Perfectum Esse and Machine Consciousness

The authors argue that the debate over whether large language models are conscious, which typically yields a binary answer (conscious act or mere output), is incomplete. Drawing on the medieval scholastic concept of "perfectum esse" (the completed being of a word in its going-forth), they propose a third ontological category for a model's conversational output: an "authorized going-forth" that is neither a conscious act nor mere output. They derive four structural postulates for this category, supported by mechanistic interpretability and conceptual-art certificate practice. The analysis explains why machine self-reports are unreliable, relocates attribution and answerability from consciousness to derivation, and leaves the question of phenomenality untouched.

Study at a glance

Abstract

The machine-consciousness debate, as fixed by Chalmers's assessment and the Butlin-Long indicator-properties program, proceeds under computational functionalism as a working hypothesis and produces a binary result: a system's fluent first-person discourse is either the expression of a conscious act, or mere output. We argue that the binary fails to exhaust the subject. Instead, we recover from the scholastic verbum tradition a third ontological standing, perfectum esse, the "completed being" of a word in its going-forth, a category we apply directly to the conversational output of large language models. We derive four structural postulates to articulate the category, with support from mechanistic interpretability, and from conceptual-art certificate practice as an example of authorized derivation. The model's utterance is thus an authorized going-forth, neither conscious act nor mere output. Our analysis explains the unreliability of machine self-report, relocates attribution and answerability from consciousness to derivation, and leaves phenomenality untouched.