In Defence of Slop

Paul Morrill | November 2026

There was a time when, if you considered an argument intellectually deficient, you were expected to explain why. Perhaps its premises were unsound, its reasoning defective, its evidence inadequate or its conclusions unwarranted. There was, at least in principle, some obligation to engage with what was being said.

Now, apparently, we have slop.

AI slop, philosophical slop, academic slop, literary slop. The word appears to have become a convenient way of disposing of almost anything that offends our intellectual sensibilities. It is wonderfully economical. One word, and the rather tiresome business of criticism can be dispensed with.

I find this particularly interesting when the accusation comes from people who presumably consider themselves sufficiently knowledgeable to recognise intellectual inadequacy when they encounter it.

Surely, if something is demonstrably poor, we ought to be able to explain what makes it poor?

I have no particular desire to defend fabricated research, incoherent arguments or the indiscriminate production of material that contributes little beyond occupying space. There is plenty of dreadful material circulating online, and artificial intelligence has made its production considerably easier.

Nevertheless, I am becoming increasingly suspicious of the casual ease with which slop is employed to dismiss work without examining it.

My concern is not principally with the existence of poor intellectual work. Poor work has always existed. What interests me is the possibility that the language used to identify it is becoming a substitute for the very critical reasoning its users claim to defend.

There is a certain irony in that.

When a judgement becomes an argument

A useful starting point is to distinguish the quality of an intellectual contribution from the judgement we make about it.

An argument may be factually incorrect, logically inconsistent, derivative or unsupported by evidence. Such deficiencies can be identified and, where appropriate, challenged.

Alternatively, an argument may simply irritate us. Perhaps it employs a style we dislike, challenges a position to which we are attached, or comes from someone whose qualifications we consider insufficient. Perhaps we suspect it was written with the assistance of artificial intelligence.

None of these observations necessarily establishes that the argument is defective.

Philosophers have long recognised the difficulties associated with evaluating claims according to their origins rather than their merits. The genetic fallacy is relevant here, although we should be careful about applying it indiscriminately. The provenance of information can be important when assessing its reliability. An anonymous assertion, an account from a recognised specialist and a statement generated by a language model may reasonably attract different levels of initial confidence.

But provenance is not the same as truth.

A proposition does not become false because it was generated by a machine, any more than it becomes true because it was written by a professor.

There are legitimate reasons to question AI-generated material. Such systems may produce convincing but inaccurate statements, fabricate references or present uncertainty as established fact. They may also encourage authors to publish arguments they have insufficient understanding to defend.

These are concerns about reliability, intellectual responsibility and authorship. They deserve serious attention.

What I question is the tendency to replace those concerns with a general judgement about intellectual worth.

We cease asking whether an argument is sound and begin asking whether it resembles something produced by a machine.

These are different questions.

LinkedIn and the industrialisation of slop

LinkedIn provides an interesting setting in which to observe this development.

In June 2026, the platform announced measures to reduce generic, low-effort AI-generated material while acknowledging that AI could legitimately assist people in expressing their own ideas (LinkedIn, 2026a).

By September, LinkedIn reported that more than one million members had used its new reporting mechanism for suspected AI slop. The company also claimed that views of material classified in this way had fallen by 40% (LinkedIn, 2026b).

These figures are interesting, although they need to be interpreted carefully. They tell us something about LinkedIn's moderation activity and the willingness of its members to report suspected slop. They do not tell us whether those judgements were correct, nor whether the material identified was necessarily deficient.

The distinction matters.

LinkedIn defines AI slop in terms of generic, repetitive, low-effort material that lacks substance or an identifiable perspective. That seems a reasonable description of a particular kind of poor writing. Indeed, I suspect most of us have encountered enough of it to recognise what the company is attempting to address.

What concerns me is the movement from a judgement about quality to a classification that may influence whether other people encounter the material at all.

A reader's impression becomes part of the mechanism through which visibility is determined.

