AI unreliable in identifying retracted research papers, says study

LLMs don’t reliably identify retracted papers, a new study finds. (Image: DALL-E)

Large language models should not be used to weed out retracted literature, a study of 21 chatbots concludes. Not only were the chatbots unreliable at correctly identifying retracted papers, they spit out different results when given the same prompts.

The “very simple study,” as lead author Konradin Metze called it, used LLM chatbots like ChatGPT, Copilot, Gemini, and others to see whether they would successfully identify retracted articles in a list of references.

Metze and colleagues compiled a list of 132 publications. The list comprised the 50 most cited, retracted papers by Joachim Boldt, a prolific German researcher who also sits at the top of the Retraction Watch Leaderboard. Another 50 were Boldt’s most cited non-retracted papers. The rest were works by other researchers with the last name “Boldt” and first initial of “J.” The study authors prompted each chatbot to indicate which of the listed references had been retracted.

Continue reading AI unreliable in identifying retracted research papers, says study

Research integrity conference hit with AI-generated abstracts

The first of three themes for next year’s World Conference on Research Integrity will be the risks and benefits of artificial intelligence for research integrity. In an ironic and possibly predictable turn of events, the conference has received “an unusually large proportion” of off-topic abstracts that show signs of being written by generative AI.

The call for abstracts for the conference, set for May in Vancouver, closed a month ago. Last week, peer reviewers received an email with “URGENT” in the subject line.

“If you haven’t already reviewed the 9th WCRI abstracts that have been allocated to you, please take note of the following,” the email read. “We’ve received several signals that an unusually large proportion of the abstracts are completely off-topic and might have been written by some form of generative AI.”

Continue reading Research integrity conference hit with AI-generated abstracts

Computing society pulls works for ‘citation falsification’ months after sleuth is convicted of defamation

Solal Pirelli

An international computing society has begun retracting conference papers for “citation falsification” only months after the sleuth who flagged the suspect articles was convicted for defamation in a lawsuit filed by one of the offending authors.

So far, the Association for Computing Machinery (ACM) has pulled at least 27 of the papers, but dozens more remain, according to Solal Pirelli, a software engineer in Lausanne, Switzerland, who raised concerns about the articles more than two years ago. Some of the proceedings allegedly include plagiarized works, while others are plagued by citation stuffing.

The retraction notices from September 10 state:

Continue reading Computing society pulls works for ‘citation falsification’ months after sleuth is convicted of defamation

Weekend reads: Debunking ‘When Prophecy Fails’; ‘Godfather of AI’ first to reach 1 million citations; ‘Cake causes herpes?’

Dear RW readers, can you spare $25?

The week at Retraction Watch featured:

Did you know that Retraction Watch and the Retraction Watch Database are  projects of The Center of Scientific Integrity?  Others include the Medical Evidence Project, the Hijacked Journal Checker, and the Sleuths in Residence Program. Help support this work.   

Here’s what was happening elsewhere (some of these items may be paywalled, metered access, or require free registration to read):

Continue reading Weekend reads: Debunking ‘When Prophecy Fails’; ‘Godfather of AI’ first to reach 1 million citations; ‘Cake causes herpes?’

Botanists plant a stake in oral cancer research with case report, now under investigation

Elsevier is investigating a case report of a person with aggressive cancer, written by three plant researchers working far afield of their specialty. 

The three authors of the study, published June 2024 in Oral Oncology Reports, purport to diagnose a 63-year-old man with a rare, aggressive form of oral cancer. The journal is a companion title to Elsevier’s Oral Oncology according to the homepage, but is not indexed in Clarivate’s Web of Science. 

Corresponding author Velmani Sankaravel told Retraction Watch he and his colleagues found the case report from an “online open-access source” and then used it “to support our research on plant-based diagnostics for oral cancer.” However, the paper lists CT scans, biopsies, and other routine diagnostic tests and makes no mention of plant-based diagnostic tools.

Continue reading Botanists plant a stake in oral cancer research with case report, now under investigation

AMA ethics journal shutters after 26 years

The American Medical Association will cease publication of its ethics journal at the end of this year. 

