Academia was engulfed in scandal this week after research papers across dozens of academic journals were found to have been written by AI.
An investigation found over 100 papers that were likely written, at least in part, by ChatGPT.
These papers slipped through because of lax or nonexistent peer-review processes at for-profit journals, stoking wider fears that the body of human scientific knowledge is being rapidly infiltrated by low-quality computer-generated garbage.
Many of these papers were published in obscure scientific journals, but news of this kind of scientific fraud hurts public trust in science, many scientists agree. Even before AI-generated text in scientific papers, so-called ‘paper mills’ have been pumping out low-quality and even plagiarized papers for years.
This article, published in the International Journal of New Media Studies, bears the telltale sign of being written by AI: ‘As of my last knowledge update.’
This chapter from a book on media studies covers issues important to youth in India. It contains odd phrases such as ‘Youth is the time of life when one is young.’
The influence of AI on research papers was exposed by a new report from the tech journalism site 404 Media.
It found there were 115 results for ‘As of my last knowledge update’ in Google Scholar, Google’s search tool for academic papers. This phrase, which as of Tuesday returned 188 results, is often used by the bot to tell users how current its information is.
404 Media reported that the dates in the papers containing the phrase corresponded to the dates of actual ChatGPT knowledge updates.
The papers covered topics including spinal injuries, battery technologies, rural medicine, bacterial infections, cryptocurrency, children’s wellbeing, and even artificial intelligence.
Many of the journals publishing these fraudulent papers are raking in money, charging the authors processing fees to publish their articles that do not contain any new ideas.
A few of the search results are not meant to be fooling anyone, and in fact are about the pitfalls of using ChatGPT as a tool for research.
‘Don’t Trust ChatGPT: A Case Study of a Defective Research Tool,’ is the title of one such paper.
Removing any results with ‘ChatGPT’ in them still yields a staggering 135 papers that include the odd phrase.
Some of these articles appear legitimate, while others are barely intelligible: ‘Global Education Iducation and International Education Advocacy’ is one such example of garbled titles.
The phrase even appears in a paper about how to make AI systems more transparent and explainable.
Bellingcat researcher Kolina Koltai posted a screenshot of one academic paper on X where the researchers included ChatGPT’s cheery reply in their academic paper: ‘Certainly, here is a possible introduction for your topic.’
Besides the fact that using ChatGPT is easy, professional pressure is a major reason this is happening.
ChatGPTis regularly updated with new knowledge about the world. When it gives answers to questions, it often clarifies when its knowledge was most recently updated. It is a useful note for ChatGPT users, but it also offers a clue that an academic paper was written by ChatGPT.
Bellingcat researcher Kolina Koltai posted this screenshot to X, showing a scientific paper whose authors did not even bother removing ChatGPT’s signature cheery reply.
Academic researchers face immense pressure from their universities to publish papers, because they are one of the major ways scientists are evaluated when it comes time to get a new job or receive a promotion.
This challenge is so common, that there is an ominous phrase used to describe it: ‘publish or perish.’
And for those looking to take the easy way out, plenty of academic journals are happy to oblige.
Normally the publishing process for a reputable academic journal is lengthy, involving peer review by other scientists, often followed by multiple rounds of revisions.
Scientists may even need to perform additional experiments to satisfy the requests of journal editors and reviewers.
This process can take months or even years, depending on what changes are needed.
This paper contained two different versions of the telltale phrase. The journal it is published in charges a mere $4 processing fee.
This academic article was published in December. It has been cited zero times since then, suggesting that other researchers did not find it useful
But generally speaking, the end result is a solid piece of work that has been scrutinized, polished, and perfected.
Not so with these ‘paper mills’ that accept nearly every submission – as long as the author also pays the publishing fee.
The International Journal of New Media Studies, for instance, has published two different papers containing the telltale phrase, ‘As of my last knowledge update.’
The journal claims to conduct peer review, and even has a page dedicated to describing its process.
‘The practice of peer review is to ensure that only good science is published,’ reads the introduction to the page on the journal’s website.
‘It is an objective process at the heart of good scholarly publishing and is carried out by all reputable scientific journals.’
Peer review involves sending out a manuscript to multiple experts – usually three – who carefully read the paper and offer their critiques. Even with no major revisions, this process takes time, because reviewers are working scientists who must find time in their schedules to read the paper.
This paper is titled ‘Global Education Iducation and International Education Advocacy.’ It was published in January of this year, and it contains only four references to other studies
A second use of the key phrase in this paper shows that it was almost certainly written by AI
Yet the journal promises that articles will be published one to two days after they are submitted.
DailyMail.com has reached out to the journal’s editors regarding its policies on AI-generated submissions.
The International Journal of New Media Studies charges authors $50 to publish an article, another $20 to assign the article a DOI number (used for online indexing), and yet another $50 to receive a hardcopy of the journal.
This journal is a glaring example, but it is hardly unique.
Searching for ‘As of my last knowledge update’ yields a huge list of papers published in all sorts of journals, covering topics as diverse as materials science, youth wellbeing, and quantum physics.
It is no secret that paper mills crank out low-quality papers for a fee.
But as AI chatbot tools become easier to use and more widely available, these low-value papers seem to only be getting more abundant.