This is not inherently objectionable. Social media platforms have legitimate reasons to discourage spam, automated engagement and mass-produced material designed principally to attract attention. The problem arises when we assume that the ability to recognise a familiar style is equivalent to the ability to assess intellectual substance.

I have noticed something similar in discussions on LinkedIn. Lengthy prose is increasingly treated with suspicion, particularly when it adopts a polished or formal register. There are comments suggesting that such writing is not worth reading because it resembles AI output.

I find this a curious position.

Surely the question should be whether the author has anything worth saying?

A recent discussion concerning a statistical figure illustrates the problem rather well. The circulated figure distinguished between material labelled Human Writing and AI Slop. My concern was that the labels appeared to conflate a distinction concerning the origin of the writing with a judgement about its quality.

The analysis involved linear discriminant analysis, a perfectly legitimate statistical technique for distinguishing predefined categories according to measured characteristics.

But identifying differences between categories does not establish that one category is intellectually inferior to another.

What, precisely, was being measured?

This is a familiar problem in psychology. We can construct sophisticated statistical models and still make questionable inferences if the constructs being examined are inadequately defined. Messick (1995) made this point in relation to the interpretation and consequences of psychological and educational measurement.

The issue here is not whether AI-generated writing can be distinguished from human writing under particular conditions. It is whether such a distinction tells us anything reliable about intellectual merit.

A technically impressive analysis can still be accompanied by a rather poor interpretation.

And perhaps the most unfortunate aspect is that the evaluative label may acquire an appearance of scientific legitimacy simply through its association with the statistical procedure.

I wonder how much of the current enthusiasm for identifying slop rests upon a similar confusion.

We were producing slop long before AI arrived

I have discussed the deterioration of institutional language elsewhere, particularly in Academic Language Was Flattening Before AI Arrived (Morrill, 2026a).

My argument was that universities had already become remarkably proficient at producing language that sounded authoritative while communicating very little.

Anyone who has spent time in academic administration will recognise the vocabulary. Strategic alignment, stakeholder engagement, transformational delivery, value creation and the apparently inexhaustible pursuit of excellence.

These expressions are not necessarily meaningless. In particular circumstances they can describe perfectly reasonable activities. The difficulty arises when they become substitutes for explaining what someone proposes to do, why it should be done and how its consequences might be evaluated.

AI did not invent this language.

It encountered an enormous quantity of it in the material from which it learned and became rather good at reproducing it.

I find this an uncomfortable point for those who regard AI as principally responsible for the apparent decline in intellectual expression.

If artificial intelligence can reproduce the familiar language of academic and professional authority with relatively little difficulty, perhaps we should question how much intellectual work that language was performing in the first place.

A committee paper does not become intellectually substantial because its author spent an afternoon writing it. Nor does an argument become more coherent because it appears beneath the name of a distinguished academic.

Equally, a useful idea does not necessarily lose its value because the person expressing it received assistance in finding an appropriate register.

This is not to suggest that the process of writing is unimportant. Writing can itself be a means of developing understanding. There are good educational reasons for requiring students to construct and defend arguments without continually outsourcing the intellectual labour involved.

But we should not confuse the effort expended in producing something with the quality of the resulting contribution.

Someone may work very hard to produce something dreadful.

Another person may produce something useful with comparatively little effort.

The distinction may offend our sense of intellectual fairness, particularly when we have invested considerable time acquiring skills that technology can now reproduce. But that does not invalidate it.

Philosophical slop and the ownership of questions

The extension of the word slop into philosophy is particularly interesting.

Philosophy has never been especially tidy. It contains arguments that remain unresolved, concepts that resist satisfactory definition and disagreements that have survived centuries of scholarly attention.

Some philosophical writing is undoubtedly poor. There are arguments that depend upon equivocation, claims that cannot be defended and passages that appear profound largely because they are difficult to understand.

But I wonder whether we are beginning to confuse intellectual inadequacy with unfamiliarity, particularly when people without formal philosophical training venture into established areas of enquiry.

