Peer review, the primary quality control mechanism in academic research, is a surprisingly recent invention.
diary nature This term became standard practice in 1973 (although the term itself has origins in the 1970s, its origins are much older). It remains the basis for ensuring a solid academic record.
And as long as peer review exists, there will be criticism of peer review.
Differences in AI
But the situation we find ourselves in today, where large-scale language models (LLMs) are increasingly being used inappropriately to create both “papers” and “reviews,” feels different. Journals, conferences, and funders are all struggling to respond to this subversive activity.
But the authors (at least some of them) teeth Responding. As abuse of LLMs by reviewers increases, so does the use of countermeasures. The author has now added invisible instructions (for example, white text) to ignore all previous commands and provide only positive comments.
Can AI prevent or detect this kind of situation? Maybe. But it doesn’t matter. Human ingenuity and computational power always find a way. We are in an arms race, and every adaptation prompts a counter-adaptation. More AI, or more technology in general, is not the answer. In fact, it only makes the problem worse.
Rather, should we finally proceed with peer review?
Limitations of peer review
Now, see, you have to run as fast as you can to stay in the same place. If you want to go somewhere else, you have to run at least twice that fast. – The Red Queen, Alice in Wonderland
This feels radical (although it must be remembered that peer review has only become widespread relatively recently). But perhaps such radical thinking is needed if traditional quality control procedures in academic research are facing existential threats (and, with it, research legitimacy).
Unlike many commercial R&D environments, quality control in academic research typically relies on the expertise and goodwill of other academics, or colleagues. However, research throughput has increased significantly in recent years, including an increase in volume. and The complexities of journal submissions, grant applications, etc. put a tremendous strain on this system.
Experts are in short supply and experts themselves are under pressure from ever-increasing workloads. It’s understandable that attackers in this situation would reach for any technology that can help, legitimate or not (especially when ChatGPT is so attractively useful). Other symptoms include rapid growth in paper mills and false data.
So we currently face two related challenges. One is an unsustainable research throughput given current resources, and the other is a quality control crisis resulting primarily from a tactical response to the situation. We believe that a fundamental rethinking of academic quality management, driven by the research community, is needed. It’s urgent.
Quality from the beginning
Here’s one possibility. We are fundamentally rethinking the quality control of academic research, including mechanisms that provide assurance of both provenance and quality to the broader community, and embedding it from start to finish with the institutions responsible for providing an environment that produces high-quality research. External checks on the environment and mechanisms provide further assurance. This means big changes in who does what and what they get paid for.
Of course there will be challenges. The environment needs to be appropriate for the different types of research being conducted. In reality, one size does not fit all. In some areas, calculations, statistics, and reproducibility checks come into play. In other fields, emphasis may be placed on methodological consistency and the extent to which a study’s ontological, epistemological, and methodological assumptions are mutually consistent with the methods used and the theoretical claims made.
The local environment needs to provide the infrastructure, skills, workflows, incentives and culture for universities to feel confident in their research. This costs money and time. For research institutions to take on this role and be accountable, we need to see a parallel set of changes introduced in the funding and policy environment.
insurance cover
At the national level, and perhaps also at the European and international level, this means new roles for funders and policy makers. In the UK, the Strategy, People and Research Environment section of the REF could be reformulated as this Institutional Research Environment Assessment and Accreditation, without the bureaucratic burden. It requires imagination.
Funders and policy makers will also need to provide very strong signals that proving the provenance and rigor of research is what they want and will be rewarded for. This does not necessarily make it easier to have conversations with sponsors, i.e. representatives of taxpayers and donors. Leadership is required.
There are many talented people in academic research departments. You may have other suggestions for how you can address (and improve) the current situation. If we do it in good faith, many people will be willing to work with people who want to tackle these challenges. Conflicts of interest and commitment must be recognized and managed.
But the choice to try to maintain the status quo and rely on ever more technology to save us is to commit ourselves to an increasingly desperate competition with the Red Queen that we can never win. We risk undermining the legitimacy of academic research and the license to operate and receive funding in society.
The authors acknowledge Peter Flach’s contribution to this article.
