Towards Artificial Virtuous Agents: Games, Dilemmas and Machine Learning

Virtue ethics offers a promising alternative to dominant AI ethics approaches: The paper argues that virtue ethics—which focuses on developing good character traits rather than following rules (deontology) or maximizing outcomes (utilitarianism)—has been underexplored in machine ethics despite its potential advantages, suggesting AI developers should consider this complementary ethical framework.

Role-playing games with moral dilemmas can train AI agents to make virtuous decisions: The authors propose using games like Papers, Please and Life is Strange—where players face genuine ethical trade-offs—as training environments to help AI systems develop virtuous decision-making patterns, moving beyond simple rule-following.

Current AI ethics frameworks struggle with real-world complexity: The paper identifies key challenges in formalizing virtue ethics for AI, including difficulties in codifying virtues and resolving genuine ethical dilemmas, highlighting why practical implementation of virtue-based AI has lagged behind other ethical approaches.

Modern AI techniques can make virtuous agents understandable: By combining affinity-based reinforcement learning with explainable AI methods, the authors suggest we can both train virtuous agents and examine their reasoning through an ethical lens, addressing the "black box" problem in AI decision-making.

This approach represents early-stage foundational work: The paper positions virtuous AI agents as "a first step" toward formalizing virtue ethics in practice, indicating this is emerging research that could influence how we develop more ethically sophisticated AI systems in the future.