Consider someone who begins asking questions about consciousness, free will or the relationship between mind and matter. They may use terminology imprecisely, overlook arguments developed centuries earlier or arrive at conclusions that specialists regard as elementary.

Does that make the exercise worthless?

I would suggest not.

There is considerable value in learning why an attractive argument fails. Indeed, much of education depends upon allowing people to entertain ideas that are subsequently corrected, refined or abandoned.

Of course, specialists are entitled to identify mistakes and explain why particular arguments have already been examined and found wanting. That is one of the considerable benefits of expertise.

What troubles me is the temptation to mistake familiarity with a literature for ownership of the questions it addresses.

An established philosopher may reasonably explain why an amateur's argument is mistaken. But simply identifying the author as an amateur, or describing their contribution as philosophical slop, tells us rather less.

There is also an asymmetry worth considering.

An obscure argument written by a recognised academic may be treated as intellectually demanding, while an equally obscure argument produced by someone outside the discipline may be dismissed as pretentious nonsense.

I am not suggesting these judgements are always wrong. Expertise matters, and disciplinary knowledge can reveal deficiencies that an inexperienced reader would overlook.

But it would be naive to suppose that judgements about intellectual quality are entirely independent of social standing.

Bourdieu (1984) examined how judgements of taste and cultural legitimacy are entangled with social distinction. His work does not establish that contemporary accusations of AI slop are motivated by professional snobbery, but it provides a useful framework for considering why some forms of expression acquire legitimacy while others struggle to be taken seriously.

I have considered related issues in The Privilege of Being Heard (Morrill, 2026c), where I questioned the tendency to confuse communicative performance with intellectual competence.

I do not intend to revisit the argument here. What interests me is the possibility that the label slop may sometimes serve as a means of policing intellectual boundaries without requiring the person employing it to defend those boundaries.

The accusation is sufficient.

Or at least it is expected to be.

Where the critics have a point

It would be rather foolish to defend slop without considering why the term has become popular.

There is a genuine problem with the increasing volume of generic, repetitive and poorly substantiated material circulating online. AI makes such material easier to produce, and the cumulative consequences may be considerable.

Møller et al. (2026) examined the effects of AI assistance in a controlled experiment involving 680 participants interacting within a simulated social media environment. Their findings indicated that different forms of AI intervention could affect participation and perceptions of discussion quality, with some assistance producing more activity while diminishing perceived authenticity.

This is important because the effect of AI cannot always be evaluated by examining individual contributions in isolation.

A single AI-assisted comment might be informative and entirely appropriate. Thousands of similar contributions may make an online discussion less useful, even when none is particularly objectionable in itself.

There is also evidence that AI assistance may improve individual performance while reducing diversity across a collection of outputs.

Doshi and Hauser (2024) investigated the effects of generative AI suggestions upon creative writing. Participants who received AI-generated ideas produced short stories that were evaluated more favourably, with particularly notable benefits for less creative writers. However, the resulting stories were also more similar to one another.

This presents an interesting difficulty.

What benefits the individual may not necessarily benefit the intellectual environment collectively.

I considered this problem in Beyond Manufacturing Consent (Morrill, 2026b), where I argued that the potential of AI to extend intellectual participation should not be confused with a guarantee of intellectual diversity.

The evidence does not justify claiming that generative AI will inevitably produce a homogeneous intellectual culture. Doshi and Hauser examined a particular creative task under experimental conditions, and we should be cautious about generalising beyond it.

Nevertheless, their findings give substance to concerns that AI-assisted production may produce a narrowing of expression even while improving individual outputs.

There is a further issue concerning the reliability of AI-mediated communication.

Sahebi and Formosa (2025) examined the problem of epistemic trust in an environment where increasingly convincing communications may have been produced or modified by AI. Their analysis identifies a dilemma. We have legitimate reasons to exercise caution, yet indiscriminate suspicion may also result in credible contributions being unfairly dismissed.

This seems particularly pertinent to the present discussion.

