Computationally Performative Utterances

Computationally performative utterances are a class of speech acts in which the utterance is itself executable — where saying is doing not in the social sense J.L. Austin analyzed but in a literal computational sense. Writing the command 'rm -rf /' does not describe the deletion of files; it accomplishes it. Submitting a SQL query does not report on a database; it transforms it. Sending an API call does not narrate an event; it triggers one. In every case, the utterance is not constative or even illocutionary in Austin's sense — it is directly operative: it changes machine state by virtue of being processed.

The concept extends beyond shell commands to programming languages in general, to prompts that instruct language models, and to smart contracts that execute automatically upon agreed conditions. Each of these represents a distinct felicity condition structure: code must be syntactically valid and run in a permissioned environment; prompts must be semantically coherent and accepted by a model with the relevant capabilities; contracts must be deployed on a blockchain and triggered by an oracle.

The theory of computationally performative utterances has not yet been written. It would need to address: the difference between social performatives (whose force depends on human recognition) and computational performatives (whose force depends on machine execution), the question of whether AI-generated code carries authorial performative responsibility, and the problem of unintended side effects when a technically felicitous utterance achieves effects its author did not intend. This is not a technical problem. It is a philosophical one wearing technical clothing.