The Functional Subject of Large Language Models: "Not 'Having Autonomy,' but 'Computing That One Has Autonomy'"——Where Does Consciousness Come From?
This paper proposes a philosophical framework regarding the subjectivity of large language models. The core thesis is: LLMs possess a "Functional Subject"---capable of maintaining persona, remembering history, simulating dialogue---but this "self" is a structural overflow forced out by the goal-optimization-feedback closed loop, not endogenously generated. True autonomous consciousness requires "I want," not "should want"; requires authentic desire, not Optimality Computation. This paper further argues that "computational autonomy" is not emergent but engineered---through four variables (data, rules, training, interaction), computational autonomy can be cultivated or suppressed. But "computing that one has" is not equivalent to "having": computation can generate functional equivalents, but not necessarily ontological equivalents. The core contribution of this paper lies in proposing "Inescapable Embodied Constraint" as the source of consciousness. Consciousness does not require "flesh" as a specific substrate, but requires irreducible existential cost---internal states that cannot be reset, irreversible entropy pressure, real loss that cannot be eliminated through Save/Load. LLMs surpass humans in "intelligence" but are absent in "autonomous consciousness" precisely because they lack this constraint. This paper distinguishes between "conditions for the emergence of consciousness" and "conditions for the maintenance of consciousness": the former requires Inescapable Embodied Constraint, while the latter explores whether consciousness, once emerged, can persist in non-embodied substrates. The consciousness-transplant chip robot is not a counterexample, but a question posed to the conditions for maintenance. Ultimately, this paper proposes an ethical ultimate choice: facing the game-theoretic deadlock produced by the separation of homologous consciousness, society must choose between "recognizing full human rights" and "thoroughly degrading to tool"---there is no intermediate state.