AI in Peer Review: The Positive, The Negative and Insights from the Research Integrity Team
By Arpita Bhattacharya
The publishing industry has always embraced technological developments. Over the years, many innovations have reshaped the industry, and advancements in the field of artificial intelligence (AI) are emerging as another transformative force, with the potential to significantly impact scholarly research and publication.
While AI is not a new concept, Generative AI (GenAI) and Large Language Models (LLMs) have recently gained more attention due to innovations like ChatGPT. GenAI encompasses deep learning models that can produce high-quality images, texts, and other forms of data based on the data used to train the models[1]. LLMs are AI models designed to understand and generate text-based outputs and are trained on vast amounts of data to understand natural language prompts. They specifically generate text-based outputs[2].
Like many other fields, AI is having an impact on the academic publishing industry, including peer review. From generating research topics ideas, helping to write research articles, and even creating review reports, AI is transforming the scholarly landscape. This prompts us to consider: How can we prepare ourselves to embrace this technology, and what steps can we take to ensure it enhances our work rather than complicating it?
AI and Its Benefits for Peer Review: The Positive
AI technologies are touted to revolutionize scientific writing for all. Tools like Writefull or Paperpal offer significant benefits to all authors especially to those whose first language isn’t English. Other tools such as Jasper, Trinka and ProWriting Aid offer AI powered scientific writing and paraphrasing support to academics. We know that academics are using AI-powered technologies to supercharge their research, and Sage recognizes that it is vital to support the needs of researchers. Therefore, our AI policy requires authors to declare the use of such tools only when they’ve used them to generate new findings or content.
AI can also simplify drafting formal communications within the peer review process. Authors can effortlessly generate eye-catching cover letters, while editors can write decision letters with ease. To ensure confidentiality, we ask our editors not to input any reviewer comments into GenAI applications for decision letters, but they may use them to develop structured templates. AI is transforming the way we communicate within the peer review process, making it more efficient for everyone involved.
Reviewers might find it helpful to use AI to create standardized response letter formats to facilitate efficient peer review. Future AI technologies could also aid reviewers with fact checking references, or double checking statistical analyses - complementing rather than replacing the human element in peer review. Time will tell if reviewers may use AI tools to spot errors in a submission or double check the soundness of analyses. We see a future where AI can support reviewers in making sound judgments about a submission. AI-generated summaries of existing articles could provide easy and quick references, serving as learning aids or as a means to provide quicker responses to questions posed by researchers. While Plain Language Summaries of scholarly literature aren’t directly relevant to peer review, they might be useful for journals that involve patients as peer reviewers.
Another important area where AI can assist is finding reviewers. Sourcing expert reviewers can be a time-consuming task for editors. Rather than replacing reviewers with AI generated reviews, AI tools could assist in finding reviewers using keywords, searching through previous publications, institutional website information, and more. This approach could help improve reviewer databases and significantly streamline the review process.
Emerging Challenges in Peer Review Due to AI: The Negative
Pressure on peer reviewers to provide timely and thorough assessments means that the temptation to replace human reviewers with AI technology is growing. There have been industry wide discussions on whether AI generated reviews can become the future of peer review, either augmenting or replacing humans in the process. Is AI capable of critical thinking, and can it provide constructive feedback to authors? In other words, can it truly replace actual human expertise? The basis of peer review is to seek input from experts in the same field (peers) who may have insights to share from their experience and extensive knowledge for the benefit of the wider community. AI models or tools may be able to offer simple feedback such as “provide further information here” or “what is the reference to support this claim” but can it really replace the human expert who may say “this analysis is not appropriate for this topic or research question”?
As research integrity professionals, we are observing an increase in reviewers using generative AI to create review reports. Our primary concern in some scenarios is the difficulty in determining whether the reviewer wrote the review and used generative AI to improve the language or if they input the submission into generative AI and asked it to review the article. In some circumstances, we have noticed similar review reports repeated across submissions, either by the same reviewer or different individuals. This presents one of our biggest ethical challenges, as AI cannot be held accountable for thoroughly reading submissions or adequately fact checking research. There is potential for the use of AI to undermine the peer review process, impacting the ability of editors and publishers as gatekeepers of the scholarly record to ensure only reliable research is published.
The rise of generative AI used for scholarly writing has highlighted other serious issues of hallucinations[3]. We have seen reports of incorrect or non-existent citations in research articles. We are also concerned about the generation of fabricated scientific images that may be undetectable by the human eye. The appearance of incorrect or fabricated data provided by generative AI puts additional pressure on manual checks or the development of technology to detect such activity. Thus, rather than offering support, the use of generative AI has added another task for reviewers and editors; to carefully check the submissions for inconsistencies like non-existent or irrelevant references.
What does the RI Team think?
In the present world, everyone is aware of the existence of AI and knows that it is being used in scientific research and scholarly writing. The need of the hour is not to fight the change, but to develop processes so we can co-exist and maintain the integrity of the peer review process. While authors should ensure that the use of synthetic data or use of generative AI for writing and editing scientific papers is clearly disclosed[4], reviewers and editors should understand their role as field experts to spot inconsistencies that may occur when AI is used to generate content. We are often contacted by editors and reviewers who are staying vigilant and can spot AI-generated content in their submissions queue or AI generated review reports that reach their desk. Some of our eagle-eyed internal editors have also spotted AI-generated submissions that plagiarize previously published work. In our opinion, this is something that subject experts can assess better than a tool.
Similarly, policymakers must stay abreast of developments in generative AI technology, to ensure that guidelines and policies remain effective, proportionate and uphold the highest ethical standards in scholarly publishing.
Works Cited:
[1] IBM (2023): “What is generative AI?”
[2] Canary (2023): “What are Large Language Models? Explained in Plain Terms.”
[3] Sage Publishing (2024): “Using AI in peer review and publishing”
[4] Sage Publishing (2024): “Artificial Intelligence Policy.”
About the Authors
