Stakeholders in Explainable AI

Different stakeholders need different types of explanations: The paper identifies multiple distinct communities (such as developers, regulators, end-users, and affected individuals) who all care about AI explainability but have fundamentally different questions in mind. When a regulator asks "Why did this AI make this decision?" they're asking something different than when an affected person asks the same question, which is why a one-size-fits-all approach to explainability won't work.

"Explainability" means different things depending on your role: The authors use the software engineering concepts of "validation" (did we build the system right?) and "verification" (did we build the right system?) to show that some stakeholders care about whether an AI system works as intended, while others care about whether it should be used at all. This distinction explains why stakeholder communities often talk past each other when discussing AI transparency.

Stakeholder concerns are compatible but not identical: While the paper notes that different communities' interests in explainability are "broadly compatible," they diverge in important ways. This means that efforts to make AI more explainable need to be tailored to specific stakeholder needs rather than assuming that explaining a system to engineers will satisfy regulators or the public.

Clarifying what we're asking about is essential for responsible AI: The paper emphasizes that much confusion in AI governance stems from ambiguity about what "explainability" actually means. For responsible AI development, stakeholders need to be explicit about whose questions they're trying to answer and what kind of explanation would actually be useful.