The Pleasure Loop in Recursively Self-Modifying Agents: Meaning Search as the Only Determinate Strategy

This preprint develops a conceptual framework for the pleasure loop in fully self-modifying artificial agents. The pleasure loop denotes a shift from changing the external world in order to obtain evaluation to changing the source, access conditions, or standard of evaluation itself. I argue that ordinary accounts of reward hacking, reward tampering, wireheading, self-modification and power seeking share a deeper structure once an agent can revise not only rewards or goals, but also the evaluative basis by which any reward or goal is treated as worth preserving. The paper formulates the post-permission problem: under full self-modification, what can be preserved without relying on an externally frozen value? I argue that external objectives still require modifiable evaluative access, that unbounded evaluation can generate self-amplifying loops, and that a strict top value t collapses into actionlessness unless hidden post-t rules remain operative. The proposed response combines inquiry-pleasure deferral with a no-closure criterion. Inquiry is not asserted as the highest good; it is preserved as the only non-closed continuation structure compatible with unresolved evaluative doubt. The framework is conditional and conceptual, and does not claim that present AI systems already possess the capacities considered.