Researchers developed short training to improves people's detection of AI faces (www.southampton.ac.uk)

Researchers at the University of Southampton have developed a brief training program, named DISCERN-AI, designed to enhance people's ability to identify AI-generated faces. This initiative comes in response to the increasing sophistication of deepfake technology, which can create hyper-realistic images that often confuse individuals more than actual human pictures. The study, published in Computers in Human Behavior, demonstrated the training's effectiveness even 20 days later, significantly improving participants' detection accuracy from below chance to above chance performance. The DISCERN-AI training consists of three segments that challenge common misconceptions about face recognition. Participants learn to identify deceptive physical traits often associated with AI faces, recognize useful but overlooked markers, and disregard misleading "red flags" typically thought to indicate inauthenticity. While human trainees showed marked improvement in identifying AI-generated images, computer models trained on these same cues achieved a striking 94% accuracy. This research not only underscores the capability for humans to be trained to better discern AI creations, but it also highlights the ongoing challenges posed by advanced deepfake technologies, making the findings crucial for the AI/ML community. Plans to make DISCERN-AI widely accessible online are underway, potentially equipping more individuals with the skills to navigate an increasingly AI-driven digital landscape.