The Thousand Faces of Explainable AI Along the Machine Learning Life Cycle: Industrial Reality and Current State of Research

XAI research is misaligned with industry needs: While academics have developed many explainability methods, most focus on the model evaluation phase for data scientists. However, industry practitioners need explainability tools across the entire ML lifecycle—from initial problem definition through deployment and monitoring—revealing a significant gap between what researchers study and what businesses actually require.

Explainability tools remain inaccessible to non-experts: The paper confirms that current XAI methods and frameworks are primarily designed for technical specialists. For responsible AI deployment, organizations need explainability approaches that allow non-technical stakeholders (business leaders, regulators, end-users) to understand AI decisions, but existing tools fall short of this goal.

Different industry sectors have varying XAI priorities: The research involved stakeholders across multiple industries, suggesting that explainability needs are not one-size-fits-all. Different sectors likely require different approaches to making AI systems transparent and trustworthy, yet current research doesn't adequately address this diversity.

The ML lifecycle requires integrated explainability strategies: Rather than treating explainability as a single technical problem, the paper suggests organizations need explainability solutions at multiple stages—from data preparation through model deployment. This systemic approach is underexplored in current academic research.