A Means-End Account of Explainable Artificial Intelligence

There is no one-size-fits-all explanation for AI systems. The paper identifies fundamental disagreement in the field about what should be explained, who needs the explanation, how to explain it, and why—suggesting that the "best" explanation depends entirely on the specific context and purpose.

Different stakeholders need different types of explanations. The paper emphasizes that explanations must be tailored to their audience (topic, stakeholder, and goal), implying that an explanation suitable for a regulator may not work for an end-user, and vice versa.

We can evaluate XAI methods by checking if they match their intended purpose. Rather than debating whether one explanation method is universally superior, the paper proposes assessing whether a chosen method actually fits the specific topic being explained, the person receiving the explanation, and the goal the explanation is meant to achieve.

The field needs clearer frameworks to avoid mismatched solutions. By organizing XAI contributions into a structured taxonomy based on means-end relations, the paper suggests that current confusion in the field stems partly from applying explanation methods without clearly defining what problem they're meant to solve.

Responsible AI requires intentional design of explanations. The paper implies that organizations deploying AI systems should deliberately determine why they need explanations before choosing how to generate them, rather than applying generic explanation techniques.