As generative tools like StyleGAN3 become increasingly sophisticated, the traditional method of identifying AI-generated content—such as looking for extra fingers or distorted earrings—is proving less effective. Researchers are now focusing on subtler perceptual qualities, including symmetry, proportionality, attractiveness, distinctiveness, and emotional expressiveness.
In studies conducted by teams in the UK, Australia, and Canada, participants were trained to notice these nuances. Initial findings show that after brief training sessions, individuals can improve their detection accuracy from around 40% to 80%. This process relies on developing a refined instinct rather than searching for specific technical flaws. While this skill is becoming vital for cybersecurity—given the rising prevalence of deepfakes in fraud and political espionage—experts note that AI models are also evolving, creating a continuous cycle of advancement between detection capabilities and machine generation.
