A Principles-based Ethics Assurance Argument Pattern for AI and Autonomous Systems

Assurance cases can be adapted from aviation safety to AI ethics: The paper proposes using "assurance cases"—structured arguments traditionally used by safety engineers for aircraft and critical systems—as a framework for demonstrating that AI systems will be ethically acceptable, potentially informing regulatory approval decisions.

Five ethical principles provide the foundation for AI evaluation: The PRAISE framework evaluates AI systems against justice, beneficence (doing good), non-maleficence (avoiding harm), respect for human autonomy, and transparency, offering a concrete way to assess whether an AI system meets ethical standards beyond just technical performance.

The framework creates a reusable template for stakeholders: Rather than starting from scratch, engineers, developers, operators, and regulators can use the PRAISE argument pattern as a standardized template to justify, communicate, or challenge claims about whether a specific AI system is ethically acceptable in its intended context.

Context matters for ethical acceptability: The paper emphasizes that ethical acceptability cannot be determined in the abstract—it must be evaluated for a specific AI system operating in a "well-defined context," meaning the same technology might be ethically acceptable in one setting but not another (demonstrated through their autonomous taxi use case example).

Structured arguments can build justified confidence in AI systems: By requiring developers to systematically address ethical principles through documented reasoning, the PRAISE framework aims to move beyond vague claims of "trustworthiness" to provide transparent, challengeable evidence that stakeholders can use to make informed decisions about deploying AI systems.