Bad Bunny

This artificial intelligence (AI) stuff clearly cuts both ways. As Benito Antonio Martínez Ocasio has said about his performance name, “Bad Bunny,” it is something that looks good (A bunny, what’s not to like?) but may indeed be bad (Think “Snowball” in The Secret Life of Pets). That seems to me to be the storyline for AI. It can be a great tool, and it can get you into a lot of trouble. One of our jobs as scientists, educators, writers, and just plain old human beings and world citizens is to know the difference between good and bad. In this case, although it is true that AI can help you with your writing and your research, its use can get you into practical and professional trouble if you use it unwisely. Here is a story for you. I just rejected a paper that three different software programs reported as 74%–97% AI generated. I was shocked. Although many of my editor colleagues have been reporting to our various editors’ groups on the increasing detection of AI in submitted papers, this was the first paper I have received that I could clearly and unequivocally say was AI-generated. It was most certainly the first paper that made it through my desk reject screening. Thankfully, a reviewer contacted me with concerns about the paper, giving us both a chance to consider options, the first of which was to run the paper through AI-detection software. The reviewer and I independently used different detection software; however, we received similar results of “strong evidence” that large portions of the manuscript were “not human-generated.” In particular, the paper, which was a scoping review, was found to have multiple red flags, including “statistical perfection.” There was also “perfect” alignment between the narrative and the tables, something that rarely happens. Papers always have minor errors and human-driven “idiosyncratic” writing. The paper was just too formulaic, something that the excellent reviewer identified independent of AI-detection software. Software scanning of the document confirmed the reviewer’s concerns. The detection software also noted “implausible DOI (digital object identifier) patterns.” I subsequently submitted the titles of the “reviewed” papers to various indices and discovered that not one of them was real (let that sink in). On my closer examination, I could see that the titles of the articles that were claimed to be part of the review were themselves suspicious; they were of nearly identical language, syntax, and length across multiple entries. The “articles” were reportedly from high-quality journals, but in fact, a search of those journals showed no evidence that any of the so-called reviewed papers had been published by the journal in question, or by any journal for that matter. I cannot tell you how grateful I am to the reviewer, who did exactly what editors hope reviewers will do—read carefully, note concerns clearly, and contact us when there appears to be an issue. Because of the reviewer’s quick work, we at the journal were able to cancel the second reviewer, so more time was not wasted. We also did not end up accepting this “well-written” paper, only to get to the editing phase before finding that a host of fabricated articles were included in the review. Yes, we wasted time and energy, but not as much as we might have. Now, you may be wondering why I sent this paper to a reviewer in the first place. Me, too. I did so because the topic was interesting and even a bit novel. It also came from authors new to the journal, and it came from a country from which we rarely receive papers. The writing was good, and the paper conformed to format guidelines. I thought it was worth sending out for review. To be clear, before we send papers to reviewers, we submit them to iThenticate. This paper had a low iThenticate number; it was not plagiarized. Unfortunately, it was fabricated, and the journal’s version of iThenticate does not scan for fabrication or AI generation. Later, I used different software to detect AI. Thinking back to about 2020, when AI started becoming a big thing in publishing, I realize that occasionally a paper has gotten all the way through review and been accepted, only for me to then discover there were citations I could not find, even with extensive searching. Since we will not publish a paper that has unverifiable or inaccessible references, if we find such references during editing, we ask authors to provide a correct citation. Sometimes authors can do this, but sometimes they cannot. We omit any unverifiable or inaccessible citations from the final version. I can remember this happening only a handful of times, and trust me, before we publish a paper, we check and double-check every reference, including cited websites. One recent paper concerned me more than others, as there were four citations out of 40 that I could not find. The author could not produce them, so I removed them from the paper and had the author correct the text. We did publish the paper; I am still queasy thinking about it, even though we verified all remaining citations. Only now do I wonder if that paper, and perhaps others, were “written” with AI assistance. It is a troubling idea to consider. After the current discovery of AI misuse, I had a discussion with colleagues about this issue. We all acknowledged using AI for some aspect of our writing. In particular, we use AI to correct our grammar and spelling, to shorten long sentences, or to make our writing clearer in other ways. Some colleagues noted that they would often run an entire paper through AI to address these editorial issues. However, most colleagues only used AI editing for specific, well-delineated sections of papers. Colleagues also noted that they use AI to write code for statistical analyses or for creating figures; they do not use AI to run the analyses or create the figures. Both editing and code creation can be done without AI, as can many other aspects of writing and editing. The point, however, is that AI can make these tasks easier and quicker. Moreover, we were all clear that using AI tools does not absolve us of the responsibility to be accurate; we know we have to check AI’s work. Checking behind AI is a good strategy to keep that bunny from turning bad. Where you need help—creating titles, generating code, editing papers—use AI. However, carefully check what AI gives you for solutions. AI software may lead to edits that do not retain the intent or the accuracy of your writing. AI has no ethics; it is not human, and it does not care. Thus, AI systems will fabricate data and references and “write” misinformation. Remember, too, AI leaves a mark—that bad bunny is not subtle. Rather, AI software will change your writing style, “smooth,” accurate but imperfect writing into nonexistence, fabricate references, and lie. And be warned—even though AI detection software is imperfect, it will detect all these bad bunny behaviors. I personally welcome timesaving tools, especially for spelling and grammar. There is nothing wrong with using AI for these and other mundane writing tasks. As scientists, however, we must be transparent about our use of AI (Khalifa & Albadawy, 2024). Thus, in your cover letter to the editor, you need to say you used AI and how you used it. I understand that authors are worried that acknowledging AI use will be frowned upon. However, transparency and honesty do not go out of style. Moreover, a lack of transparency will receive no mercy, at least, not from me. Not only will your AI-generated paper not be published in the journal, but I am also likely to report your fabrication to your university. In short, it will be quite costly to you if I discover you used AI without acknowledgment. I too can be a bad bunny. AI can revolutionize our work as scientists. We can use it to summarize large amounts of information and analyze complex data. With ethical use, AI may help us make discoveries within our data that would take us years to sort out without these automated systems. Moreover, with ethical use, AI could result in greater accuracy and transparency in scientific writing and publication. But we are not there yet, and unless we use AI responsibly now, while it is still in its “infancy,” we run the risk of eliminating the last little shred of trust the public has in scientists. So let us agree not to blow this chance we have to further develop and use what could be a wonderful productivity tool. Let us agree that we understand that AI will change our work in good ways only if we commit to using it with integrity and within established academic standards. Let us acknowledge that AI cannot find solutions to improve human health, our disciplinary goal, unless we carefully ensure its ethical use. We know that AI will continue to develop in sophistication and to grow in use. That is fine as long as we keep the bad part in check. Our work in human science is a privilege, and our journey as explorers of human health needs curiosity as well as ethical commitment. Please, do not be a bad bunny.