Science literature is being polluted with fake research papers, but the numbers have been greatly increased. Just focusing on cancer research to get an idea of the scale of the problem…
A January 2026 study in The BMJ delivered one of the sharpest measurements yet of the crisis’s scale. A team led by biostatistician Adrian Barnett of Queensland University of Technology trained a BERT-based text classifier — a bidirectional language model — on 2,202 known paper mill papers drawn from the Retraction Watch database, then turned it loose on 2.6 million cancer research papers published between 1999 and 2024. The model achieved 91% accuracy in distinguishing genuine from fraudulent manuscripts. What it found when applied to the broader literature was sobering: nearly 10% of all cancer papers screened — more than 261,000 studies — showed textual patterns consistent with paper mill origin, according to the machine-learning cancer paper mill study
The study drew particular attention to a paradox built into the fraud problem: flagged papers accumulated significantly more citations than non-flagged papers. Fraudulent research, in other words, was not sitting ignored at the margins of the literature — it was being cited and built upon, according to Barnett’s BMJ analysis of paper mills..
You beat me to it, Goofy. I realised AI overview generated “insights” were a dodgy bet when I asked Google a question a year or so ago (on the likelihood of domestic violence evaluators being bamboozled by AI generated narratives) and one heavily relied on source was…me. From a post I placed on this very board!
You’ll notice that the OP used the present tense. For sure, outdated research becomes outdated and/or redundant as new data, evidence, interpretation etc changes the landscape. That’s very different from literature containing fraudulent work that’s being published in the here and now.
“Publish or perish” in academia puts much pressure on participants. Volunteers review papers submitted before they are published. Use of AI to improve papers must be tempting. If the research reported is plausible it might get through. But can still be discovered when others repeat it.
Best solution might be to require some to repeat experimental to verify them. But that would severely delay publication.
This is a concern mostly in the sciences. Not likely a problem in history. Could AI fake newly discovered historical documents? The long lost letter from Robert E Lee?
It wouldn’t surprise me. I read there’s a ‘documentary’ coming out soon on Elon Musk. He refused to participate in the project so they used computers to gin up a fake interview. Which will now become part of history.
How would this help? Remember, the issue is with FAKE publications…which, by implication suggests that authors did this knowingly. I doubt they’d be any more honest with a do-over.
Back in the early aughts, it was suggested that Andrew Wakefield submit more data or review what he’d already submitted. He refused and was subsequently “invited to resign” his gig at the Royal Free. This was the research that was ultimately determined to be fraudulent. It seems it’s discovered more quickly now
Fake publications are usually discovered when someone determines the stated conclusion is false. Asking an outside lab to repeat a key experimental is one way to verify authenticity. Very practical sometimes, difficult or impossible sometimes. But would increase delays in publication.
Partnerships in this sort of thing are very possible. I will repeat and verify your experimental if you will do mine. Connections, relationships, etc can make this feasible.
Keep in mind students sometimes embellish results to impress the boss. Is boss sharp enough to spot faked results? A famous one is student using marker to fake an animal transplant.
Remember…this thread pertains specifically to FAKE research publications. It’s hard to imagine you actually think that a fraudster would join forces with another entity in order to give the findings the seal of approval pre publication. Unless, of course, it was another fraudster willing to get involved with the subterfuge. Could certainly happen…but as a “solution” to fraudulent publications is a laughable idea.
See, FRAUDSTERS do their fraudulent stuff knowingly and with the intention to get their work published somewhere, somehow…and probably without any concern over whether it ultimately is exposed. Researchers who cut these capers … and Andrew Wakefield would be a prime example…even go so far as to deceive their co-authors (at least the ones who’re acting in good faith) They do not want their fraudulent results looked at closely by unrelated researchers who aren’t in on the Big Bamboozle.
This shouldn’t be a difficult concept to take on board when discussing FRAUDULENT publications.
I think, however, the concept is one of not having the work published until someone else replicates it. Impracticable, but the fraudster would have to have someone in a different institution in on it, as well as being able to tell the journal editor to whom to refer it.
Kinda. Although…it’s not like the AI Musk is saying anything the real Musk hasn’t said. Apart from Musk refusing to participate, the use of AI was intended to be ironic.
No need to do that. For the fraudsters who’re only interested in getting their names recognised…if only temporarily before retraction/embarrasment…they rarely go for the peer reviewed journals. It’s the pay-to-publish, predatory rags that serve their needs. No peer review, and oftentimes no editorial oversight either.