6 days ago
Crossed Wires: The swindling of science and the explosion of academic retractions
Scientific fraud was rare before the internet. In 1990, only 50 papers were retracted across the academic spectrum. In 2023-24, there were 45,000 retractions.
Since childhood I have retained a quaint and somewhat naïve picture of the 'scientist', a breed I have always thought heroic. I picture a slightly socially awkward man or woman of uncommon genius and determination, who rises above the daily concerns of mere mortals to live a life of the mind, feeding their curiosity and using their powers of observation and logic to discover the nature of … anything and everything we care about.
It's a silly idea, of course — most scientists are no more or less awkward than anyone else. Many of them are not geniuses either, merely smart enough to earn a PhD and follow a career in scientific research. And now it seems they are just as likely to be flawed and fallible as the rest of us, which is a great disappointment to me. Naïveté is often stalked by the small stings of letdown.
A recent article in The New York Times about the rise of fraud in scientific (and other academic) publishing sent me down a rabbit hole, perhaps because I didn't want to believe it. After being submerged for a while, I resurfaced with a single disenchanted question — oh my word, what has happened to my heroes?
Take Larry Richardson. A mathematician who racked up 130 citations in four years on Google Scholar, relating to a dozen papers on complex mathematical subjects he had written.
Except that Larry is a cat who belonged to Northwestern University graduate student Reese Richardson's grandmother.
The papers were all 'gibberish', according to Richardson. It was an experiment he dreamt up with another graduate student to test how easy it is to fake citations. Having many citations of your work means potentially a better job or promotion, a higher salary or tenure. The papers were eventually taken down by Google Scholar, but the citations remain (full story here). This is an amusing incident, but it reflects a growing problem.
Broken model
Scientific fraud was rare until about 2010. Scientific discovery, its methodologies and results are almost entirely mediated by the global academic publishing industry, which has always wielded enormous responsibility and power when it comes to which papers are selected for publication.
There is a well-established process of filtering, editorship and peer review, and, historically, retractions were rare enough to sometimes make mass media headlines, such as the retraction in 2010, 12 years after publication, of the Andrew Wakefield paper that sought to draw a linkage between MMR vaccines and autism, the effects of which are still being felt today.
Before the internet launched in the mid-1990s, there were about 6,000 scholarly journals (across all disciplines) which printed about 600,000 articles a year (according to Web of Science). The journals were largely governed by professional bodies, and high printing costs meant small distributions and short individual submissions to restrict journal page length.
The internet broke that model. Print costs went to zero, as did distribution costs, and journal length was no longer a constraint. Add to this the increasing pressure to 'publish or perish', the astonishing rise of research in China and India, the salary arms race for top talent, and the rise of 'open access' (free) academic journals, and you have a perfect storm for deception and outright fraud in what is euphemistically called the 'academic prestige' industry.
Those tame figures from the 1990s have exploded into an estimated 24,000 journals and four million articles this year. Most of the journals are of poor quality, with questionable review policies, generated by 'paper mills' — a vast global ecosystem of commercial research paper factories. An unscrupulous academic can simply order up a paper or, even more shadily, just pay for an author's slot.
Until recently, the papers were written by ghostwriters with some experience and a skilful hand at plagiarism. But things have changed since 2003. Guess who writes the papers now? AI. And they are getting better, sometimes sliding through strict peer review.
Unsurprisingly, the good guys are pushing back. For instance, in 2023-24, a publication called Retraction Watch reported 45,000 retractions across the academic spectrum (which includes the humanities, but the fraud is mainly in the sciences).
In 1990, only 50 papers were retracted. From 50 retractions to 45,000 in 35 years is alarming, to say the least (China leads the walk of shame by a factor of five). It is interesting to note that even this number may be underestimated — the prestigious publication Nature reports that there may be hundreds of thousands of suspect papers as yet undetected in the scholarly publication canon.
There are other pushbacks from the industry. Cross-journal collaborations are seeking out fraudulent papers (the two best known are STM Integrity Hub and United2Act), and there are new technologies which scan papers and alert when there is a hint of suspicion, like Papermill Alarm and the Problematic Paper Screener. Amateur sleuths and whistleblowers also play a critical role.
AI's unstoppable march
Academics who are caught are sanctioned, but I fear that this is a losing battle because of AI's unstoppable march. The technology is improving so rapidly that it is inevitable that detection technologies will lag behind the ingenuity of cheaters, just as they have in other areas of AI encroachment.
Also, legitimate scientific breakthroughs are increasingly being made by AI, and the resulting papers will also be written by AI. The AI papers may be worthy contributions, but unethical researchers will find it tempting to claim authorship if they can get away with it.
Have AI papers made it through to any reliable journals? The evidence is increasingly yes. For instance, there was a high-profile episode in 2024 when a Frontiers in Cell and Developmental Biology article published AI-generated figures that were anatomically nonsensical; the paper passed editorial checks and peer review but was later retracted after community scrutiny.
There is a further challenge. It takes time to retract a paper. There are legal issues at play; authors can sue. Retractors must be incredibly careful and thorough. It took 11 years for the Wakefield autism paper to be retracted. Timelines are much shorter now, but, until the paper is retracted, it forms part of our knowledge base and may be acted upon by the industry (for instance, in drug development).
What does this mean for the rest of us? Some malfeasance in academic publishing is certainly regrettable, but perhaps not much more than that? Some unnecessary expense is incurred, perhaps some unfair advantage is gained by crooked researchers. One could sigh and say that the world still moves on — after all, there are crooks all over the place, not just in research.
Which brings me back to my admiration for the field of research and discovery, particularly in the sciences. This grand human endeavour, the application of intellect, experience, creativity and rigour to satisfy our unquenchable curiosity — surely this is where truth and honesty matter more than anywhere else?
Or is that just my naïveté talking? DM
Maverick451 in SA and Legend Times Group in the UK/EU, available now.
