In the AI era, the human struggle can no longer be the point of research

For a body of people trained in healthy scepticism, academics are remarkably willing to take certain claims at face value when they align with their preconceptions. And that is especially true when it comes to artificial intelligence (AI).

I recently participated in a panel discussion on AI and academic research, with an audience of international business school faculty. The conversation, which centred on the findings of a recent paper, covered ground familiar to anyone in a research-intensive institution: the massive increase in journal submissions after the arrival of large language models, the apparent decline in quality accompanying that surge, and a peer review system kept operational only by a dramatic increase in volunteer labour.

Faculty who’d listened with visible concern to the reports of increased submissions relaxed perceptibly when the quality problem was flagged. The negative appraisal of AI-assisted work was somehow reassuring – evidence, perhaps, that machine cognition was no substitute for the real thing.

The relief was short-lived, however, when it was pointed out that “AI slop” was more a reflection of the author’s lack of talent and care, and that high-quality work is a very real possibility when the process is managed by somebody with domain-specific knowledge and a talent for prompting.

Yet what followed was more revealing still: a sustained, at times impassioned, defence of the human struggle in scholarly work. The intrinsic value of academic output, for many of those present, was fundamentally related to the cognitive effort expended in its production. 

The case for struggle in the humanities and social sciences rests on a set of interrelated convictions. That unaided human thought produces something qualitatively irreplaceable and transcendentally important. That intellectual toil develops capacities that cannot otherwise be cultivated. And that the value of a contribution is inextricably linked with the intellectual challenge of producing it.

These convictions are more than convenient institutional beliefs. They developed within disciplines that have philosophical traditions with genuine depth, in which interpretation is fundamental, meaning is almost always contested, and the researcher and the work they produce are inseparable. When unaided human thought was the only available means of production, the weight of this stance was difficult to dispute.

The arrival of AI has, however, presented a fundamental challenge. When a tool can replicate and, in some cases, exceed the results of human synthesis, the belief that the struggle is the point requires urgent consideration. I’d argue the case is genuinely compelling in one area – the development of new researchers, where wrestling with literature, failing productively and learning to craft a problem are critical capacities that need to be developed. This process might itself become AI-augmented, but the acquiring of such formative competencies remains essential.

The case has far less relevance to how research should be produced, however, by those whose capacities are already formed – and even less relevance as a principle that speaks to the value of final research.

Source: 

Getty Images montage

When AI can bridge the gap from concept to intelligent synthesis within hours, clinging devoutly to human process is about more than philosophical conviction or a belief in the value of craft and artisanship – particularly given that it comes against a backdrop of longstanding and largely unaddressed criticisms of research relevance and impact. It is an act of intellectual indulgence. It supports longstanding institutional arrangements – notably those related to academics’ desire to strategically develop professional identity within the field and, more broadly, to self-actualise: to prioritise personal growth, creative freedom and self-fulfilment above other considerations.

Stripped of its romantic rhetoric, then, the response of many in the academy to the question of human versus machine outputs is closer to that of a professional guild or religious order than a scientific community.

Consider how the logic plays out in fields less invested in the question. Medical imaging has been transformed by algorithms detecting narrow, high-volume patterns with an accuracy exceeding that of specialist clinicians. Would we contest a cancer diagnosis because the imaging was algorithm-assisted? Climate modelling, genomic research and drug discovery depend on computational tools performing work no human could perform unaided. Would we reject climate projections because a machine ran the model? The answer is self-evidently no.

Directed by humans in these fields, AI is fast evolving into an engine of discovery. One example is the Nobel prizewinning AlphaFold, which solved a 50-year challenge in structural biology. Another is autonomous networks saving centuries of lab work by predicting millions of novel inorganic crystals, revolutionising materials science. Here, the machine doesn’t threaten the idea of science. It’s expanding the boundaries of what’s knowable.

The argument that humanities and social sciences are different – that the synthesis and interpretation they specialise in shouldn’t be delegated to AI – is an assertion that mischaracterises what is happening in the technology-augmented writing process. The intellectual responsibilities of a researcher are unchanged by the tools they use. The ability to identify a meaningful question, frame it correctly, steer inquiry productively, interpret, iterate, validate what emerges, and then stake professional reputation on the output – none of this is delegated to a machine.

A researcher who adds their name to a paper accepts accountability for what it claims, whether they write every sentence in longhand or lean on AI tools to help draft a literature review. What changes is the speed and, in skilled hands, the quality of execution. This is not a threat to scholarly integrity – it’s what tools do.

If the contribution is sound, rigorously grounded and accurately cited, the mechanism of its production is irrelevant to its value. To argue otherwise is to confuse the craft of academic writing with the value of academic research – and to protect the former at the expense of the latter.

Source: 

Getty Images montage

When a publishable manuscript can be produced in so little time, however, the question of whether it was worth producing in the first place does become impossible to ignore. As the marginal cost of production collapses, the volume of outputs inevitably skyrockets, and the signal-to-noise ratio plummets. The immediate responses from publishers are likely to include submission fees dressed up as quality filters (how very convenient), the proliferation of symbolic AI-detection tools and sanctions, a tiered publishing model, with a smaller number of reputable journals maintaining human peer review, and an increasing switch to data harvesting as an alternative business model. These are the reactions of a system protecting its existing architecture and revenues. Whether the longer-term pressure forces the academy itself to undercut all this by reorienting its incentives towards research impact over output remains uncertain.

