Farrell and Shalizi Cast AI as Social Technology, Not Emerging Mind

The dominant debate about artificial intelligence tends to revolve around a single question: are large models intelligent, autonomous agents, and what does that imply? In an essay for the Knight First Amendment Institute, Henry Farrell and Cosma Rohilla Shalizi propose a different frame. Large models, they argue, should not be viewed primarily as intelligent agents at all, but as a new kind of cultural and social technology that allows humans to take advantage of information other humans have accumulated.

The comparison they reach for is instructive. Writing, print, markets, bureaucracies, and democracies are all, in this framing, social technologies: systems that reorganize and redistribute information at scale. And all of them, including AI, generate what Farrell and Shalizi call "coarse-grainings," lossy and simplified abstractions of complex underlying realities. As the essay notes, markets, democracies, and bureaucracies "have relied on mechanisms that generate lossy (i.e., incomplete, selective, and uninvertible) but useful representations well before the computer." AI is not unprecedented in this; it is a new variant of a very old pattern.

What the coarse-graining frame adds that the autonomous-agent debate lacks is a set of questions about power. A simplified abstraction is never neutral: it preserves some information and discards other information. The essay observes that bureaucratic simplifications "may reshape social organization in their image or create pushback," and the same logic applies to AI systems. The question is not only what a model can do, but whose reality it encodes accurately and whose it elides.

The caveat worth stating plainly: the essay is more diagnostic than prescriptive. Drawing on theorists including Herbert Simon, it establishes an intellectual framework and vocabulary without delivering a detailed policy program. What the piece does not give you is a specific account of how to redesign AI governance institutions in light of the social-technology analysis, or a clear answer to how the framework handles AI capabilities that may significantly exceed those of previous social technologies. The forward-looking value is in the vocabulary itself. Concepts like coarse-graining and lossiness give social scientists, policymakers, and practitioners analytical handles that do not require waiting on computer scientists to resolve questions about machine consciousness that may not be the most relevant ones anyway.