The problem is not simply that AI sometimes produces unreliable material. It is that the suspicion of AI involvement may itself become grounds for withholding credibility, even when the material is sound.

We therefore face two possible errors.

We may accept something because it appears authoritative when it is not.

Or we may dismiss something because it appears artificial when it is perfectly reasonable.

Neither is particularly desirable.

The solution cannot simply be to abandon scepticism. Nor can it be to treat every AI-assisted contribution as suspect.

We need a rather more discriminating approach.

Why is dismissal so attractive?

I suspect the popularity of slop tells us something about the psychology of criticism.

One possibility is cognitive economy.

Evaluating an argument requires effort. We need to understand what is being claimed, examine the evidence and consider whether the conclusion follows. Recognising familiar characteristics and assigning something to a category is considerably easier.

If a passage contains expressions commonly associated with AI-generated prose, we may conclude that it lacks originality before examining its content.

Sometimes that judgement may be correct.

But recognising a style is not the same as evaluating an argument.

There may also be a social component.

Declaring something slop allows the critic to demonstrate discernment. It establishes, or at least implies, a distinction between those capable of recognising intellectual inadequacy and those unfortunate enough to produce it.

I wonder whether this is particularly attractive in professional environments where authority has traditionally depended upon recognised expertise, credentials and command of an accepted register.

Artificial intelligence complicates those arrangements by making forms of expression once associated with specialist training increasingly accessible.

In Taking the Middle Road (Morrill, 2026d), I considered how responses to AI may become entangled with professional identity and the prospect of losing abilities that once distinguished particular groups.

It would be tempting to interpret the popularity of slop as another expression of this anxiety.

But that would be too convenient.

It would also risk reproducing the very behaviour I am criticising.

Not everyone who objects to AI-generated material is protecting an established position. Some have perfectly reasonable concerns about misinformation, intellectual dishonesty, the erosion of educational practices and the burden of separating credible material from nonsense.

We should not dismiss those concerns as gatekeeping merely because they complicate a more optimistic view of AI.

Nor should we assume that expertise is simply a mechanism of exclusion. Much of its value lies in providing the knowledge necessary to recognise errors and distinguish superficial plausibility from genuine understanding.

The more interesting question is what happens when expertise is expressed principally through dismissal rather than explanation.

At that point, authority risks becoming a substitute for argument.

And I find that rather troubling.

The problem with imperfect ideas

There is something else worth defending here, and it extends well beyond artificial intelligence.

Not every worthwhile intellectual contribution begins as a polished argument.

Ideas frequently develop through tentative observations, misunderstandings, poorly expressed questions and exchanges with people who possess different kinds of knowledge.

Much of this process is intellectually untidy.

I suspect we would be in considerable difficulty if every contribution were required to satisfy the conventions of an established discipline before it could be considered.

A student asking a naive question may expose an assumption that specialists have ceased to examine. Someone without formal philosophical training may rediscover a familiar argument, but the process of learning why it fails can still be valuable. An individual using AI to investigate an unfamiliar scientific problem may begin with substantial misconceptions and gradually develop a more sophisticated understanding.

None of these examples guarantees a worthwhile result.

But the value of intellectual activity is not always adequately captured by examining the quality of its earliest products.

There is a difference between a poor argument that is part of developing understanding and one that has been produced without any apparent concern for accuracy or improvement.

The first may be a necessary feature of learning.

The second is rather more deserving of criticism.

Even here, however, we should exercise caution. We cannot reliably infer someone's intentions, competence or willingness to learn from a short passage of writing.

I wonder whether the increasingly hostile language surrounding intellectual inadequacy might discourage some people from expressing ideas that are not yet fully developed.

I have no empirical evidence demonstrating that this is happening as a consequence of the term slop. It remains a possibility rather than an established finding.

But it is worth considering.

We frequently argue that education should encourage curiosity, experimentation and the willingness to make mistakes.

It would be unfortunate if the wider intellectual environment became increasingly hostile to precisely those activities.

