Mitigating Adversarial Norm Training with Moral Axioms

This paper addresses the issue of adversarial attacks on ethical AI systems. We investigate using moral axioms and rules deontic logic in a norm learning framework to mitigate training. model intuition construction provides systems with guard rails yet still allows for conventions. evaluate our approach by drawing inspiration from study commonly used development research. questionnaire aims test an agent's ability reason conclusions despite opposed testimony. Our findings suggest that can correctly situations learn conventions training environment. conclude adding axiomatic prohibitions inference makes it less vulnerable attacks.