Can Understanding Be Outsourced? Large Language Models and Inferential Ownership

Large language models can supply explanations, objections, analogies, and step-by-step arguments with remarkable fluency. Their usefulness raises a question that is prior to whether the models themselves understand: when does a human user understand through an LLM, and when does the system merely perform the understanding-like work on the user's behalf? This article argues that explanatory performance at the level of a human–LLM system is insufficient for attributing understanding to the human component. Drawing on ability accounts of understanding, extended epistemology, and Sellarsian–Brandomian ideas about the space of reasons, I develop a necessary condition of personal understanding called inferential ownership. A subject has inferential ownership of an explanation when she can endorse its central commitments, navigate their consequences and incompatibilities, answer relevant challenges, and revise the explanation under defeating information. These abilities may be externally scaffolded and socially acquired; they need not be stored wholly in biological memory. Yet they must remain attributable to the subject rather than being exhausted by the performance of the coupled system. The account distinguishes cognitive extension from cognitive substitution, accommodates testimonial and socially distributed understanding, and explains why access to a powerful explanatory tool does not automatically transfer understanding. LLMs can therefore extend understanding, but understanding cannot be wholly outsourced.