A multi-component framework for the analysis and design of explainable artificial intelligence

XAI research lacks a unified framework: The paper identifies that despite rapid growth in explainable AI research, there is "yet to appear a principled framework" grounded in scientific principles of explainability, meaning current XAI tools and techniques are being developed without consistent standards or shared understanding.

Five levels of XAI maturity exist: The authors propose a framework that calibrates "five successive levels of XAI," suggesting that explainability is not binary but exists on a spectrum—organizations should understand where their AI systems fall on this scale rather than simply labeling them as "explainable" or not.

Historical scientific principles should guide XAI development: The paper emphasizes the need to "respect the literature of explainability in the history of science," indicating that XAI should draw on established scientific methods for explanation rather than inventing entirely new approaches, which could improve the quality and reliability of explanations.

Regulatory pressure is driving XAI adoption without sufficient rigor: The "perfect storm" of industrial demand and regulatory requirements has created urgency to deploy XAI tools, but this pressure may be outpacing the development of principled, well-designed approaches—a concern for organizations implementing XAI to meet compliance requirements.

Strategic inventory of XAI requirements is needed: The authors call for a systematic cataloging of what explainability actually requires in different contexts, suggesting that organizations should first clarify their specific XAI needs before selecting tools, rather than adopting generic solutions.