Ethical AI Triplet: A Framework for Stress-Testing Fairness in Digital Twins in Healthcare

Abstract

With “Digital Twins” tools integrating more Artificial Intelligence agents, those become powerful tools for optimizing domains like public health. Ensuring their ethical alignment with core medical rules (e.g. empathy, equity, justice etc.) is paramount. This paper proposes a counter framework to balance them: the Ethical AI Triplet. Our novel architecture acts as an automated “ethics committee” that validates policies proposed by an optimization AI. The “Triplet” has three specialized agents: a Population Generator that creates diverse societies, a Simulation Engine that runs proposed policies against the societies, and an Ethical Auditor that evaluates the outcomes against metrics of fairness and harm. Unlike traditional AI governance, which relies on post-hoc audits and manual reviews, the Ethical AI Triplet showcases an automated, adversarial framework designed to stress test policies for ethical robustness before deployment, identifying in time harmful possibilities that existing frameworks may miss.

DOI: 10.61416/ceai.v27i3.9745