This is a key reason why AI is so uncomfortable: the value problem it threatens to expose isn’t new, and it isn’t confined to a single discipline. C. Wright Mills was ridiculing grand theorists of sociology in the 1950s, Stanislav Andreski accused the social sciences of “sorcery” in the 1970s, and Alan Sokal cruelly exposed the pretensions of cultural theory in the 1990s. More recently, Mats Alvesson, Yiannis Gabriel and Roland Paulsen have highlighted the perils of “publish or perish” logics and the proliferation of “meaningless research”: work of little value to society, and of only modest value to its authors beyond securing employment and promotion.

Similarly, drawing on experiences from the business and management domain, Dennis Tourish describes a disconnected research culture in which nonsense prevails. And almost 25 years after Pfeffer and Fong’s withering critique of business school research’s relevance and usefulness, it’s a stretch to identify a single research contribution in the intervening period that has substantively influenced practice. To be clear, AI slop is not corrupting a healthy system. It’s forcing the sector to reap what it has sown – by delivering more of the same, at scale.

It is misguided to think that any comfort is offered by the disclosure framework, based on principles of caution, human accountability and transparency, that academic publishers have converged around. As a recent large-scale analysis of more than 5.2 million abstracts across more than 5,000 leading journals reported, while 70 per cent of journals had AI policies in place in early 2025 – most requiring disclosure – the study’s authors found no discernible difference in AI adoption between journals with rules and those without.  

A deeper dive into 164,000 full-text papers in the same study found that fewer than 0.1 per cent of those published since 2023 had explicitly disclosed AI use. If that figure is accurate, I’d suggest that academics are either the most virtuous or the most tormented of all professionals, aware of the possibilities of AI yet constrained by higher principle from capitalising on it. Alternatively, they are either genuinely confused by the rules or not entirely honest in their disclosures. You decide.

Source: 

Getty Images montage

Whatever the response of researchers, it’s clear that publishers are tying themselves into pretzels when it comes to AI disclosure rules. They’re requiring transparency for a technology that cannot be reliably detected, defining rules that are difficult or impossible to enforce, and aiming to restrict a tool they themselves use – and will increasingly depend upon for peer review purposes.

We even have the bizarre spectacle of journals in areas like innovation and management publishing papers that demonstrate the value of AI to ideation and discovery while those journals simultaneously regulate its use in research submitted to their own pages. And, in a perverse slap in the face for authors, we see certain publishers discreetly licensing academic research to AI companies for training purposes – without compensation for the authors, naturally.

Yet that arrangement does flag up a legitimate reason, beyond a defence of existing commercial arrangements, that publishers have to care about human contribution. That is the “model collapse” that AI researchers have warned of: the degradation of systems trained recursively on machine-generated content as scarce original knowledge thins out, variance is lost and outputs converge on a bland, potentially biased statistical average. Some will no doubt see this as a vindication: proof that machine text corrupts while human text enriches.

But this is an oversimplistic interpretation. An AI model is starved when text contains no new information: recombination without discovery. But it cares less whether a sentence is typed or generated.

Slop, in other words, shouldn’t be defined by its method of production, but by its lack of informational density – and the handcrafted variety has been accumulating in journals for decades. What protects a model from collapse is precisely what might invigorate academic research: less verbiage and more original data, novel observations and bold interpretations – validated by a researcher willing to stand behind them.

That the producers of this valuable raw material are the one party in the licensing chain likely to go uncompensated should be the real cause for concern. But publishers and researchers are not separate actors in this story. They’re mutually dependent within the established system, and the responses each has reached for reflect that dependence with some predictability.

When faculty defend the primacy of unaided human effort, they’re drawing on a genuine and longstanding conviction that effortful intellectual engagement defines the scholar in ways that matter. While deserving respect, however, this philosophical position of substance becomes a regulatory instrument of self-preservation when it is translated into disclosure requirements, boundary violation taxonomies, and submission guidelines that prohibit AI use for “intellectual tasks”.

Its primary function is not to advance understanding but, instead, to protect a specific kind of knowledge production: one that allowed slop to flourish but is legible to journal editors, navigable by review committees, and optimised for metrics that determine status. The struggle being defended is not the struggle to understand the world. It is the struggle of remaining actualised within a system that has become, in large part, self-referential and self-serving.

AI, deployed seriously and without the burden of stigma, could begin to address this – not by accelerating the production of text but by enabling synthesis that is broader, more cross-disciplinary and considerably more ambitious. 

In a world where AI augments intellectual inquiry at negligible cost, the struggle can no longer be the objective of knowledge production. A system that continues to make such a choice dresses that choice in the language of integrity and tries to regulate the tools that challenge its assumptions. That system is not defending scientific progress – it’s defending itself.

Ian N. Richardson is a faculty member and director of executive education at Stockholm Business School at Stockholm University. He is co-founder of the national Swedish programme AI for Executives, which seeks to drive board-level understanding and organisational adoption of AI across industries and sectors.