Can the assessment change the decision?
An AI impact assessment has ethical value when it subjects a proposal to credible challenge and demonstrably informs an accountable decision. A completed template alone does not establish that it did.
This white paper sets out the ethical foundations an organization needs before completing an assessment template. It draws lessons from earlier impact assessment regimes, examines impacts on workers, communities and other affected parties, and addresses the governance of acquired AI systems.
Five principles
- Necessity and proportionality: Could a less intrusive means achieve the purpose, and do supported benefits justify the intrusion that remains?
- Human dignity, agency and context: How does a system change ordinary life when information moves beyond the setting in which it arose?
- Power and procedural fairness: Can affected people raise concerns, challenge consequential results and seek accommodation?
- Accountability and human oversight: Who verifies outputs before harm, and who can change or stop the deployment?
- Transparency and traceability: What does the system observe and infer, who uses that information, and what record remains?
The paper also considers assessor independence, uncertainty, illustrative scenarios and questions for leaders reviewing a proposed deployment.
Download the white paper (PDF, 25 pages) ↓
Open the PDF in your browser ↗ · Discuss the paper on LinkedIn ↗
This paper is general information and discussion, not professional advice. See the notice and references in the PDF.