MARIA: A Framework for Marginal Risk Assessment without Ground Truth in AI Systems

TL;DR

This paper introduces MARIA, a framework for assessing the relative risk of AI systems without relying on ground truth, focusing on predictability, capability, and interaction dominance to guide safe deployment.

Contribution

It proposes a novel marginal risk assessment framework that evaluates AI systems comparatively, bypassing the need for ground truth or absolute risk measurements.

Findings

Enables risk comparison without ground truth

Identifies AI system strengths and risks relative to existing processes

Provides actionable guidance for responsible AI deployment

Abstract

Before deploying an AI system to replace an existing process, it must be compared with the incumbent to ensure improvement without added risk. Traditional evaluation relies on ground truth for both systems, but this is often unavailable due to delayed or unknowable outcomes, high costs, or incomplete data, especially for long-standing systems deemed safe by convention. The more practical solution is not to compute absolute risk but the difference between systems. We therefore propose a marginal risk assessment framework, that avoids dependence on ground truth or absolute risk. It emphasizes three kinds of relative evaluation methodology, including predictability, capability and interaction dominance. By shifting focus from absolute to relative evaluation, our approach equips software teams with actionable guidance: identifying where AI enhances outcomes, where it introduces new risks, and how to adopt such systems responsibly.