Philosophy of Mind
Belief, understanding and representation, tested on systems that produce language without a life.
For most of its history, philosophy of mind had one case study: us. Systems that produce fluent language give it a second, and second cases are where concepts show their seams.
My way into the field is through epistemology. Whether a language model has belief-like states is now partly an empirical question. Herrmann and Levinstein propose standards an internal representation must meet to count as belief-like — accuracy, coherence, uniformity across domains, and use, which requires that the representation actually govern what the model outputs. Use is the standard that matters for whether a system can be a source of knowledge, and it is the least established.
Understanding is a separate question. On an externalist view like Williamson’s, to understand an expression is to participate with sufficient fluency in the practice of a linguistic community, and on that view whether these systems understand what they produce is a serious and open question. It is also less decisive than it looks: understanding is compatible with every epistemic failing the knowledge norm of assertion marks. A human bullshitter understands his sentences perfectly.