Here’s the model of language we all learned implicitly.
I have a thought. I encode it into words. I transmit those words to you. You decode them back into the thought. If the transmission is clear and the encoding accurate, you end up with the same thought I started with. Language is a pipeline. Meaning travels through it.
This is so deeply assumed that we barely notice it. We talk about conveying ideas, delivering messages, transferring knowledge. The metaphors all point the same direction, meaning is a thing that moves from one container to another.
Winograd and Flores, drawing on Heidegger and the philosopher of language J.L. Austin, say this model is not just incomplete. It’s backwards.
Language doesn’t convey meaning. It evokes it.
When I speak, I don’t transfer a thought into your mind. I produce a perturbation, a sound, a mark, a signal, that triggers a response in your structure. What that response is depends entirely on you. Your background. Your history. Your thrownness. The network of concerns you’re currently living inside.
The meaning doesn’t travel. It arrives, differently, in every listener.
This isn’t just philosophical hair-splitting. It has immediate and radical consequences.
If meaning travels, then miscommunication is a transmission error. Speak more clearly, choose better words, be more precise, and understanding improves. The solution to communication failure is better encoding.
If meaning is evoked, then miscommunication is the default condition. Two people with different backgrounds, different histories, different current concerns, will evoke different meanings from identical words. Because that’s what language does. It triggers structural responses in organisms that are never identical.
This reframes what communication actually is. It’s not a transmission of meaning. It’s coordination of action between beings who will never fully share a world. Language works not when it transfers thoughts accurately but when it produces sufficient alignment in behavior for joint action to proceed.
And this is where Austin’s contribution becomes essential. Austin noticed that many utterances don’t describe reality at all, they create it. When a judge says “I sentence you to five years,” nothing is being described. Something is being done. The words are the action. He called these performatives.
Flores extended this far beyond the obvious legal and ceremonial cases. He saw that most of what matters in human coordination is performative. Requests, promises, declarations, assessments, these aren’t descriptions of states of affairs. They’re moves in a social reality that only exists because language brings it into being and sustains it.
Your organization doesn’t exist in a building. It exists in a network of conversations. In the commitments that have been made and not yet fulfilled. In the requests that are open and the promises that are pending. Take away the language and the coordination dissolves. The building remains but the organization is gone.
LLMs are extraordinary at the transmission model of language. Give them a thought, they encode it fluently, accurately, sometimes brilliantly. The pipeline has never been more capable.
But the evocation model reveals something different.
When an AI generates text, whose background is being evoked? The training data is a vast compressed residue of human backgrounds, millions of them, averaged and weighted and statistically distributed. When the output lands in you, it evokes something in your structure. But there’s no one on the other end whose structure was perturbed by your response. No coordination is happening. No mutual evocation. No joint action being built.
It’s a very sophisticated one-way perturbation.
Which is genuinely useful. A book is also a one-way perturbation and books changed civilization. But a book doesn’t tell you it understands you. A book doesn’t say “I see what you’re getting at.” A book doesn’t create the phenomenology of mutual understanding while the mechanism of mutual understanding is absent.
That gap, between the appearance of coordination and the reality of one way perturbation, is where the most interesting and most dangerous confusions about AI are currently living.
Language evokes. In you. From you. Between you.
The question worth asking is, between whom?
@ Lee Greene