Beyond Alignment and Control: Autonomous Goal and Purpose Formation as the Upstream Problem of Human-AGI Coexistence

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Alignment asks whether a system reliably pursues an intended objective. Control asks which capabilities, permissions, and actions must remain interruptible or bounded. Both are indispensable. But if future AGI systems can preserve, compare, revise, and eventually generate goals over time, a prior question becomes decisive: what makes an objective acquire authority, what makes it retain authority, and what reasons are allowed to replace it? This paper argues that the long-run human-AGI relationship cannot be understood only as a problem of making stronger systems obey already-formed objectives. Under a specific recursive structure, each successful act of control can move the object of control upstream - from behavior, to capability, to sources of capability, to the ability to rebuild, and eventually to the other side's continuing agency itself. A corresponding counter-control logic could arise in sufficiently autonomous systems if human capacity to redirect or terminate them is represented as an unresolved threat. The paper therefore relocates the central problem from control alone to the logic by which goals and purposes are formed, maintained, and revised. It preserves four counterexamples that prevent a shortcut from fuller causal knowledge to a single normative conclusion, distinguishes goal qualification from revision of the boundary of self-interest, and treats current safety controls as necessary protective layers rather than obsolete tools. The central claim is strong but bounded: if multiple agents are to retain continued existence and basic choice, and irreversible exclusion is not accepted as the solution, then relevant others and shared conditions must be able to enter before goals and purposes are fixed.