False Positives and Red Flags: Navigating AI in Peer Review
By Christa Walker
Since the launch of ChatGPT in 2022, there has been concern among educators, authors, editors, and reviewers about the use of Large Language Models (LLMs) in writing research papers. An influx of “AI Detectors” appeared online seemingly overnight, claiming to identify content not written by an actual human. These new tools sparked a wave of emails from reviewers and authors asserting that the paper or review they were looking at was written by AI, based on these detectors’ results. However, these tools come with their own set of problems.
So, how can we ethically navigate this brave new world of AI use?
Avoid relying on AI detectors to determine if a paper was written by an LLM. As of this post’s publication, AI detectors lack reliability. OpenAI, the creator of ChatGPT, discontinued their own AI detector after only a few months due to its high false positive rate, erroneously flagging works like Shakespeare and the US Constitution. LLMs are trained to mimic human writing, making it challenging to distinguish their output. False positives can also occur from normal parts of the writing process, like the use of Grammarly or other grammar checking programs and tools.
Shift the focus to content quality rather than speculating about AI use! For authors, concentrate on what your readers need to understand from your research. During revision, pay attention to what the reviewer is asking you to improve. Do they point to something specific? You don’t have to agree with them, but you shouldn’t assume reviewers are using AI simply because you disagree with them or dislike their feedback.
For reviewers, evaluate the appropriateness of the methods used in the study. Were they applied correctly? Do the conclusions make sense for the study? Were the participants treated ethically? Remember, your goal is to enhance the quality of research, not to get wrapped up in AI detection. More information on how to write a good review can be found in Sage’s Reviewer Gateway here.
Don’t put papers into ChatGPT, other LLMS, or any other kind of trained AI tool to check, summarize, or review a paper. You don’t have permission to put someone else’s writing, be it a paper or a review, into programs like these. Using AI in Editorial work presents confidentiality and copyright issues and risks poor or unethical research being disseminated inappropriately.
Instead of relying on AI, set aside time to evaluate the work. If you don’t have the time to review thoroughly, consider declining the invitation or requesting an extension. Ask the authors for more supporting information if something seems incorrect or flag it in your review for the Editor. Be sure to explain your claim. Apply the CRAAP test to the sources if something seems off. Finally, review Sage’s Artificial Intelligence policies here and here for more information on what is and isn’t acceptable use. Please visit these pages regularly as the field of AI is constantly changing.
Do not overanalyze the language used in the paper to try and determine if an LLM wrote it. Remember, this is not your primary focus. Details such as “tortured phrases” or generic sounding sections are not always a good indicator of AI use. This could simply reflect that the writer is not a native English speaker or that writing is not their strongest skill. It’s also crucial to acknowledge that AI checkers have been found to be biased against non-native English speakers. Relying on these tools can disadvantage authors who are writing in a second language.
Instead, be mindful of your own biases when reading others’ work. Consider whether the authors are from a different part of the world where English is spoken differently, or perhaps not at all. Can you understand what the author is trying to say? How is the language used impacting your understanding of the paper? If clarity is an issue respectfully note this to the author so it can be improved. Be sure to point to a few specific examples in the text.
Sage has also provided an Inclusive Language Guide that can be found on our Author Gateway here. This is a helpful resource for evaluating your own biases as well as writing your reviews and papers.
Do not jump to conclusions! Accusing someone of using an LLM is an accusation of academic misconduct and could damage a researcher’s career.
If you strongly believe that AI was used inappropriately, then either email the editor directly, put it in your review, or both. In your message, be sure to cite as many specific examples as possible. Are there a large number of references that don’t actually exist or don’t match any inline citations? Are there red flag phrases in the text, such as “I cannot generate that” or “Sure I can create that for you”? Support your conclusions in your statement of concern to help ensure the issue is addressed appropriately.
AI has caused a global shift in countless areas over the last few years, including academic publishing. Its technological advancements have undoubtedly reshaped how we conduct and disseminate research. But at the end of the day, as we navigate this growing landscape the best approach is the same as it has always been: Think critically and be respectful.
About the Author