Science has been built on trust for centuries. Once upon a time you could know most of your contemporaries in a field and whether their work could be relied upon or not. As the scientific world grew this became impossible – today there are simply far too many researchers to keep track of. However, the idea has remained that scientific treatises could, on the whole, be trusted, despite occasional embarrassments, such as the Schön scandal that featured faked data on single molecule semiconductors. It now seems we’re entering a new era where it’s becoming increasingly difficult to take anything on trust.
One recent study has examined the scale of the problem of paper mills, organisations that will produce a research paper for authors, for a price, that is either based on low-quality research or entirely made-up data. Some researchers seek out these shady businesses to pad CVs and secure promotions. One of the team’s startling conclusions is that the number of papers that are suspected to have come out of paper mills is doubling every 18 months. Another depressing conclusion they reached is that some editors at respected journals are complicit in this pollution of the scientific literature, acting as brokers who then shepherd paper-milled publications into their journal.
Many of these paper-milled products are spotted by diligent researchers examining the images that accompany these articles – frequently post publication. Often the articles reuse images and figures from other papers or the figures are doctored or manufactured to support their conclusions. But now there’s growing concern about the use of AI to produce images. Some instances of this fakery have been met with hilarity as the images were so bad they would fool no one who gave them more than a cursory glance. Unsurprisingly, however, the technology has improved. A couple of years ago there were warnings that AI could produce convincing fake images of cancers and western blots that could fool experts. Now, a new paper is saying the same thing about nanomaterial microscopy images.
So, what’s the answer to this rising tide of fakery? If the solutions were simple, they would already be in place. Post-publication peer review will continue to be one of the main routes for discovering fraudulent papers. Journals are already putting greater emphasis on supplying raw data, which should be more difficult to fake. Other approaches include delisting journals found to have lax standards and taking a poacher-turned-gamekeeper approach to AI and using it to find faked images. But one of the hardest things to address will be the ‘publish or perish’ culture that can lead to researchers seeking short cuts to publications. Sadly, with no easy answers in sight, trust in the literature will continue to be eroded.
No comments yet