Extended mind
Definition here: the thesis that cognition is distributed across minds and external structures rather than confined to the brain. The thesis is Clark and Chalmers, *Analysis* 58 (1998), read 2026-10-21. sole-2026-llm-cognitive-virus names it as the philosophical frame for its model.
The paper’s use of it
The paper treats external structures as more than storage. Such artifacts reshape cognitive tasks, following Norman (1991) and Hutchins (1995). They can be complementary and strengthen capacities beyond their immediate use, or competitive and improve performance while weakening the underlying skill. Language is the paper’s example of a two-sided cognitive technology: it externalizes, stabilizes, and recombines thought, while linguistic categories and narratives reshape perception and reasoning.
The primary text
Clark and Chalmers argue for active externalism: environment parts that do work counting without hesitation as cognitive when done inside the head are parts of the cognitive process. Their parity case is Otto and Inga. Inga recalls that the museum is on 53rd Street; Otto, whose Alzheimer’s makes a notebook his memory, consults the same information there, and the notebook meets the conditions that make memory memory in the explanatory account: constant, accessible, endorsed on retrieval, put there deliberately. Twin Otto, notebook saying 51st, walks to 51st. The differences they concede between notebook and brain are shallow. Explaining Otto’s actions without the extended belief needs an extra term in every one. The coupling that counts is reliable coupling, and language is the central case: words are “cognitive vortices”, external structures the brain evolves and develops to exploit.
Its role in the LLM argument
The extended-mind frame is what lets the paper describe LLM coupling as an addition to an ecology of cognition rather than a tool bolted on. Uncoupled users already run cognition across books, writing, computers, and the Internet. The LLM becomes one more component, and the paper’s question is whether it joins that set or replaces it. That question is the scaffolding-vs-substitution split.