When Reviewers Are Machines
Trust, integrity and the crisis of peer review
This event is part of a demonstration of what Lera offers for running knowledge events. Its content is written with care and draws on real sources, but the conference has not taken place, and apart from the convener its participants are fictional.
The year 2023 set an unwelcome record in the history of scholarly publishing. Richard Van Noorden reported in Nature that more than ten thousand papers were retracted that year, more than in any year before, most of them from the journals of a single publisher that had until then been known simply as an open-access publisher. Many of these papers had appeared in special issues whose review had been manipulated: fake reviewers, recycled reports, and networks that produced and placed papers like merchandise. A year earlier the Committee on Publication Ethics and the International Association of Scientific, Technical and Medical Publishers had warned about these “paper mills” in a joint report.
A new layer has now been added to the crisis. Weixin Liang and colleagues examined the reviews submitted to several large machine-learning conferences and estimated that between 6.5 and 16.9 per cent of the text of review reports had been substantially modified with language models, beyond spell-checking and minor edits. More telling still, the share was higher in reviews whose authors reported lower confidence, that were submitted close to the deadline, and whose reviewers were less likely to respond to the authors’ rebuttals.
The whole picture is this: the author, under pressure to publish, writes with the help of a machine; the exhausted reviewer reviews with the help of a machine; and the editor relies on a machine to find reviewers. A closed circle has formed in which the machine writes for the machine and the human signature is a ritual stamp. This should not be reduced to individual misconduct. Peer review is unpaid work laid on researchers who are themselves under the pressure of quantitative evaluation. A system that rewards publishing and gives nothing for reviewing hands review over to whatever can be done at the lowest cost.
In the Global South the crisis takes another form as well. Researchers in many countries are, on the one hand, under heavier pressure from indexes and rankings than others and are targeted by the advertising of paper mills; on the other, when a scandal breaks, their entire scholarly community is treated as suspect. The cost of a system designed at the centre is paid at the margins.
The way out does not lie in detection tools. It lies in moving where value sits: review should be recognised as research work, carrying the reviewer’s name, open to publication, and counted in career evaluation. Smaller, slower, signed review is better than a mass of anonymous reviews whose writers no one can any longer identify.
Questions for discussion
- If reviews are signed, how are young and vulnerable reviewers protected from retaliation by powerful authors?
- Can reviewing be given the same weight as publishing in researchers’ career evaluation?
- Where does acceptable use of machines in reviewing end?
References
Van Noorden, Richard. 2023. “More than 10,000 Research Papers Were Retracted in 2023 — a New Record.” Nature 624: 479–481.
Liang, Weixin, et al. 2024. “Monitoring AI-Modified Content at Scale: A Case Study on the Impact of ChatGPT on AI Conference Peer Reviews.” arXiv:2403.07183.
COPE and STM. 2022. Paper Mills: Research Report from COPE & STM. Committee on Publication Ethics and International Association of Scientific, Technical and Medical Publishers.