For decades, academia has operated on a convenient assumption: academic productivity is a reasonable proxy for academic quality.
We count publications, citations and impact. These numbers influence promotion, grants, rankings and academic reputation. Over time, however, the proxy has increasingly become the objective itself.
A recent paper by Thomas Manabu, The Machine Consumes Itself: Artificial Intelligence and the End of Publish-or-Perish, presents a provocative argument: generative AI is not creating the problem. It is exposing a weakness that was already built into academia.
Academia created the perfect product for AI
Academic publishing has become highly standardised. Papers follow predictable structures, styles and conventions. This is necessary for scientific communication, but it has also made academic outputs remarkably reproducible.
And generative AI is very good at reproducing them.
AI can assist with literature searches, analysis, manuscript writing and responses to reviewers. It is increasingly appearing in peer review itself. Producing something that looks like scholarship is becoming faster and cheaper.
If publication numbers are academic currency, AI has effectively made that currency much easier to manufacture.
Publication is becoming a weaker signal
Traditionally, a published paper implicitly represented substantial intellectual labour: reading, thinking, analysing, writing, revising and defending an argument.
AI weakens this relationship.
Someone producing twenty AI-assisted papers is not necessarily demonstrating five times the scholarship of someone producing four carefully developed papers.
This does not mean AI-assisted research is poor research. AI may actually allow good researchers to do better work. The problem is that publication volume becomes increasingly unreliable as a proxy for scholarly quality.
The bigger concern is the scholar
Perhaps the more important issue is not whether AI writes our papers, but what happens when it increasingly performs the intellectual work through which scholars themselves develop.
We become scholars by struggling with ideas, reading critically, making methodological decisions, being challenged and revising our thinking.
If AI increasingly performs these tasks, an uncomfortable possibility emerges:
The scholarly product may improve while the scholar producing it develops less.
An interesting irony
There is an intriguing aspect to this particular paper. Thomas Manabu has very little readily identifiable conventional academic footprint beyond this publication. More interestingly, the paper itself openly acknowledges substantial use of generative AI in drafting and structuring sections, identifying literature, improving prose, and assisting in developing its theoretical framework and arguments.
That does not invalidate the paper.
Arguments should ultimately stand on their reasoning and evidence, not simply on the credentials of their author. And the central questions raised here remain valid.
But there is an extraordinary irony.
A theoretically sophisticated, peer-reviewed paper about AI weakening the relationship between publication and scholarly labour was itself produced with substantial AI assistance by an author whose conventional academic background is difficult to establish.
Perhaps the paper inadvertently demonstrates its own argument: in the age of AI, the existence of a polished peer-reviewed publication may tell us increasingly little about the scholarly journey, expertise and intellectual labour behind it.
My suggestion: value what scholarship actually does
Manabu points towards broader and more qualitative ways of recognising scholarship. I agree with the direction, but there remains a practical problem. Measures such as intellectual leadership, originality and scholarly contribution are inherently subjective. Publication numbers became attractive precisely because they are easy to count.
For the natural and health sciences, perhaps there is another way.
Rather than replacing publication counts with another set of subjective assessments, we could place greater emphasis on what the research actually creates, enables, changes or solves.
Did it produce a new discovery, diagnostic tool, treatment, device, technology, dataset, method or intervention? Did it change clinical practice or policy? Was something implemented? Did it improve health, services or systems? Did it solve a real societal problem?
This changes the emphasis from academic output to demonstrable contribution and consequence.
Under such a system, the paper remains important, but its role changes. It documents and communicates the scholarship rather than serving as the principal evidence that scholarship occurred.
This also changes how we should think about AI. In natural and health sciences, I see little reason to object to AI being used as a language or writing tool if the underlying science is rigorous and the substantive contribution can be independently demonstrated. The value lies in the discovery, experiment, data, intervention or innovation, not in whether every sentence describing it was manually constructed.
The situation may be different in the arts, humanities and philosophy, where the argument, interpretation or creative expression may itself be the scholarly contribution. Here, perhaps the value lies not simply in what is written, but in the intellectual dialogue it generates: whether it challenges existing ideas, provokes new questions, offers new interpretations or stimulates meaningful scholarly discourse.
This suggests that we should not search for another universal metric to replace publication counts. Different disciplines may need different evidence of scholarship: in natural and health sciences, what the work discovers, creates, changes or solves; in the arts, humanities and philosophy, what it questions, interprets, creates or brings into meaningful intellectual dialogue.
A simpler principle may therefore be more useful:
Do not assess scholars primarily by what they write. Assess them increasingly by what their scholarship creates, enables, changes or solves.
AI may not be the problem after all. It may simply be forcing academia to recognise that the paper should be a record and communication of scholarship, rather than its primary measure.
Reference
Manabu, T. (2026). The machine consumes itself: Artificial intelligence and the end of publish-or-perish. Postdigital Science and Education. Advance online publication. https://doi.org/10.1007/s42438-026-00666-0