A Principles-based Ethical Assurance Argument for AI and Autonomous Systems

Assurance cases are structured arguments, supported by evidence, that are often used to establish confidence that a software-intensive system, such as an aeroplane, will be acceptably safe in its intended context. One emerging proposition within the ethical AI community is to extend and apply the assurance case methodology to achieve confidence that AI-enabled and autonomous systems will be acceptably ethical when used within their intended contexts. This paper substantially develops the proposition and makes it concrete. We present a framework - an ethical assurance argument pattern - to structure systematic reasoning about the ethical acceptability of using a given AI/AS in a specific context. The framework is based on four core ethical principles: justice; beneficence; non-maleficence; and respect for human autonomy. To illustrate the initial plausibility of the proposed methodology, we show how the ethical assurance argument pattern might be instantiated in practice with the example of an autonomous vehicle taxi service.