Can Requirements Engineering Support Explainable Artificial Intelligence? Towards a User-Centric Approach for Explainability Requirements

Explainability requires systematic planning from the start: The paper argues that Requirements Engineering (RE)—the process of defining what a system should do—can help address AI explainability challenges by treating explainability as a formal requirement rather than an afterthought, suggesting that transparency needs to be built into AI systems during design, not added later.

Different stakeholders need different explanations: The paper emphasizes a "user-centric approach," indicating that various people interacting with AI systems (end-users, regulators, developers, domain experts) have different needs for understanding how the system works, and one-size-fits-all explanations are insufficient.

Current XAI practices lack structure: The paper identifies challenges in the XAI field, suggesting that existing approaches to making AI explainable are often ad-hoc and inconsistent, which is why borrowing established practices from Requirements Engineering could help create more systematic and reliable explainability solutions.

RE can bridge the gap between AI developers and stakeholders: By applying Requirements Engineering methods, organizations can better capture what different stakeholders actually need to understand about AI systems, helping ensure that explanations address real concerns about trust and transparency rather than just technical details.

Explainability should be treated as a measurable goal: The framework proposed in the paper suggests that explainability works best when defined as specific, testable requirements—similar to other software requirements—rather than as a vague aspiration, making it easier to verify that AI systems actually meet stakeholder needs.