Artificial intelligence (AI) is rapidly becoming a tool in scientific peer review, but its increasing involvement has sparked a mix of curiosity, concern, and debate among researchers and publishers alike. In a February incident, ecologist Timothée Poisot found that a peer review of his manuscript seemed to be generated or heavily modified by AI, raising alarms about the potential for AI to undermine the peer review process. In a blog post, Poisot stressed the importance of authentic, human-driven feedback and argued that if AI were allowed to take over peer review, it would undermine the trust and social contract inherent in academic publishing.
While AI applications are already being used to streamline certain aspects of the peer review process—such as flagging errors in text, checking references, or even guiding reviewers toward more constructive feedback—the growing concern is that AI could eventually replace human reviewers entirely. For many scientists, like Poisot, this prospect is worrisome, as they fear that AI might lead to shallow or inadequate reviews, potentially reducing the quality and integrity of scientific publication.
AI tools like large language models (LLMs) have begun to gain traction in peer review. Publishers and researchers are exploring AI’s potential for automating certain tasks, such as summarizing findings, checking for statistical errors, and improving the clarity of written reviews. Some even offer services where AI generates entire peer reviews with a single click. However, the widespread use of AI in this context has raised several key questions. Could AI tools, which lack the critical judgment and expertise of human reviewers, become too powerful in the peer review process?
A survey of nearly 5,000 researchers conducted by publisher Wiley revealed that about 19% of respondents had already used LLMs to enhance their review processes. Yet, concerns persist about AI’s reliability, especially when it comes to generating substantial feedback. Many critics, including evolutionary biologist Carl Bergstrom from the University of Washington, caution against over-reliance on AI. While AI might improve prose style, Bergstrom argues that it cannot replace the critical thinking required for a thorough peer review. “Writing is thinking,” he emphasizes, and using AI for superficial tasks could result in a shallow analysis of scientific work.
Some researchers have explored using offline LLMs to avoid privacy concerns and speed up the review process. These AI systems, when used in conjunction with human oversight, can help sharpen reviewers’ notes or even point out areas of improvement. However, critics argue that AI systems are still prone to errors and that many outputs can miss critical aspects of the paper that only a human reviewer could catch.
Despite these concerns, there is a growing push from some in the scientific community for AI to play a greater role in peer review. A study published in Nature compared AI-generated reviews with those written by human experts. Interestingly, about 40% of researchers said AI-generated reviews were either more helpful or as helpful as those written by humans. These results indicate that AI may have the potential to assist in some aspects of peer review, particularly in tasks like reference checking or flagging methodological issues. Some developers have created AI tools that evaluate reviewer feedback, suggest improvements, and even recommend relevant references. However, proponents of AI in peer review stress that these tools are meant to complement human reviewers, not replace them entirely.
At the forefront of this AI-driven peer review revolution are companies like Grounded AI, which has developed a tool called Veracity to validate citations and check the accuracy of claims made in manuscripts. There are also AI tools like Paper-Wizard, which can perform a detailed review of manuscripts, spotting issues with statistical rigor and methodology. These tools are designed as ‘pre-peer-review’ products to assist authors before submission. However, their potential to alter the peer review process remains controversial.
The ethical implications of AI-driven peer review are still being debated. Some publishers, such as Elsevier, ban the use of AI in peer reviews, while others, like Wiley and Springer Nature, allow limited use with proper disclosure. As AI tools continue to evolve, more publishers are exploring how to integrate them responsibly into the peer review workflow, but many are hesitant to fully embrace the technology.
For some, the rise of AI in peer review could reshape the scientific publishing landscape. Jason Priem, co-founder of OurResearch, predicts that within the next few years, AI-driven reviews could surpass human reviews in speed and accuracy, fundamentally changing the process. While some researchers fear this could lead to a loss of human insight, others see the potential for AI to offer a more efficient and unbiased approach to evaluating research.
Ultimately, as AI tools continue to evolve, the scientific community will need to find a balance between leveraging AI to improve efficiency and ensuring that peer review remains a rigorous, human-centered process. Many experts argue that transparency is key—if AI is to play a role in peer review, the details of its use must be openly disclosed to maintain the integrity of the scientific process.
