Just last month, we discovered that 17 papers on arXiv had secret AI prompts embedded in them. This leads to damning ethical challenges to the legitimacy of the peer review process within academia. Nikkei Asia reported that these papers, affiliated with 14 academic institutions across eight countries, primarily focused on computer science topics.
Participating institutions included powerhouses like Waseda University, KAIST, Columbia University and the University of Washington. Yet the manuscripts subtly inserted covert triggers to influence reviewers’ judgments. They employed white text and even microscopic fonts to hide these impacts. This approach did a pretty good job of hiding the prompts from public facing view.
In response, one of the Waseda University professors who advocated for the inclusion of these concealed prompts made the case that they provide a vital function. He stated that they act as “a counter against ‘lazy reviewers’ who use AI” to evaluate submissions without engaging deeply with the content. This narrow and uninformed perspective has sparked a much broader conversation about artificial intelligence’s place (or lack thereof) in academic review processes. With many conferences now banning its use, throwing more gas on the fire.
The new find has led to reconsideration of established peer review practices. Critics contend that using secret prompts violates the basic morals of accepted academic practice and transparency. This should alarm anyone who cares about the integrity of the review process. If authors employ undisclosed tactics to manipulate reviewers, the integrity of the process is broken.
Those papers, as extreme as they seem, are part of a worrying trend that we’ve begun to notice in academic submissions this year. However, as new technologies develop so too does the context of academic research and publication. Sadly, this incident represents a larger trend of hostility towards innovation against the backdrop of normal scholarly practice.