The AMA Journal of Ethics, an open access, peer-reviewed journal was founded in 1999 under the name Virtual Mentor. 

“The loss of the AMA JoE will be most acutely felt by medical students and trainees, since it had a unique production model that included them in the process,” said Matthew Wynia, a physician and bioethicist at the University of Colorado whose work has been featured in the journal and who previously led the AMA Institute for Ethics.

Continue reading AMA ethics journal shutters after 26 years

BMJ places expression of concern on heavily criticized stem cell paper

The BMJ has issued an expression of concern for a paper claiming stem cell therapy can reduce the risk of heart failure. The move comes after sleuths and scientists critiqued the “complete mismatch” between the study data and the article itself. 

As we reported last week, the October 29 paper included results of a phase III clinical trial in Shiraz, Iran. Critics quickly began pointing out discrepancies in the data on PubPeer, including psychologist Nick Brown, who pointed out a “curious repeating pattern of records in the dataset” every 101 records. 

According to the expression of concern published today, The BMJ acknowledged issues “apparent from the data that support the paper” including data irregularities, discrepancies in the age criteria and the ages of participants included in the study, and undeclared conflicts of interest. 

Continue reading BMJ places expression of concern on heavily criticized stem cell paper

Math is back as Clarivate boosts integrity markers in Highly Cited Researchers list

This year’s Highly Cited Researchers are from 61 countries and regions, but 86 percent of them work in the top 10.

The analysis behind this year’s Highly Cited Researchers list, released today by indexing giant Clarivate, includes several tweaks aimed at reducing attempts to game the metric and excluding researchers who engage in questionable publication practices.

Those changes include removing highly cited papers from the calculations authored by researchers excluded from last year’s list for integrity issues. The company also applied specific removal criteria — including excessive self-citation rates, papers retracted for integrity concerns, and prolific publication rates — more comprehensively this year. In past years, the company had done so manually for particular geographic areas or disciplines.

“We’re trying to make sure that the indicators are valid and reliable, which means we have to include these kinds of filters or screens and quantitative tests that indicate some kind of quality, qualitative character,” David Pendlebury, head of research analysis at the Institute for Scientific Information at Clarivate, told Retraction Watch.

Continue reading Math is back as Clarivate boosts integrity markers in Highly Cited Researchers list

Bug in Springer Nature metadata may be causing ‘significant, systemic’ citation inflation

Millions of researchers could be affected by a “dramatic distortion of citation counts” likely caused by flaws in how the academic publishing giant Springer Nature handles article metadata, according to a new preprint.

The bug means a large number of citations are automatically attributed to the first paper in a given journal volume, instead of to whichever paper in that volume they were intended for. The issue appears to affect many of the publisher’s online-only titles, such as Nature Communications, Scientific Reports and several BMC journals.

“It seems that millions of scientists lost a few citations, while tens of thousands, the authors of Article 1s, gained all these, leading to insane citation counts,” Tamás Kriváchy of the Barcelona Institute of Science and Technology, in Spain, told us. His findings appeared earlier this month on arXiv.org. And those citation losses and gains are through no fault (or intention) of the authors themselves. In fact, one author we spoke with has tried, without success, to get mistaken citations removed from her paper. 

Continue reading Bug in Springer Nature metadata may be causing ‘significant, systemic’ citation inflation

Author changes name, publishes 10 papers in journals that banned him

How to render a publishing ban moot? Change your surname and just keep submitting.

That’s what happened in the case of Hashem Babaei, aka Hashem Gharababaei. In 2010, the Institution of Mechanical Engineers (IMechE), a professional society based in the U.K., banned the mechanical engineering researcher from the University of Guilan from submitting his work to its journals. 

But over the next 10 years, (Ghara)Babaei managed to publish at least 10 articles in the society’s journals, simply using the abbreviated version of his name while continuing to use the same email address from his institution in Rasht, Iran. 

Continue reading Author changes name, publishes 10 papers in journals that banned him