Science Journals to Investigate Image Fraud with AI Tools

The journal Science announced on Jan 4, 2024 that they will begin using the tool Proofig to identify instances of image fraud in publications submitted to its six journals. The policy is summarized as follows:

Proofig will be applied after a research paper is revised by authors. After analyzing the images, the tool generates a report flagging duplications and other abnormalities, such as rotation, scale distortion, and splicing. The paper’s editor reviews the findings and determines whether the AI-detected issues may be problematic. (In some cases, figures may have intentional rotations or duplications that are explained in the paper.) If so, the editor contacts the authors to request an explanation. During the pilot period, authors generally provided a satisfactory response and fixed the problems so that the paper could proceed to further review or acceptance. However, papers that should not be published were detected. Going forward, if the authors’ response is unsatisfactory or raises additional concerns, we will probe further and take steps that could include rejecting the paper. If image integrity concerns are raised about a paper that a Science journal already published, we will use Proofig to carefully examine the suspicious images, which will inform subsequent actions (e.g., correction or retraction).

Of particular interest are photographs involving microscopy, flow cytometry, and western blots which can be easily manipulated in Photoshop. Because of the relatively high rate of false positives in AI identification, authors will be given a chance to justify the images if the images are flagged for review. It’s highly recommended that researchers establish policies for their laboratories (especially new students) for proper image manipulation and documentation. Science’s policy will likely expand to other journals in the near future.

Science Announcement
ArsTechnica Article