KEY TAKEAWAYS
- AI use in scientific research is increasing both productivity and the size and complexity of datasets.
- Adoption of AI tools by publishers could enable them to streamline the peer review process and safeguard against circulating flawed data.
Artificial intelligence (AI) is transforming scientific research and increasing productivity. But how can publishers keep up with the consequent surge in submissions, when peer reviewers are already at capacity and the current system may not be fit for purpose? In a recent article for the London School of Economics Impact Blog, Simone Ragavooloo calls on publishers to harness AI to:
- match the pace of AI-driven scientific output, and
- protect research integrity.
Can AI-enabled peer review match increased scientific output?
The Organisation for Economic Cooperation and Development’s 2023 Artificial Intelligence in Science report states “raising the productivity of research could be the most economically and socially valuable of all the uses of AI”. To realise this potential, however, all steps of the research-to-publication process must align. Ragavooloo argues that publishers must “meet like with like”, utilising AI to streamline the peer review process. For example, Ragavooloo envisions AI doing the “heavy lifting” in areas like statistical analysis, where lack of expertise or statistical training can be limiting factors for reviewers. This would free up human reviewers to focus on aspects requiring greater human insight.
Protecting scientific discourse: can AI catch faulty data?
AI is producing increasingly large and complex datasets. This brings an increased risk of error, which, if unchecked, could lead to widespread dissemination of faulty big data. This prompts another role for AI: AI can identify methodological or statistical errors within vast quantities of information at a rate that is simply impossible for humans. While tools such as Frontiers’ Artificial Intelligence Review Assistant (AIRA) and the STM Integrity Hub are already available to help reviewers triage submitted articles, Ragavooloo believes there is still an unmet need for AI-assisted peer review applications, to ultimately prevent circulation of flawed data.
AI can identify methodological or statistical errors within vast quantities of information at a rate that is simply impossible for humans.
Looking ahead
While recognising we are in a transitional phase, Ragavooloo emphasises that publishers “have the scale and technological expertise” to develop more AI tools, calling on them to put their trust in AI and create “an open path forward” for AI-driven innovation.
————————————————–
Categories