AI Is Secretly Writing Thousands of Research Papers
Just like a novice writer, AI chatbots have their own verbal tics—a habit of overusing certain words. Now, researchers are using this very tendency as a digital fingerprint to uncover the hidden use of AI in academic work.
The Telltale Signs of AI Writing
According to a report by the New York Times, this method has led scientists to a startling conclusion: hundreds of thousands of academic papers may have already been written with significant AI assistance. A new study published in the journal Science Advances details this phenomenon. Researchers from Germany's University of Tübingen pinpointed 454 words that large language models (LLMs) favor, including terms like "garnered," "encompassing," and "burgeoning." By scanning for these words, they determined that between 13.5 and 40 percent of abstracts in biomedical articles were likely written with help from AI.
A Flood of AI in Academia
To put that into perspective, the academic database PubMed indexes about 1.5 million new papers annually. This suggests that a minimum of 200,000 of those papers could have been at least partially generated by an LLM. As the NYT points out, this number is likely a conservative estimate, as it doesn't account for authors who intentionally edit the AI-generated text to mask its origin.
From Obvious Errors to Outright Fraud
While some academics try to hide their use of AI, others have been surprisingly careless. In one glaring example shared on X, a low-quality journal published an article that included a direct apology from the chatbot itself: "I'm very sorry, but I don't have access to real time-information or patient-specific data as I am an AI language model."
Other cases are more subtle but equally concerning:
- The blog Retraction Watch discovered papers containing the phrase "regenerate response," an option specific to ChatGPT.
- The same blog highlighted a paper on millipedes that was withdrawn for using completely fabricated references, only to be republished on another database with the same fake sources.
- One journal was famously forced to retract a paper after it was found to contain nonsensical text and an AI-generated image of a rat with bizarrely oversized anatomy.
An Unprecedented Shift in Scientific Writing
The situation is further complicated by a new trend: human academics are now consciously altering their own writing to avoid words like "delve" that are overused by LLMs, fearing their work might be falsely flagged as AI-generated.
The Tübingen researchers suggest that this widespread, undeclared use of AI could represent an "unprecedented impact on scientific writing in biomedical research, surpassing the effect of major world events such as the COVID pandemic."
For study coauthor Dmitry Kobak, the practice is baffling. "I would think for something as important as writing an abstract of your paper," he told the NYT, "you would not do that."