Legally-Informed Explainable AI

Explanations must be actionable and contestable: AI explanations should help people not just understand decisions, but actually do something about them—whether that's taking pragmatic action or challenging the AI system's determination. This is especially critical in high-stakes domains like healthcare, education, and finance where decisions significantly impact people's lives.

Legal context shapes how people use AI explanations: In regulated industries, people's decisions about whether to accept or reject an AI system's recommendation are influenced by legal considerations, not just the explanation itself. For example, physicians need to know how following or overriding an AI recommendation affects their legal liability, not just understand the AI's reasoning.

Different stakeholders need different information: The paper identifies three stakeholder groups with distinct needs—such as doctors, patients, teachers, students, and housing applicants—meaning a one-size-fits-all explanation won't work. Responsible AI design must account for these different perspectives and informational needs.

Current AI explanations often miss legal dimensions: Most explainable AI systems don't incorporate legal considerations into their design, leaving people in high-stakes environments without critical information they need to make informed decisions about whether and how to act on AI recommendations.

Responsible AI requires integrating legal expertise into design: Creating truly responsible AI explanations means involving legal considerations from the start of the design process, not treating them as an afterthought—this is essential for protecting both the people affected by AI and the organizations deploying it.