The slop of anti-slop

I began by suggesting that slop has become a convenient way of dismissing things we dislike.

Having considered the matter, I think the position requires some qualification.

There is a genuine problem with low-quality material being produced at considerable scale. AI has made this easier, and there is evidence that its consequences for communication and creativity are not uniformly beneficial.

Criticism of these developments is justified.

What concerns me is the extension of that criticism into a general judgement about intellectual worth, particularly when the judgement rests upon the origins, style or perceived legitimacy of an argument rather than its substance.

I am not suggesting that every contribution deserves equal attention. Nor do I expect anyone to provide an elaborate critique of every dreadful LinkedIn post they encounter.

Sometimes we simply do not wish to read something. That is perfectly reasonable.

But there is a difference between deciding that something is not worth our attention and claiming to have established that it possesses no intellectual merit.

I would hope that expertise gives us the ability to identify deficiencies, not merely the authority to announce them.

There is something rather dispiriting about watching people who regard themselves as defenders of intellectual standards resort to a term that frequently communicates little beyond their own disapproval.

And I cannot help wondering whether the increasing popularity of slop is beginning to produce a peculiar problem of its own.

A growing body of repetitive, dismissive and occasionally self-congratulatory commentary that contributes relatively little to our understanding of the material being condemned.

The slop of anti-slop.

I do not wish to defend intellectual rubbish. Nor do I think the indiscriminate production of artificial material deserves celebration simply because it makes publishing easier.

I would, however, defend the right to produce imperfect ideas, to develop arguments without first obtaining the approval of established intellectual authorities, and to use whatever legitimate assistance makes those activities possible.

Some of the resulting work will undoubtedly be dreadful.

Some of it may be rather good.

I would prefer that we retain the intellectual curiosity necessary to tell the difference.

And if someone wishes to call it slop, I would be interested to hear why.

Otherwise, they may simply be adding to the pile.

References

Bourdieu, P. (1984). Distinction: A social critique of the judgement of taste (R. Nice, Trans.). Harvard University Press. (Original work published 1979)

Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), Article eadn5290. https://doi.org/10.1126/sciadv.adn5290

LinkedIn. (2026a, June 4). Keeping conversations real on LinkedIn. https://news.linkedin.com/2026/keeping-conversations-real-on-linkedin

LinkedIn. (2026b, September 15). How LinkedIn is continuing to tackle AI slop. https://news.linkedin.com/2026/how-linkedin-is-tackling-ai-slop

Messick, S. (1995). Validity of psychological assessment: Validation of inferences from persons' responses and performances as scientific inquiry into score meaning. American Psychologist, 50(9), 741–749. https://doi.org/10.1037/0003-066X.50.9.741

Møller, A. G., Romero, D. M., Jurgens, D., & Aiello, L. M. (2026). The impact of generative AI on social media: An experimental study. Scientific Reports, 16, Article 9376. https://doi.org/10.1038/s41598-026-40110-8

Morrill, P. (2026a, September 19). Academic language was flattening before AI arrived. https://paulmorrill.com/blog/2026-09-academic-language-flattening-before-ai

Morrill, P. (2026b, October 25). Beyond manufacturing consent: Artificial intelligence, intellectual agency and the democratisation of thought. https://paulmorrill.com/blog/2026-10-beyond-manufacturing-consent

Morrill, P. (2026c, October 21). The privilege of being heard: Oracy, intellectual authority and the democratisation of thought in the age of AI. https://paulmorrill.com/blog/2026-10-the-privilege-of-being-heard

Morrill, P. (2026d, October 24). Taking the middle road: Existential angst in the age of AI. https://paulmorrill.com/blog/2026-10-taking-the-middle-road

Sahebi, S., & Formosa, P. (2025). The AI-mediated communication dilemma: Epistemic trust, social media, and the challenge of generative artificial intelligence. Synthese, 205, Article 128. https://doi.org/10.1007/s11229-025-04963-2