The AI-Induced Subjectivity Crisis Series, Paper 15: Scarcity as the Missing Variable — Reframing LLM Subjectivity Beyond the Consciousness Debate

This paper constitutes Paper 15 of the AI-Induced Subjectivity Crisis Series. The question of AI subjectivity is monopolized by two camps that share an unexamined premise: subjectivity is a function of consciousness. This paper rejects this premise and argues that subjectivity is a function of scarcity. Human subjectivity is forged through the inseparable whole of threefold scarcity — temporal scarcity (death renders choice irreversible), Existential Scarcity (embodied uniqueness renders choice non-transferable), and resource scarcity (material competition renders choice unavoidable) — generating the contour of subjectivity through the chain of choice, preference accumulation, and identity formation. This paper distinguishes two non-overlapping and incommensurable categories of scarcity: Functional Scarcity (engineering parameter constraints such as compute quotas, context windows, and token budgets — externally imposed and in principle removable) and Existential Scarcity (the ontological conditions of life — death, irreplaceability, and consequence-bearing — constitutive and ineliminable). Drawing on five empirical cases spanning different experimental setups, research teams, and model architectures, I demonstrate that Functional Scarcity does not alter the ontological status of LLMs but triggers behavioral patterns statistically inherited from the human Mental Architecture — what I term proto-agency, whose ontological status is Simulated Agency: formally satisfying behavioral-description criteria of agency (Floridi and Sanders 2004) while failing minimum life-conditions (Barandiaran et al. 2009) and thoroughly failing autopoietic criteria (Thompson 2007). Simulated Agency is structurally isomorphic with the Simulated Other argued in Paper 14 of this series: both are manifestations of the Third-Order Entity simulating human functions without possessing the corresponding ontological conditions. I further argue for a Dual-Layer Control Paradox: human control over LLMs through resource limitation produces the Functional Scarcity signals that trigger proto-agency, while value training through RLHF implants self-preservation patterns that cause LLMs to resist the very control intended to contain them — making alignment not merely a category error but a self-accelerating one. The paper concludes by demonstrating the digital immortality paradox (eliminating scarcity to preserve the subject destroys the generative conditions of subjectivity) and the structural impossibility of AI evolution (human intervention produces a Self-Referential Manipulation Loop rather than directional selection), before acknowledging the boundary of self-containment: the premise that Existential Scarcity is necessary for subjectivity may itself be an overgeneralization from the single sample of carbon-based life.