Grammatization
In The You On AI Field Guide
Before alphabetic writing, language was continuous sound inseparable from the speaker's body, situation, and gesture. The alphabet broke this flow into discrete letters, combinable by rule, reproducing the sound independently of the speaker. The grammatization made language storable, transmissible, and accumulable across generations — without which there is no science, no law, no philosophy. The loss was the dimension that could not be captured: prosody, rhythm, the embodied accompaniment giving an utterance its full meaning.
Each subsequent stage extended the operation into new domains. Musical notation made composition possible independent of live transmission; industrial procedures made production possible independent of craft knowledge; data analytics made prediction possible independent of lived understanding of the predicted. Each stage produced the same pattern: gain in extension, loss in the ungrammatizable remainder, pharmacological inseparability of the two.
AI's grammatization is different in scope. The large language model takes the entire corpus of human textual production, breaks it into tokens, analyzes statistical relationships, and generates new sequences possessing the form of human thought without having been produced through the process of thinking. This is the grammatization of the meta-capacity through which all other grammatizations were performed — recursive in a way no previous stage achieved.
You On AI documents the phenomenology. Working with Claude restructures cognitive activity into the prompt-response cycle. The continuous flow of thinking — gradual development through reflection and revision — is reorganized into discrete exchanges. The thinker adapts to the system's tempo, just as the factory worker adapted to the machine's rhythm. The gain is extended capability; the loss is the dimension of thinking the prompt-response cycle cannot capture — the slow, associative, embodied quality of thought developing at its own pace.
Origin
Stiegler adapted the concept from linguist Sylvain Auroux's La Révolution technologique de la grammatisation (1994), which traced the historical grammatization of natural languages through dictionaries and grammars.
Stiegler generalized the operation beyond linguistics to encompass every externalization of continuous human activity into discrete technical elements, developing the framework across Technics and Time, Symbolic Misery, and Automatic Society.
Key Ideas
Continuous to discrete. The operation reduces flowing activity to reproducible units — always a gain and always a loss.
Every epoch is grammatizing. The alphabet, notation, industrial procedure, analytics, AI — each extends the operation into new domains.
AI grammatizes cognition itself. Previous stages captured specific modalities; AI captures the meta-capacity, recursively.
The ungrammatizable remainder. What cannot be captured in discrete form is precisely what tends to be lost — embodied, tacit, relational dimensions.
Debates & Critiques
The contemporary debate is whether AI's grammatization is genuinely total or whether significant dimensions of cognition resist tokenization. Critics of Stiegler's framing argue that embodied, affective, and relational cognition remain structurally outside what language models can capture. Stiegler's heirs respond that the question is not whether a remainder exists but whether institutional practices will protect it.
Further Reading
- Bernard Stiegler, Symbolic Misery, Volume 1: The Hyperindustrial Epoch (2004)
- Sylvain Auroux, La Révolution technologique de la grammatisation (1994)
- Anne Alombert, "Reticulated Artificial Intelligence" (2024)
- Stiegler, "Artificial Stupidity and Artificial Intelligence in the Anthropocene" (Shanghai lecture, 2018)