Researchers have sounded the alarm over the use of fake AI-generated images

The authors of an editorial published in the American Journal of Hematology, claim that “generative artificial intelligence can be used to create fake scientific images either from scratch or by altering existing visual materials to enhance the realism of the final fabricated product.”

Authors Enrico M. Bucci, professor of biology at the Sbarro Cancer Research Institute at Temple University’s College of Science and Technology, and Angelo Parini of the University of Toulouse, highlight a growing problem in science: the use of artificial intelligence (AI) to create fake images that look like real research data. The paper, titled “The Synthetic Image Crisis in Science,” describes how artificial intelligence is being used to create realistic but completely fake scientific images.

These images can evade detection because they are not edited versions of real photographs. Instead, they are created from scratch using artificial intelligence tools that can evade detection because they contain no features to distinguish them from the real thing.

“These tools can now be used by anyone, regardless of their scientific background. With the right prompting, they can model the entire visual arsenal of a study in a matter of minutes,” the authors say.

Modern artificial intelligence systems can create fake images based on simple text descriptions. For example, a user can request an image of a Western blot analysis showing a particular protein in treated cells, and the AI will generate a plausible image even if the experiment was not performed. The same tools can be used to make subtle changes to real scientific images. These changes could include adjusting colors, moving parts of the image, or adding elements, all without the traces that conventional editing tools leave behind.

The tools used in image generators are trained on real scientific images and are now widely available to the public. As a result, reviewers and journal editors are beginning to find the generated images in submitted scientific papers.

“It is critical that the scientific community and the peer-review system prepare for this looming threat of falsified scientific data,” says Antonio Giordano, MD, PhD, professor at Temple University and founder and director of the Sbarro Health Research Organization (SHRO).

The concerns raised by researchers call for updating protocols such as documentation, transparency, and accountability in response to the new reality of a world with artificial intelligence.