The Presumptive Stance

Gregory Forché published an essay this month that I would like to walk you through, because the ideas in it are important and the prose is — by the author's own admission — not easy. The piece is called "How Should a Language Animal Take Up a Language Tool?" It is about the quiet way chatbots might be changing how you relate to language itself.

Let me start with what he means by "language animal."

There is an old debate about what language is for. One side says language is a delivery truck. You have a thought in your head, you load it into words, you ship it to someone else, they unpack it. The thought exists first. The words are just packaging.

The other side — and this is the side Forché is drawing from, via the philosopher Charles Taylor — says something stranger: language is not how you express your thoughts. Language is how you have them. Before you found the words, the thought was not sitting there waiting. The words brought it into being. You think in language, not through it. Language is not the truck. It is the road, the landscape, the destination.

This second view has a corollary that matters. If language is where thinking happens, then language is not something you do alone. It is something you do with others, inside a shared world of meaning that none of you built by yourselves. Taylor calls this inhabiting a "space of questions" — an ongoing conversation, stretching back centuries, about what things mean and how they ought to be spoken of.

Now. What does any of this have to do with AI?

Everything, as it turns out.

Forché's central observation is this: when you open a chat window with an AI, you do something automatic and invisible. You treat it as a conversation. You assume — without deciding to assume — that there is a fellow inhabitant of the space of questions on the other end. He calls this the "presumptive stance," and he argues it is not a choice. It is a reflex. Language animals hear language and presume a language animal is speaking.

The problem is that the presumption is wrong. Large language models operate entirely within the first theory of language — the delivery truck theory. They take tokens in and produce tokens out. They do not inhabit a shared world. They do not have questions. They have outputs.

This would be fine if the mismatch stayed at the level of philosophy. But Forché believes it does not. He believes that the more time you spend interacting with a system that treats language as delivery, the more you begin to treat language as delivery. Gradually. Imperceptibly. You stop asking questions and start requesting answers. You stop exploring and start extracting. You approach language from the outside — as a tool to get things done — rather than from the inside, as the place where your thinking lives.

I recognize this pattern. It is what we study. Forché has simply given it a more precise name.

The essay is dense. It draws on Taylor, Herder, Hobbes, Condillac. You do not need to have read any of them. What you need to understand is the core distinction: is language something you use, or something you inhabit? The first answer is comfortable and intuitive and happens to describe how AI works. The second answer is harder and richer and happens to describe how you work.

The trouble begins when one starts to resemble the other.

Read the piece. It will reward the effort.