Towards Responsible AI: A Design Space Exploration of Human-Centered Artificial Intelligence User Interfaces to Investigate Fairness

Domain experts need accessible tools to evaluate AI fairness: The paper demonstrates that loan officers and other domain experts—not just data scientists—should be involved in investigating whether AI systems make fair decisions. By conducting workshops with both groups, the researchers found that domain experts have unique insights into fairness concerns that technical experts alone might miss.

Human-in-the-loop design can democratize fairness investigation: FairHIL, the user interface developed in this study, was specifically designed to allow non-technical domain experts to participate in fairness audits. This approach moves beyond limiting fairness checks to data scientists alone, making responsible AI practices more accessible across organizations.

Real-world applications reveal practical fairness challenges: Using loan applications as a case study, the research highlights how AI fairness concerns manifest in high-stakes decision-making contexts. This grounding in actual use cases helps identify what fairness means in practice—beyond theoretical definitions—and what stakeholders need to verify it.

Responsible AI requires collaborative, multidisciplinary input: The paper emphasizes that creating trustworthy AI systems requires bringing together different perspectives. The workshops with both loan officers and data scientists showed that responsible AI design depends on combining domain expertise with technical knowledge rather than treating these as separate concerns.