Deep fakes are hardly recognized as such anymore. This is the result of a recent study that presented subjects with AI-generated and real portraits. […]
Images of faces generated with the help of artificial intelligence – so–called deep fakes – are getting better and better. Yes, they are now so deceptively real that many people cannot recognize them as such.
On the contrary, many viewers consider the AI portraits to be more “real” and “trustworthy” than unadulterated photos of people. This is the conclusion of a study published in the journal “Proceedings of the National Academy of Sciences USA” (Pnas).
Low hit rate
Specifically, in one experiment, several people were to assess whether pictures of different faces were real or fake. In the first test, the hit rate was only 48 percent, so only so many recognized whether an image was real or generated using AI. Only after several attempts did the hit rate settle at 59 percent.
Images of fair-skinned people were the most difficult to expose as deep fakes, which the researchers explain by the fact that the algorithms were presented with more images of fair-skinned people than of dark-skinned people, so the AI has more “experience” in these cases and therefore works better.
In another experiment, the participants were also asked to rate which of the submitted face portraits looked the most trustworthy. Again, the fake pictures won over the real ones. This even though there were more and more intense smiles on the real pictures.