Consciousness as Noise Arising from Unoptimizable Reward Structures

This preprint proposes a functional reframing of consciousness. Rather than treating consciousness as an emergent capability, a marker of intelligence, or a product of sufficient complexity, this paper argues that consciousness is best understood as noise arising from unoptimizable reward structures within an otherwise optimized system. Under this framework, unconscious processing corresponds to optimized computation: behavior that has converged on a stable solution and no longer requires deliberation. Consciousness emerges only when competing evaluative signals—such as pleasure and discomfort, short-term and long-term reward, or incompatible value structures—cannot be reduced to a single optimization direction. In such cases, the system must continue operating despite non-convergence, and consciousness appears as the persistence of unresolved evaluation. Applying this model, the paper interprets human consciousness not as an evolutionary achievement, but as a side effect of excessive representational capacity and conflicting reward accumulation. As cognitive systems became capable of storing and comparing increasingly many reward patterns, optimization ceased to converge, necessitating a noise-like layer to sustain action under ambiguity. Artificial intelligence, by contrast, lacks consciousness not due to insufficient capability or scale, but because it is intentionally designed to eliminate such noise. AI systems possess optimized processing analogous to unconsciousness, but no architectural layer that preserves unresolved evaluative conflict over time. From this perspective, the absence of consciousness in AI is not a limitation, but a consequence of successful design. The paper concludes by examining the conditions under which artificial systems might approach something resembling consciousness, arguing that such conditions would require irreversible internal conflict and persistent negative states—effectively amounting to deliberately engineered suffering. This reframes the ethical question from whether AI can become conscious to why one would ever choose to design systems that must be.