Review in the age of generative AI

Generative AI technologies can offer sustainable and impactful solutions, writes Markus Kaindl

Scholarly publishing faces challenges that affect researchers, editors, and peer reviewers. They range from: the increase in volume submissions and maintenance of high-quality standards; ensuring a fair, timely review process; and balancing high quality peer review with their own research and professional commitments. 

Having worked with researchers for over a decade, across different sectors, I believe that Generative AI (GenAI) technologies can offer sustainable and impactful solutions to many of the issues I have seen.  A recent Nature Biomedical Engineering Editorial outlined an inspiring vision - that within the next few years, AI-driven systems will have transformed peer review in research. How? Well, human experts will focus on high-level insights and checking, while AI will handle the more fundamental elements. This has the potential to transform how research is published, evaluated, and discussed.  

As the research ecosystem has become more complex, evaluating research outputs has increased in complexity, too. So, we need to ask ourselves - how can we make this more effective for everyone involved?  Let me share some examples of the experiences I’ve had so far at Springer Nature, and what we have learned from our pilots. 

Familiar challenges and benefits of AI  

For authors:

I often hear frustrations about long publication times and multiple rounds of revision from authors. These delays can significantly hinder the dissemination of research. Although the adoption of preprints across disciplines has changed matters significantly, the potential for propagation of unverified information still is a risk. I've also noticed that many authors struggle with finding the most appropriate journal and then explaining both discipline-specific and cross-disciplinary work to reviewers. 

I'm excited about how AI can help authors by providing actionable feedback earlier in the drafting and submission process. From what I've seen in our early pilots with data availability or ethics statements, this timely intervention can lead to fewer amendments and quicker publication times. Additionally, GenAI's potential to improve the clarity of cross-disciplinary research has deeply impressed me as it breaks down the entry barrier of complex papers with simpler language. 

For editors:

In my role, I work closely with editors who face the challenge of making reliable decisions amidst an increasing volume of submissions. One editor told me that finding qualified reviewers who can provide comprehensive, cross-disciplinary feedback, can be difficult, and when the day only has 24 hours, the time pressure on editors is big.  

This is where I see AI support tools making a real difference. Drafting pre-review editorial notes or checking if a paper fits a journal’s scope are promising use cases that significantly speed up recurring steps in our editors’ workflows. In our pilot experiments, I've been astonished by GenAI's ability to “rescue” papers by providing personalised, interactive insights. By analysing papers and highlighting key strengths and deficiencies, these tools could transform the editorial process, speeding up process where necessary, whilst maintaining high editorial standards. 

For referees:  

In my discussions with reviewers, I often hear about time constraints and the challenge of providing detailed, constructive feedback, especially for highly specialized or interdisciplinary work. One reviewer confided in me that maintaining objectivity and avoiding personal biases is an ongoing challenge, especially when you're rushed. The good news is that many disciplines circumvent that challenge with double blind peer review, and with peer review system Snapp we are focused on helping our communities further. 

I'm optimistic about how AI will help reviewers by providing initial review drafts. In our early trials, I've seen how this enables reviewers to apply their expertise more effectively, making the experience more constructive and rewarding. Personally, I'm most excited about the AI's capability for cross-disciplinary, non-selective reviewing. I believe this will both broaden and deepen the peer-review process, helping reviewers navigate complex interdisciplinary submissions and augment their capabilities and expertise. 

Defining our role as a publisher exploring generative AI 

Being a part of this exploration phase into the integration of generative AI in academic publishing has been one of the most exciting challenges of my career. Our teams are focused on developing AI support tools that can transform the editorial process while maintaining the highest standards of quality and integrity. Springer Nature is already rightfully investing heavily in protecting research integrity, most recenby by launching Gepetto and Snapshot, two new AI tools to protect research integrity. So, what about GenAI which can potentially be used to quickly spot pseudoscientific or problematic papers with dual-use research of concern? Our approach is simple:

1. Human-centred. As I never cease to stress, these AI technologies are here to aid and enhance human capabilities, not replace them. 

2. Complementary integration with existing services and platforms. I've been impressed by how our AI tools build upon the capabilities of our already highly accurate submission screening systems. 

3. Active engagement with our communities. Speaking with authors, editors, and reviewers to ensure our AI tools meet and exceed their expectations for reliability, consistency, and quality has the highest priority for us. 

4. Taking bias and manipulation through AI very seriously we implemented rigorous testing and feedback loops, deeply coupled to Springer Nature’s AI principles. 

5. Pilot projects that explore GenAI’s novel interrogation skills on contextualization and interpretation in the review process. I continue to be amazed by the insights these trials are generating for editors once they engage and interact with it. 

6. A safe environment for testing that does not disclose any unpublished manuscripts. 

Throughout our pilot phases, we have been clear about our ambitions: to keep pace with technological advancements to drive discovery, promote equity, and protect research quality across all disciplines. Generative AI is already making its way into assisting peer review. We want to intentionally explore it in partnership with the communities we serve so that together we understand the limitations firsthand. Embracing this new technology and being part of the conversation means that together we can help define practices for sustainable and effective use of GenAI for the benefit of our communties and the future of scholarly publishing.

Markus Kaindl is Director of Content Innovation at Springer Nature