With the rapid advances in artificial intelligence, computer-generated images, including hyperrealistic faces, are becoming increasingly common. Many of these images are so convincing that it can be difficult to distinguish them from real photographs.
A recent study tested people’s ability to identify real faces versus AI-generated ones, revealing that most participants missed a majority of the AI-generated faces. Even “super-recognizers,” individuals known for their exceptional facial-processing skills, could correctly identify fake faces only 41% of the time, while the general participants identified only about 30%.
However, the study indicated that participants’ ability to detect fake faces improved after just five minutes of training. This training focused on common rendering errors, such as unnatural skin textures and odd hair placement. Following the training, super-recognizers identified 64% of fake faces, and typical recognizers identified 51%.
Given the challenges faced even by skilled participants, there are questions about individual confidence in spotting AI-generated faces.
—
