The relationship between professional and general ethics in generative AI

General ethics and professional ethics often conflict in AI systems: The paper identifies a growing gap between broad ethical principles used to guide AI development and the specific ethical codes that govern professional fields like medicine, law, and translation. AI systems deployed in these domains are effectively implementing professional ethics on the ground, but this hasn't been adequately addressed in AI ethics research and policy.

Professional ethics operate within a hierarchical structure relative to general ethics: Rather than existing separately, professional ethics are structured hierarchically with general ethics, meaning they build upon and sometimes override general ethical principles in specific professional contexts. Understanding this relationship is essential for designing AI systems that work appropriately in specialized domains.

AI systems have "favored ethics" that can be identified and studied systematically: The paper proposes that researchers can assess which ethical priorities an AI model actually implements by testing it across similar but slightly different scenarios in professional contexts. This empirical approach reveals the nuances of what ethics a system actually follows in practice, not just what it claims to follow.

Equilibrium between competing ethics determines real-world AI behavior: When general and professional ethics conflict, AI systems reach some form of equilibrium that determines which ethical principles get prioritized in actual decisions. Identifying how this equilibrium forms is crucial for ensuring AI behaves appropriately in professional practice.

Interventions can be designed to change which ethics AI systems prioritize: The paper suggests that once we understand how AI models favor certain ethics, we can develop targeted interventions to shift their ethical priorities in professional contexts, offering a path toward more responsible AI deployment in specialized fields.