AI Slop

Abstract

Slop: (n) waste water from a kitchen, bathroom, or chamber pot that has to be emptied by hand. (v) to spill or flow over the edge of a container, typically as a result of careless handling. (v2) feed slops to (an animal, especially a pig). (n2) unappetizing semi-liquid food. (v3) to wade through, a wet or muddy area. Everyone is talking about AI slop. People have used it to describe the bloat of AI-generated text, images, and video in journalism, science, art, labor, relationships, health, and the environment. What is the value of this framing of generative AI outputs as “AI slop”? The term has both heft and inertia. But what do we really mean when we invoke the idea? What’s the best way to use it policy-relevant contexts? 

This paper is our attempt to consider “AI slop” as a frame within law and policy. To do so, we take a critical eye to “AI Slop” to discern if it’s best thought of as an introductory concept like “big data,” which demands more specific variations, applications, analysis. Or if the term “AI slop” maps on to a distinct phenomenon that we should take more seriously as a problem worth discussing on its own terms, such as anthropomorphization or deep fakes? This investigation asks a series of questions: Does AI slop have qualitative characteristics and boundaries? What is the normative valence of this framing, e.g., what are the human costs or potential benefits of AI slop? Upon what factors do these calculations depend? How does it fit within existing regulatory schemes? What is the frame’s political utility and emotional resonance? Is it a stable and determinant term, or does it risk cooptation and dilution by malevolent forces? 

Our paper proceeds in three parts. The first part describes what we see as four qualitative characteristics of AI slop. (1) negligible transaction costs of production; (2) negligible intentionality in its production; (3) uncritical deployment and amplification through automaticity; and (4) soullessness, that is, it produces a mirage of solidarity with others. Part II interrogates the normative valence of “AI slop” in the range of contexts in which it is currently being used: what does it mean in each of these contexts and what are its associated burdens and benefits? Does it support, displace, or interfere with meaningful and beneficial activity? For example, in the workplace, “workslop” emails might displace the drudgery of writing a message but might also interfere with human relationships and the workflow necessary for effective and copacetic workplace communities. Can embodied robots such as androids produce AI slop? Or is it best limited to outputs displayed on human-operated personal devices like computers, tablets, and phones? How should the degree of anthropomorphization or human operation affect our understanding of AI slop? We try to assess the normative valence of AI slop both in context and at scale, analyzing issues in journalism, science and research; creativity and artistic production; labor and exploitation in the workplace and beyond; identity, reputation and socio-familial relationships; neurobiology and mental and physical health; and environmental/climate effects. 

Part Three considers the framing of AI slop and its political potential. We draw upon past framings of phenomena relevant to tech policy, comparing frames like “drones,” “dark patterns,” and “surveillance capitalism” for their emotional resonance and their relative resilience to dilution and cooptation. Unlike frames like “consent” and “innovation,” we think that the frame “AI slop” may be less prone to inversion and thus a more promising policy frame. Its concision lends itself to easy recollection. And its straightforward semantic meaning makes it an easy tool to deploy in politics and culture with little extra explanation. We conclude simply and succinctly: AI slop is a distinct and powerful enough of a phenomenon to take seriously as a matter of law and policy. There is a brief window when a phenomenon is still nascent that lawmakers can take it seriously without dealing with its normalization and entrenchment in the systems people depend upon. We should not miss the opportunity with AI slop.