News release

From: The Royal Society

(A)I spy - Super recognisers can be trained to be better at spotting AI generated faces. Researchers have found that, without training, super recognisers identify AI generated images of faces at chance level, while a typical viewer identifies synthetic faces significantly below chance frequency. Both groups significantly improved their results after being trained, putting the super recognisers above chance frequency. The researchers suggest that trained super-recognisers could be used for real-world applications, such as online identity verification. Royal Society Open Science

Training human super-recognisers' detection and discrimination of AI-generated faces

Royal Society Open Science

AI-generated faces are difficult to detect and are often judged as more realistic than real faces. We measured whether participants could accurately judge faces as 'real' or 'not real'. Our participants were super-recognisers (people who have exceptional face recognition skills), and control participants (people who have typical face recognition skills). We found that super-recognisers were better at detecting the synthetic faces than control participants, who were at chance at best. With a short training procedure, super-recognisers performed above chance, suggesting that they may be useful in real-world synthetic face detection scenarios