There are more and more fraudulent scientific publications produced by software. They sometimes succeed in fooling scientific journals. Has the Covid-19 served as an accelerator for the publication of false scientific studies? On the coronavirus alone, more than 600,000 studies were produced in 2020. Some have been published in scientific journals and peer-reviewed, others more simply on Internet platforms, such as BioRxiv or MedRxiv [sites en anglais]. But the problem is that often, under the pressure of the emergency, researchers put their work directly online. “It’s like having access to a research draft”, explains Guillaume Cabanac, lecturer in computer science at the University of Toulouse.
Of course, these studies, which are sometimes fragmentary, have made it possible to respond quickly to
many questions: how does the virus attack our immune system? How does it spread in closed places? But we prematurely publicized research that was not completed, and some of the results of which turned out to be false. Where the process of proofreading and peer review takes months, it has sometimes been completed in days. Result: according to the Retraction Watch website [site en anglais], 40,000 studies have been retracted, including more than 190 on Covid-19.
But there is something more serious: according to the report by Pierre Corvol published in 2016,
fraudulent studies now represent nearly 2% of published studies
every year. It may not seem like much, but out of 600,000 studies devoted to
Covid-19, there are in fact 12,000 potentially fraudulent ones.
It is all the easier to produce them today as there are new
digital tools to do it. The French computer scientist, Cyril Labbé, master of
computer science conferences at Grenoble Alpes University, became known
by creating a fictional character: Ike Antkare. This avatar became in 2010 more
quoted as Albert Einstein. “I used SciGen, an automatic generator of
scientific texts, created by MIT students in 2005 to test the quality
scientific conference, explains the professor. The result is a very beautiful
text in good English but that doesn’t make sense. “
To help his colleagues spot the fake articles, the computer scientist then created a software called SciDetect. Yet again this year, with his colleague Guillaume Cabanac, he identified several hundred studies generated by algorithms in scientific journals. This is particularly the case of a pseudo study devoted to “monitoring the marine ecosystem by wifi sensors and traditional music” [site en anglais]. As unlikely as it is, it was published in the journal Arabian Journal of Geosciences, which is owned by science publishing giant Springer. Several months after being reported, it was finally retracted in early November. And the phenomenon is likely to amplify, because the systems of automatic text generators continue to progress with the advent of the family of GPT (pre-trained generative transformer), a language model developed by the company OpenIA of Elon Musk.
False studies are also proposed on an industrial scale. Elisabeth Bik, a Dutch microbiologist, along with other researchers, discovered
the existence of a business of producing false scientific articles: the
“paper mills” (translation: “paper mills”). Companies hidden in
the anonymity of the Internet jungle offers false scientific papers for remuneration. “Some articles are objectively false, she says. One of them, devoted to prostate cancer, says that half of the patients are women. “
Another actor has come to muddy the waters: the “predatory magazines”. They
come to compete with great historical publishers like Wiley, Springer Nature, Taylor & Francis, or Elsevier who owns The Lancet. Today, they weigh heavily on the market, alongside learned societies which also have their publishing house and scientific journals such as IOP Publishing. [site en anglais]. Universities have to pay them very expensive subscriptions to these prestigious scientific journals so that their students and researchers have access to them.
Conversely, “predatory” publishers propose to charge less
dear researchers to publish their article. And they promise them strong
fallout, with serious peer review. But this control is not
carried out. In fact, in 24 hours, an article can be accepted, even if it has not
any sense. Sometimes the members of these so-called editorial boards do not even exist, or are long dead. To fool researchers, some of these journals also play on ambiguity: they choose a title close to that of a reputable journal. Identifying these predatory actors is not easy.
Many resources on # Predatory Journals are available on the new website published by @_Journalology
✅ An example: advice on what to do if you think you have submitted an article to a # PredatoryReview https://t.co/ZbsFlEdeft
– Open Science Pasteur (@OpenSci_Pasteur) October 1, 2021
To alert his colleagues, the librarian at the University of Denver in the
Colorado, Jeffrey Beall, did attempt to compile a list of such journals, which is
become an international benchmark. But he found himself threatened with
lawsuits. So he had to withdraw it. Today, only Cabell’s [site en anglais] has established a ranking of more than 14,000 questionable titles, according to Hervé Maisonneuve of the blog Rédactions Médicales. But it charges so much for access to its service that no French university is subscribed to it.
Even in journals for which there is a reading committee, studies
fraudulent acts can fool peers. This is what happened to the medical journal
The Lancet in June 2020. A study devoted to the health risks of hydroxychloroquine based on data collected by the company Surgisphere, had to be retracted a few weeks after its publication, given the existence of suspicions on the origin of the data. “After the Surgisphere affair we had to change our editorial policies, acknowledges Naomi Lee, physician, editor-in-chief at Lancet, We are now asking that it be very clearly specified which author of the study had access to the raw data, and that he is indeed the guarantor of this data. “
Chérifa Boukacem, university professor in information and communication sciences at the University of Lyon 1 adds: “Today there is a crisis of proofreaders. There are many more authors, in China, in Asia, in Africa. But the number of experts has not increased as much.”
When there is a suspicion of fraud, journals are often contacted by whistleblowers. But they don’t investigate themselves. They
ask the author or his laboratory to do this. Some countries
like the United States, Denmark or Sweden have organizations
independent scientific integrity inquiry. In France, it is a “referent
scientific integrity “within each research organization or
university that is to conduct the survey.
Rémy Mosseri, CNRS referent admits having had 120 reports, 66 referrals, and had to write about twenty reports for lack of scientific integrity since the start of his mission in 2018. “Almost half of the cases concerns biology. Over a third of plagiarism issues, 20% of cases are the result of manipulations of scientific data, and there are a number of important of disagreements on who should sign the studies “, explains the physicist. A joint administrative commission then decides on possible sanctions. According to the management of the CNRS, in three years, four sanctions have been imposed, and one person has not obtained his tenure, following plagiarism, fabrications or falsification of data.
If the number of false studies published continues to increase, it is because
for some, the system has become a “push to crime”. Publish your work
certainly allows you to publicize your research to build a foundation of
common knowledge, but it is also a means of obtaining funding
for his research in an increasingly competitive world. The weight taken by
the Shanghai ranking, which since 2003 has published the list of universities and
most prestigious research institutes, according to the number of citations
the publications of their researchers, has accentuated this pressure.
But the publication of these bogus studies can be damaging to
scientists, and for the citizens themselves. We saw it with the case of the
Andrew Hill meta-analysis on ivermectin. This pharmacologist from
the University of Liverpool had studied the methods and results
about twenty clinical trials on more than 3,000 patients treated with this
pest control against Covid-19. Its analysis last January estimated that this
treatment reduced mortality by 56%, and had to be prescribed outside of
clinical tests. Some people then took ivermectin thinking
avoid disease. By looking more closely at the data from certain clinical trials, in particular in Egypt, Iran, and more recently in Lebanon, researchers have however identified fraud. “We do not see any effect of ivermectin on survival, nor on clinical recovery”, had to admit Andrew Hill to the British press.
For her part, Patricia Murray, professor at the University of Liverpool is worried
the non-retraction of a study written in particular by an Italian surgeon,
Paolo Macchiarini, and published in The Lancet in 2008. It is now in
cause for fraud in the context of several other studies devoted to a
technique of operation of the trachea. Several of his patients died. “Some doctors in other countries may tell themselves, based on this article, that this is a technique that works and that they can try it on their patients”, deplores the whistleblower. The risk in the long term is that the very image of science is damaged, that the public loses confidence in its methods, and turns away from the rational to take refuge in belief.