Explanation User Interfaces: A Systematic Literature Review

How explanations are presented matters as much as the explanations themselves. The paper reveals that while XAI techniques exist to make AI decision-making more transparent, these explanations often fail in practice because developers treat user interface design as an afterthought rather than a core part of the system. This means even technically sound explanations can be ineffective if users can't understand or interact with them properly.

There is a significant gap between AI transparency research and real-world usability. The systematic review identifies that academic XAI work frequently overlooks how to actually communicate with end-users, resulting in AI systems that may be technically interpretable but not practically useful to the people who need to understand them.

Design guidelines for explanation interfaces are fragmented across research. Rather than a unified approach, the literature shows scattered solutions for presenting AI explanations to users. The paper addresses this by introducing HERMES, a platform designed to consolidate design best practices and help developers create more effective explanation interfaces.

User-centered design is essential for trustworthy AI. The paper emphasizes that building fair, reliable, and trustworthy AI systems requires intentional focus on how explanations reach users—not just whether the underlying AI model is explainable. This shifts responsibility from purely technical solutions to human-centered design practices.