Untrained participants can tell the difference between a real image and a fake one less than half the time, and after training, three out of five times. This was the result of a study by Sophie Natingale and Hani Farid, researchers at the University of California, Berkeley. In addition, the people in the fake images appear trustworthy to viewers 8% more often than the real ones.
For the study, the authors selected 400 real photos and 400 images created by a neural network. Each set included equal proportions of women and men, of different races and ages.
Participants in the experiment were shown 128 randomly selected photos and asked to determine whether it was real or fake. On average, the subjects answered correctly in 48.2% of cases. At the same time the number of errors in classifying photographs of white men was higher than when analyzing photos of women or representatives of other races. According to the authors, the reason is that more images of white males are available for AI machine learning, which means the quality of the faking increases.
The second group of study participants was pre-trained to recognize computer-generated faces. They were shown examples of image processing traces that might indicate faking. In addition, after each response, participants were told whether they had correctly classified the image. In this case, the quality of recognition of fake photos increased, but only to 59%.
The researchers also asked a separate group of 223 participants to rate on a scale of 1 to 7 how trustworthy they thought the people in the images were. The average level of trust for real people was 4.48, and 4.82 for fake people. The 8% difference, the researchers estimated, exceeds the margin of error and is significant. Interestingly, three of the four most trustworthy people were fakes, and all four of the least trustworthy were real people.
“We find that artificially created faces are more trustworthy than real faces,” the study authors write. – This is likely due to the fact that the synthesized faces look more like average faces, which have previously been shown to be considered more reliable.”
The synthesis of sound, image and video using AI has made Hollywood-level special effects widely available. Synthesized speech, fictional images and substituting one person for another are no longer just entertainment, but also opportunities for fraud and deception. “Easy access to such high-quality fake images will lead to various problems, including, for example, more convincing fake social media profiles, fraud and misinformation campaigns with serious consequences for individuals and society as a whole,” the scientists point out.
Read more:
Scientists found the healthiest people on Earth: it’s all about a unique lifestyle
The dangers of free VPNs. Why they should not be downloaded and how to protect yourself?
Physicists recreated T-1000 abilities from “Terminator 2” in laboratory
