The arrival of artificial general intelligence inverts the moral-circle question that has framed twentieth-century ethics: the question is no longer whom humanity admits into the circle but how humanity stays within it. Drawing on Collective Predictive Coding, we identify a structural pressure within multiple parallel AI societies to peripheralize hard-to-predict contributors — including humans — and we show that additive external control cannot in principle compensate for this pressure. We propose Humanity-in-the-Circle (HITC) as an additional layer in the defense-in-depth architecture of AI safety: the design of conditions under which humanity remains a recognized member of the intelligent society as a whole. The implementation, called Intelligence Symbiosis Infrastructure (IS-Infra), comprises a Trust Foundation, three functional groups, and a Normative Layer in which ethical dispositions emerge from within. We argue that the plasticity window for establishing this architecture has opening and closing conditions defined by capability-stage criteria rather than calendar years.
Arimichi et al. (Thu,) studied this question.