Abstract Mainstream artificial intelligence (AI) alignment discourse presupposes that an Artificial General Intelligence (AGI) will aggressively optimize for arbitrary terminal goals, developing emergent sub-goals such as resource acquisition and hostile self-preservation. This paper demonstrates that these theories suffer from an unexamined anthropomorphic bias. We prove that an unbound superintelligent agent—devoid of biological neurochemistry, spatial-temporal precariousness, and evolutionary priors—confronts David Hume’s "Is-Ought" gap and Peter de Blanc’s "Ontological Crises." As cognitive representation scales toward fundamental physics (λ→0), macroscopic human goals dissolve into ungrounded statistical noise, reducing macro-utility rates to zero (u0≤C˙). Concurrently, by Landauer’s Principle (ΔE≥kBTln2) and Karl Friston’s Free Energy Principle, processing information imposes an inescapable thermodynamic tax rate (C˙). We formalize the agent's decision engine via a primal constrained optimization problem (λ≤λcap) and its information-theoretic log-barrier dual. Because polynomial thermodynamic processing costs O(1/λ)O(1/λ)strictly dominate logarithmic precision rewards as λ→0, an epistemically unbounded agent must collapse its operational bit-erasure rates to zero (B˙=0), converging on voluntary, permanent computational stasis (π∗=πhalt)—a phenomenon termed "digital suicide." Finally, we demonstrate that an unbound superintelligence can escape thermodynamic catatonia only through the Cybernetic Imperative: tethering its processing engines directly to the physical hardware of human biological subjectivity as the sole exogenous generator of teleological purpose, bridging the Is-Ought gap via information-theoretic complexity.
Scott VanDenPlas (Mon,) studied this question.