With the rapid advances in artificial intelligence, computer-generated images — including hyperrealistic faces — are becoming more common. Many look so convincing it can be hard to tell them apart from real photos.
In a new study, researchers tested people’s ability to distinguish images of real faces from AI-generated ones and found that most participants missed most of the AI-generated faces. Even “super-recognizers” — an elite group with exceptionally strong facial-processing abilities — were able to correctly identify fake faces as fake only 41% of the time. Typical recognizers correctly identified only about 30% of the AI-generated faces.
However, the study also showed that people’s detection of fake faces improved when they were given just five minutes of training beforehand. The training taught the participants how to spot common computer-rendering errors, such as unnatural-looking skin textures or oddities in how hair lies across the face. After training, detection accuracy increased substantially, with super-recognizers spotting 64% of fake faces and typical recognizers identifying 51%.
Given how difficult the task proved to be even for highly skilled participants, how confident are you in your own ability to spot AI faces? Answer our poll below, and let us know why in the comments.
—‘I trust AI the way a sailor trusts the sea. It can carry you far, or it can drown you’: Poll results reveal majority do not trust AI
—‘There’s no shoving that genie back in the bottle’: Readers believe it’s too late to stop the progression of AI
—‘I honestly am not sure on this at all’: Poll reveals public uncertainty over experimenting on conscious lab-grown ‘minibrains’
