Skip to content

Why Slop Matters

Kommers et al. (2026) — The Alan Turing Institute / Purdue / Duke / Chicago / Cornell, submitted to ACM

Core Argument

The authors argue that AI-generated "slop" — low-quality content that exhibits superficial competence without underlying substance — should not be dismissed as mere digital pollution but taken seriously as a subject of study. They make two claims: first, that slop serves a social function by offering a supply-side solution to demand for content that is economically infeasible for humans to produce; second, that slop has aesthetic value, following the historical pattern where "low" cultural forms (kitsch, camp, pastiche) were initially dismissed by critics but later recognised as legitimate.

Three Prototypical Properties of Slop

The paper identifies three "family resemblance" features that characterise most AI slop:

  1. Superficial competence: the output demonstrates a kind of competence that would require real expertise to achieve without AI, but this veneer of quality is belied by a lack of underlying substance, craft, or communicative intent. AI-generated work memos have good grammar; AI-generated images are photorealistic. The content is characterised by overtly "idealised" output that flattens individual variation into canonical patterns.

  2. Asymmetric effort: slop is generated with a facile prompt and requires little effort of the kind that would be necessary to create such output without AI. The cost-benefit calculus of content production is fundamentally altered when generation cost approaches zero.

  3. Mass producibility: slop is designed to work within a digital ecosystem of mass production and distribution. The outputs may be personalised, but they are crucially mass produce-able.

Three Dimensions of Variation

While slop shares prototypical features, it varies across three dimensions:

  • Instrumental utility: some slop is generated with a specific purpose (e.g., workplace deliverables), even if the result is sloppy
  • Personalisation: some slop resembles a specific person's voice or likeness
  • Surrealism: some slop has a heightened, intensified quality — related to the "hallucinatory" quality of AI systems that can confabulate plausible material and then escalate it to absurdity

Relevance to This Wiki

For safety-critical applications, the concept of slop maps directly to output-vacuity — the failure mode where LLM output is technically unobjectionable but operationally empty. The "superficial competence" property is precisely the mechanism that makes vacuous AI advisory output dangerous: it passes surface-level quality checks while providing no actionable insight. The "asymmetric effort" property explains why slop proliferates in professional settings — when generating a plausible-sounding assessment costs almost nothing, the incentive to critically evaluate whether it actually says anything diminishes. This connects to automation-bias and trust-calibration: users may accept slop not because they over-trust its content, but because the effort to distinguish substance from surface is not worth the cognitive cost when content is cheap. The paper's framing of slop as a cultural and economic phenomenon rather than purely a technical deficiency reinforces that mitigation requires system design (structured outputs, specificity metrics) rather than just model improvement. The connection to monoculture-collapse is also notable: if AI-generated content dominates information ecosystems, the flattening of individual variation into canonical patterns reduces the diversity of perspectives available to decision-makers.