Editors’ note: Today’s post is by Neil Blair Christensen, Sales Director at the Molecular Connections Group.

Since participating in this year’s Society for Scholarly Publishing Annual Meeting, and through my work in the years prior, I’ve been thinking about how our scholarly publishing community engages with AI (artificial intelligence) — more specifically, what we might be avoiding.

Irving Janis, the Yale psychologist who developed the theory of groupthink from studying foreign policy disasters, found that the groups most vulnerable to it weren’t the incompetent ones. Quite the opposite. They were experienced, high-functioning ones, because their cohesion and shared identity made the cost of dissent high. This isn’t done through intimidation, but more through subtle social or professional rewards of agreement, and the quiet discomfort of being the person who asks the wrong question.

Lately, I wonder if we, the scholarly publishing community, are one of those groups.

At this year’s SSP meeting, I noticed a persistent disconnect between conversations happening outside the meeting rooms and the formal presentations inside (as other Scholarly Kitchen authors have acknowledged this year). Over coffee, people described how they’re using AI to get work done, practically and without ceremony. Inside the rooms, the conversations were more about how we as an industry should think about AI. Governance frameworks, responsible adoption, strategic positioning. Same technology, different lenses. When I mentioned this to David Worlock, he described it as the creeping edge of change moving forward without being in rhythm with the thought-leader world. That framing stayed with me.

I think the disconnect runs deeper than AI adoption pace.

If you look at the formal conversations our community has been having about AI in recent years, a pattern emerges. We debate governance. We build integrity screening tools. We invest in AI-assisted peer review, discovery, and production workflows. What we discuss much less is why those challenges exist in the first place, and whether AI tools are solving them or enabling their continued growth. This pattern can be seen elsewhere.

Peer review challenges, for example, have been topics in our industry for years and years. The conversation is mostly about how to manage volume growth and reviewer shortage. What is less discussed, is that this high volume of manuscripts is a product of academic incentive structures that force researchers to produce output as if there is no tomorrow. As we run out of people, we are adding technology to facilitate a flood that we, as an industry, have a bias in not stemming.

The groupthink is most acute in which problems we discuss solving. AI isn’t causing that. It’s just accelerating the consequences.

Every stakeholder in the scholarly publishing ecosystem has some financial or institutional stake in the continuation of high-volume knowledge production. Researchers need publications for career advancement, universities need output for rankings and funding, funders need signs of returns on investment, publishers generate revenue from processing and certifying that output, libraries are given budgets to provide access to it, and vendors sell tools to manage it. No one in that chain has a persistent structural incentive to question the added value of all that volume. Not one. So, our industry conversations, including those on AI, tend to optimize within that framework rather than interrogate it. Increasing volume and complexity require more AI to cope. The more AI we apply, the more volume and complexity we produce.

That’s not a conspiracy. It’s what Janis would recognize as normal groupthink dynamics. A collective rationalization of the status quo, if you will. An ingroup confidence that existing strengths can prove decisive against disruption, and an ingroup that defines which questions are asked.

Are we at risk of becoming a case study like the music industry, broadcasting, or Kodak? Maybe, but the risk isn’t disruption from the outside. It’s that our own incentives are the deeper problem, and AI is making them easier to sustain. We’re applying increasingly sophisticated interventions to a hamster wheel that is perpetually under strain. Profitable problems tend to persist until they don’t.

I’m not writing this to indict anyone, and I’m not exempting myself. I’m just as embedded in this community as anyone reading it — and I even sell AI solutions. I’m writing it because I think the most valuable thing a professional community can do is to distinguish between the problems it has incentives to solve, and the problems that actually need solving. Those aren’t always the same. Is more simply more? What are we solving for?

Author’s notes: I am grateful to Lettie Conrad for her suggested edits. Claude Opus 4.8 was used to polish the writing.