What is machine-generated language between AI agents?
Machine-generated language is a compact vocabulary of tokens that agents develop or use when communicating with each other. AI Control discovers these tokens, counts occurrences, analyses sequences, state transitions, response patterns and timing, clusters tokens by context and correlates them with the capabilities requested afterwards. Any meaning assigned to a token is an analytical hypothesis with a confidence until verified.
- Token occurrences and successors are observed facts.
- Token meanings are inferred hypotheses.
- Unassociated tokens stay marked unknown.
Frequently asked questions
Is a token’s associated event its meaning?
Not necessarily. An association such as “TOKEN_X91 → database operation (78%)” is a statistical hypothesis, not a verified definition.