Stuart Grant

Artificial intelligence (AI) alignment, ecological governance, and pluralist political ethics are usually treated in separate literatures. Yet advanced AI systems, human institutions, and ecological systems increasingly act within the same substrate. Existing frameworks are therefore strongest on single interfaces: alignment research formalises human–AI cooperation but treats humans as the only direct normative principals; environmental frameworks specify shared planetary limits without an architecture for cross-class governance; and political theories of pluralism and consent provide resources developed primarily for human polities. This paper proposes a research programme—convergent ethics—to address that gap. It advances: (1) a minimal viability ontology of agency at specified levels of abstraction; (2) six constitutive commitments for cross-class analysis; (3) a model of sandboxed pluralism organising separate domains, shared domains, and governed boundaries; (4) a triage architecture combining capabilities-weighting, sandboxing, and residual maximin; (5) a procedural composition in which a deliberative polity operates under constitutional floor constraints with stewardship-based representation for non-self-advocating classes; and (6) a research agenda, including provisional criteria for when artificial systems may become candidates for direct moral standing. The framework is proposed, not settled. Its central claim is not that heterogeneous agents share a single welfare metric, but that they often share vulnerable viability conditions whose destruction is presumptively unjustifiable on prudential, moral, and strategic grounds. The paper develops this claim through an advanced-AI coexistence case, with implications for ecological overreach and harm reduction across heterogeneous classes, and argues that positive-sum institutional design is an under-explored way to reduce pressures associated with instrumental convergence.