Inclusive Design of AI's Explanations: Just for Those Previously Left Out, or for Everyone?

Inclusive design approaches to AI explanations can benefit everyone, not just underserved users. The study found "curb-cut effects" where fixes designed to help specific user groups (in this case, using the GenderMag inclusive design method) led to overall increased engagement with explanations and better mental model concept scores across all participants.

One-size-fits-all explanations don't work for AI systems. The research confirms that "many cases of different explanations serving some users well and being unhelpful to others," highlighting why tailored approaches to XAI design matter for responsible AI deployment.

Inclusive design methods from software engineering can be adapted for AI. The study demonstrates that practitioners can adopt established inclusive design approaches (like GenderMag) to improve AI explainability systems, suggesting a practical pathway for making AI more responsible and understandable.

Improving explanations for inclusivity requires ongoing iteration. While the inclusivity fixes showed promise in some areas, the abstract notes they "did not impr[ove]" in other measured outcomes, indicating that creating truly inclusive AI explanations is an iterative process requiring continued testing and refinement.

Mental models matter for responsible AI. The study measured whether users developed accurate mental models of how the AI works, suggesting that responsible AI requires not just explanations, but explanations that help diverse users genuinely understand AI system behavior.