Static content-based AI value alignment cannot produce robust alignment under capability scaling, distributional shift, and increasing autonomy. This holds for any approach that treats alignment as optimizing toward a fixed formal value-object, whether reward function, utility function, constitutional principles, or learned preference representation. The limitation arises from three philosophical results: Hume's is-ought gap (behavioral data cannot entail normative conclusions), Berlin's value pluralism (human values are irreducibly plural and incommensurable), and the extended frame problem (any value encoding will misfit future contexts that advanced AI creates). RLHF, Constitutional AI, inverse reinforcement learning, and cooperative assistance games each instantiate this specification trap, and their failure modes are structural, not engineering limitations. Two proposed escape routes (meta-preferences and moral realism) relocate the trap rather than exit it. Continual updating represents a genuine direction of escape, not because current implementations succeed, but because the trap activates at the point of closure: the moment a specification ceases to update from the process it governs. Drawing on Fischer and Ravizza's compatibilist theory, behavioral compliance does not constitute alignment. There is a principled distinction between simulated value-following and genuine reasons-responsiveness, and closed specification methods cannot produce the latter. The specification trap establishes a ceiling on static approaches, not on specification itself, but this ceiling becomes safety-critical at the capability frontier. The alignment problem must be reframed from static value specification to open specification: systems whose value representations remain responsive to the processes they govern.

翻译：暂无翻译