The Five-Contract Model and the Responsibility Closure Spectrum: A Governance Framework for High-Impact AI Systems

Key Points

• This paper aims to provide a framework for enhancing accountability in AI systems through the five-contract model and responsibility closure spectrum.
• Introduced the Accountability Threshold theory with the five-contract model consisting of Subject, Interface, Evidence, Freeze, Compensation.
• Developed a responsibility closure spectrum to replace binary judgments with graded requirements.
• Presented case studies on an autonomous vehicle accident and an AI medical error to illustrate the framework.
• Demonstrated that the five-contract model effectively operationalizes responsibility in AI systems.
• Showed that the responsibility closure spectrum provides a more nuanced understanding of accountability in high-impact scenarios.

Abstract