Artificial intelligence to prevent scientific fraud

Laboratory life. The Dana-Farber Institute in Boston, one of the main American cancer research centers, affiliated with Harvard University, is facing a major crisis that puts its reputation and credibility at stake: has just requested the retraction of six scientific articles, signed by some of its key members, and corrections for dozens of others. These measurements come after a British biologist, Sholto David, described on the website of German fraud hunter Leonid Schneider suspected image manipulation in 58 Dana-Farber studies.

Could this have happened if the magazines publishing the contested articles some twenty years ago had artificial intelligence (AI) tools available to detect these image manipulations in advance?

In his first editorial of 2024, Holden Thorp, editor-in-chief of the American magazine Science and associated journals, announces, in any case, having adopted this innovation: manuscripts submitted to these publications will pass through the filter of this software, called Proofig. “This should help you detect simple errors and fraudulent activity before you decide to publish”, explains Holden Thorp. A necessity, due to recent “incidents” that undermined public trust in scientists and damaged the careers of some of them, “who did not detect doctored images originating from their own laboratories”.

Already adopted by other scientific publishers, Proofig has proven its ability to detect “problematic figures”, says Holden Thorp. He remembers that, seven years ago, Science uses another software, iThenticate, to detect plagiarism. If AI is used to detect image changes, humans retain control over the analysis and final decision. Science will also use Proofig to retrospectively analyze articles whose integrity may be questioned.

Dror Kolodkin-Gal is the creator of Proofig, an Israeli start-up with more than twenty employees. This former researcher was struck, during his career, by the fact that the reputation of the best scientists could be tarnished “for stupid image duplication errors”not intentional. “It was incomprehensible to me that there were no quality control tools for jobs that sometimes require years of effort and millions of dollars of investment. » He finally decided to create it.

You still have 45% of this article to read. The rest is reserved for subscribers.

Leave a Comment