Introduction to Slop Studies (simulacro.co.uk)

This piece introduces "AI slop"—a broad, coined term for low-quality, inauthentic or indiscriminately distributed AI-generated content—and synthesizes a Master’s dissertation that maps the problem both conceptually and empirically. The author builds a four‑dimension framework (themes, types, qualities, metaphors) from qualitative and quantitative analyses of sources cited on the Wikipedia page, produces the SlopNews dataset and a public archive (slopscooper), and reports that slop news is statistically less complex, less varied, and carries more positive sentiment than non‑slop news. The writeup traces the term’s rise (usage jumped ~334% in 2024), situates it among related phenomena (hallucination, misinformation, spam, “bullshit”), and ties its modern expansion to cheap, scalable generation after tools like ChatGPT lowered production costs. Significance for AI/ML: slop isn’t just a technical artifact to detect—because it’s often not strictly false or malicious, traditional fact‑checking and hallucination mitigation miss much of it. Instead slop represents a structural threat to information ecosystems (polluting open‑source repos, academic submissions, publishing and moderation pipelines) driven by incentive systems—algorithmic visibility, engagement monetization and quantity‑over‑quality workflows. The work argues for broader responses: measurement (datasets like SlopNews), institutional triage, and socio‑technical interventions that address platform incentives and epistemic harm, while framing slop conceptually as a form of “bullshitting” that erodes shared reality rather than a narrowly defined AI risk.