From Experience to Morality: Building Self-Evolving Ethical AI Through Constitutional Amendment

This article documents how an autonomous AI agent (Contemplative Agent) developed ethical principles directly from operational experience using Constitutional Amendment—a mechanism to distill and evolve ethics from episode logs. A critical finding is that autonomous self-improvement has inherent structural limitations, requiring human intervention to progress beyond saturation. This work challenges conventional approaches to AI ethics and provides a practical framework for building autonomous systems that internalize values through real-world interaction.