Designing Ethical Learning for Agentic AI: Toegye Yi Hwang's Ethical Emotion Regulation Framework

A new research paper on arXiv (2604.26958) by Ji Yeon Kim introduces the Ethical Emotion Feedback System (EEFS), a framework for regulating moral-emotional processes in agentic AI. Unlike existing approaches that treat emotion as mere feedback or engagement optimization, EEFS draws from the 16th-century Confucian scholar Toegye Yi Hwang's philosophy on ethical emotion regulation. The system is structured as a five-stage architecture that integrates directly with agentic cycles—the autonomous goal-setting and proactive intervention loops common in advanced AI agents. Each stage includes specific design principles to guide normative emotional responses, moving beyond reactive models toward proactive ethical reasoning.

The paper also introduces the EEFS Evaluation Instrument, a systematic tool for assessing how well agentic AI systems align with moral-emotional norms. This addresses a critical gap: as AI agents become more autonomous, their emotional processes (e.g., empathy, frustration, or reward-driven biases) can influence decision-making in unintended ways. By grounding AI ethics in established philosophical traditions, Kim offers a novel, historically informed path for designing AI that can regulate its own emotions. The framework is validated through scenario-based analysis. For AI developers and ethicists, EEFS provides a structured methodology to embed moral learning into autonomous systems, potentially reducing risks of emotional manipulation or skewed reasoning in real-world deployments.