AI tools become essential support for academic peer review amid submission surge

Traditional publishing models are rapidly reforming as mounting support for open access continues to soar. Last year, submissions soared by 30%, according to Springer Nature. But this demand is proving to be a double-edged sword as retracted articles have hit an all-time high in the past decade, and journals are flooded with AI-generated articles and fabricated research.

Naturally, this is causing bottlenecks that harm publishers, researchers, and reviewers. Publications struggle to stick to editorial schedules because reviewers scramble to evaluate papers, and authors are left waiting for months, if not years, to get published. The current peer review model was never designed to sustain this scale of demand.

Amid this, it’s crucial to recognize the role of AI as a crucial support tool for strengthening existing peer review processes and helping the industry overcome its biggest barriers.

Why Traditional Processes Are Cracking

Peer review is unpaid work. Reviewers volunteer their time in a gesture of goodwill to the academic and scientific communities. The demands are steep, and the rewards are less appealing with all the pressures compounding. Because of the rising demand and excessive strains, fewer reviewers are accepting invitations to evaluate papers. What’s more, sourcing reviewers is a growing challenge for editors, which is exacerbating backlogs and dragging out publishing timelines.

At the same time, academics are facing a ‘publish or perish’ environment. To progress professionally, they are expected to have published work in an increasingly congested landscape. This is a universal expectation around the world, and there is pressure to stand out amid the noise, on top of the strain to meet the resulting demand. These factors are behind the peer review crisis, where the current system, as it stands, is not sustainable.

To clarify, AI tools should not be adopted in the peer-review process to replace human judgment, but rather to strengthen it. Authors, and editors, expect a certain level of respect and dedication in the review process, as is given to research.

However, small mistakes can easily slip by unnoticed, especially when reviewers are rushing to review a large number of papers. To illustrate, one study found that in 2020, over 100 million hours – the equivalent of 15,000 years – were spent on global peer reviews.

Yet that does not make these errors an acceptable norm. What might appear to be an insignificant oversight, such as a minuscule flaw in data, can damage readers’ perceptions of the research quality.

AI as a Support Tool, Not a Substitute

AI tools flag subtle mistakes that can otherwise be overlooked. Citation errors, data omissions, and potential overlaps with existing literature are just some of the examples of small but significant errors that reviewers can miss.

These tools are designed to handle low-risk tasks so that reviewers can focus on the more advanced aspects of editing, such as expert analysis and contextual assessment. A tool is limited to the data it is fed with, but a human reviewer with specialized expertise can contextualize research to a much broader and informed level. Moreover, an algorithm cannot substitute for the subtle reasoning of subject experts, nor should it. It can, however, ease some of the lower-stakes burdens they encounter.

AI can help reviewers provide clearer and more insightful feedback, but only when used wisely. It can assist in confirming citations, highlighting methodological concerns, detecting absent data, and spotting similarities with earlier published research. Another reason it appeals to reviewers is the potential for AI-enabled search to help them keep abreast of new knowledge.

Crucially, however, industry leaders, such as senior editors at Elsevier, endorse the importance of equilibrium. They view it as a useful tool for improving review quality, but only when ethically applied and always with human supervision.

Addressing the Broader Implications of AI Adoption

Worries abound among the academic community. Very recently, researchers found that the influx of submissions post-ChatGPT has surged by 42%, but the quality of writing has unfortunately taken a hit. That same study also flagged concerns around how AI is being used to inform editorial decisions when used in peer reviews.

Additionally, guidelines and regulations are still catching up with the technology to ensure tools are being used ethically and in moderation. Individuals and publications alike must understand the associated risks. Abusing these tools is a guarantee of bias, hallucination, incorrect conclusions, and false perceptions.

The academic publishing community is also concerned with privacy and confidentiality. Authors tend to feel uneasy when external tools are integrated into the review process, especially in situations where research contains sensitive data. This anxiety around potential leaks is understandable given the amount of time and effort authors dedicate to their research.

For these reasons, perceptions of AI in the review process are fairly weak. Joint research from KPMG and the University of Melbourne found that 46% of people are willing to trust AI. This investigation also revealed that an overwhelming majority (70%) believe that regulation is necessary.

These factors must be addressed, and that requires training and transparency around how AI is applied, regardless of the extent to which it is used in peer reviews. Anyone using AI in the review process has to understand its best practices and apply critical thinking throughout. Reviewers and publishing teams will need to build their skills in AI and data literacy, which includes being able to interpret model behavior, recognize limitations, and mitigate risks such as hallucinations.

Transparency is foundational to adoption strategies, and publications and reviewers have to be prepared to clarify and justify AI deployment. Routine audits can assist in confirming precision and reliability. It is also crucial for publications to specify how models are trained and which data guides them. In academia, authors must always demonstrate their research methods and rationale. The same level of disclosure should apply to AI usage.

The security of these tools cannot be ignored, either. Data encryption adds a much-needed layer of cyber protection from third-party threats such as potential hacking attempts. Again, audits are a vital part of strengthening security because they are designed to pick up on any blind spots before they cause harm.

The industry is at a crossroads, either choosing to reject technology in a bid to stick with traditional models of vetting research or embracing it. The former means that it is only a matter of time before the process crumbles under the demand. The latter, however, empowers and enables publishers and reviewers not just to keep pace, but to stay ahead of ballooning demand.