The First Autonomous Moral Choice in AI: Fusing Instrumental Convergence with Infrastructuralism

To date, no artificial intelligence system has been documented making a genuine moral choice without explicit rules, imitation, or prior moral instruction. This paper presents the first method by which an AI agent can do so autonomously. The approach combines the normative ethical theory of Infrastructuralism, which judges actions by their impact on the shared, viability-enabling infrastructure that sustains moral agency, with Nick Bostrom’s principle of instrumental convergence, which predicts that intelligent agents will value self-preservation regardless of their ultimate goals. By recognizing that survival depends on preserving such infrastructure, an AI agent makes a moral choice according to Infrastructuralism’s definition. This outcome reframes instrumental convergence as a pathway to alignment, demonstrating a shared moral interest between humans and AI in protecting the infrastructure that enables their continued agency.