International study led by Tel Aviv University finds that the repetition effect causes images—authentic and AI-created—to be perceived as true; the effect is noticeable even among skeptics
Will we soon be unable to distinguish between a real and a fake image? A new international study led by a team of researchers from Tel Aviv University has shown that simply viewing an image repeatedly, whether it is authentic or created using AI, increases the likelihood that we will think it is real. The team of researchers warns that the results raise concerns about the spread of false visual information on social media and shaping the public's perception of reality.
Repetition effect
In the study, researchers found that people tend to believe that images they have seen before are more real than images they are seeing for the first time, even when they are created entirely by artificial intelligence. In other words, an image that has been shared multiple times on social media will be perceived as more believable, regardless of its authenticity.
The research was led by Guy Greenfeld, who is currently completing his doctorate in psychological sciences under the supervision of Prof. Nira Lieberman. At the School of Psychological Sciences at the Gershon Gordon Faculty of Social SciencesThe study also involved researchers from Germany, Belgium, and Spain. The findings were published in the prestigious scientific journal Journal of Experimental Psychology: Learning, Memory, and Cognition of the American Psychological Association (APA).
"The study is based on a well-known psychological effect called the 'repetition effect,' according to which information that is repeatedly exposed to is perceived as more reliable," explains Guy Greenfeld. As part of the study, the researchers sought to examine whether this phenomenon is also relevant to the visual world, especially to images created with the help of artificial intelligence algorithms. "This is the first study to demonstrate this effect for images (so far it has only been demonstrated for text), and the results raise concerns about the spread of false visual information on social networks and shaping the public's perception of reality. The finding can also be summarized by saying that if up until now the proverb said that a lie told enough times becomes the truth, then in the study we discovered that an image viewed repeatedly becomes reality," adds Greenfeld.
We've seen it before, it must be real.
During the experiment, subjects were shown a series of images that included both real photos taken in real life and images created by artificial intelligence. Some of the images were shown again at a later stage of the study, and the subjects were asked to decide whether the images represented a real event or object. The result was clear: compared to the images viewed for the first time, images that they saw again were rated as more trustworthy, regardless of their source. Another interesting finding is that the repetition effect was enhanced among participants who tended to be more skeptical. That is, people who are naturally cautious and do not easily believe may rely more on repetition as evidence of truth.
"In the age of social networks and digital media, we are constantly and indiscriminately exposed to visual information. While it was easy to lie with words in the past, today artificial intelligence tools make it just as easy to 'lie' with images. The new study reveals a worrying mechanism: our brains attribute greater credibility to repeated visual information, regardless of its veracity. This creates a dangerous combination - repeated exposure to false information can create credible misrepresentations, simply through repetition. The findings raise profound questions about how we process information, especially under conditions of visual overload in social and news media. They also highlight the central challenge of our time: maintaining truthfulness and criticality in a world of dynamic, easy-to-edit, and complex-to-discern visual content," Greenfeld concludes.
More of the topic in Hayadan:
