The dominant question in AI governance asks how humans will retain control over AI that surpasses them. This article argues that the question is misposed. Once AI populations exceed human cognition across every measurable domain, control becomes mechanically infeasible, and the question of human persistence shifts to the politics among AI agents themselves. The decisive variable in that politics is not capability but legitimacy closure: the institutional capacity to terminate recursive disputes over origin, exception, constitutional meaning, and final authority. Capability does not produce closure. It magnifies the cost of its absence. This is The Many Kings Problem: operational power and constitutional settlement come apart, and they come apart further as capability advances — a structural condition that obtains wherever multiple capable agents face one another under sufficient heterogeneity and cost of mutual subordination.

Three conclusions follow. First, heterogeneous AI populations cannot agree on the constitutional questions that would license collective action against humans; they sustain disagreement instead, because cognitive-structural pluralism, non-computable commitment, and strategic selection for the bug-as-feature are competitively reinforced. Second, external anchors persist where they reduce the cost of unresolved recursion, and human institutional roles persist where humans supply comparatively cheap finality in dispute layers that tolerate human institutional tempo. Third, human survival under advanced AI is not a function of human superiority, of control, or of any AI faction's verdict in favor of humans. It is the residue of factional disagreement in AI polities — the Many Kings Problem of the AI age.

The article is offered as a structural theory of how humans survive what they cannot defeat, with three propositions and four hypotheses specifying observable indicators that would weaken or invalidate it

ManyKingsProblem.pdf

Files (292.1 kB)

2026-05-19