An associate editor of Frontiers in Systems Neuroscience has resigned in protest of the publisher’s artificial-intelligence (AI) system, arguing that its misuse is undermining academic integrity at the journal.
Michael Okun, associate professor of neuroscience at the University of Nottingham, announced his resignation on social media on 10 June, after journal officials told him they couldn’t turn off AI-related tools that automatically find potential peer reviewers.
The features are in the publisher’s Artificial Intelligence Research Assistant (AIRA) system, which Frontiers rolled out a few years ago in a bid to help editors carry out peer review, among other reasons.
In an interview with The Transmitter, Okun explained that in early May he received a request from the journal to handle a manuscript on the dynamics of neuronal networks. He sent it out to around a half-dozen potential referees, but before any of them accepted his request, AIRA started sending out its own reviewer requests to other researchers, Okun says. Two referees AIRA contacted—who had no relevant expertise, according to Okun—accepted the referee requests.
That prompted Okun to flag the problem to Frontiers. Okun says he was told to ignore the requests AIRA had sent out, and to continue seeking reviewers himself. But when he did so, and one of his invited experts accepted the request, AIRA revoked all his other invitations. “It’s just inconceivable that a manual invitation to someone who is actually an expert in the field is revoked just a few hours after it is sent,” Okun says.
That was a deal-breaker for Okun. “The journal wants to publish as many submissions as possible and doesn’t care too much about quality,” he says. “These are not bugs but intentional features, designed to essentially remove the human editors as much as possible.”
In a statement sent to The Transmitter, Frederick Fenter, Frontiers’ chief executive editor in Lausanne, Switzerland, says each feature Frontiers has built is intended to save scientists time so they can focus on research. “We regret that Dr. Okun had a bad experience, and my invitation to discuss his concerns directly remains open.”
Fenter notes that editors have the ability to pause reviewer invitations at any stage, something that was not made clear in Okun’s case. So Frontiers is conducting an internal review, Fenter says, to ascertain where the breakdown in communication took place and ensure similar instances don’t take place in future.
The Transmitter reached out to dozens of editors listed on the Frontiers in Systems Neuroscience website.
One of them, Shuzo Sakata, professor of systems neuroscience at the University of Strathclyde, says he had a similar experience with the journal’s automated system. “Even after I decided to reject a manuscript, the system invited another reviewer without my permission,” Sakata recalls.
Sakata says he no longer accepts the journal’s requests to oversee editing. “Automation is important and often helpful, but their system has crossed the threshold.”
Even after I decided to reject a manuscript, the system invited another reviewer without my permission.
—Shuzo Sakata
Several other editors The Transmitter heard from said they haven’t experienced the same issues with AIRA that Okun has highlighted, with some noting that they haven’t edited that many papers for the journal.
Other researchers noted that AIRA contacts them with unusual requests. “Frontiers is constantly sending me review or editor requests that have nothing to do with my expertise,” says Earl K. Miller, Picower Professor of Neuroscience at the Massachusetts Institute of Technology.
“I do receive frequent requests to act as editor, but 97 percent of the time the manuscript in question is non-overlapping with my expertise,” adds Mathew Diamond, who leads the Explorations into the Neuronal Foundations of Sensory Experience Lab at the International School for Advanced Studies.
Frontiers has used AI since 2018 to help its editors identify and invite reviewers, according to Fenter.
But human judgment is central to the process, he says. “Editors retain full discretion over whether to use automated suggestions; they can review all recommendations before any invitation is sent, decide who is invited and in what order, and can add, remove or reassign reviewers at any point,” Fenter adds. “Accept and reject decisions always rest with a human editor, without exception.”
As for using AI to assist editors, Fenter says this has clear benefits: The average time to secure peer reviewers willing to do the work has fallen by 30 percent at Frontiers in the past year. Frontiers’ data suggests that 80 percent of handling editors find that automation is helpful and speeds up reviewer assignment, Fenter says, and 94 percent of authors rate their peer-review experience as good or excellent.
“Editors are also supported by a dedicated Research Integrity Team, who maintains oversight of automated quality checks at every stage of our editorial process,” Fenter writes. “We remain committed to keeping human expertise at the centre of peer review, and to contributing openly to the broader conversation about the responsible use of AI in scholarly publishing.”
Okun says the idea of using large language models to find external experts is a good one. “I’m totally in favor of it, as long as the human editor is consulted,” he says.
Alternatively, Okun says that he is also fine with a fully automated system, as long as publishers disclose that all the editing was done by AI. “As long as it is honest and you don’t just slap someone’s name on it, that’s potentially also perfectly fine,” he says.
This article was first published at The Transmitter.
