Explainable AI: roles and stakeholders, desirements and challenges

The Stakeholder Playbook aims to assist developers of explainable AI (XAI) systems in addressing the diverse needs of different stakeholders or role-holders seeking insights into AI systems. The research involved cognitive interviews with experienced AI professionals to identify the requirements of stakeholders, revealing that these individuals often require access to trusted engineers and vendors to develop comprehensive mental models of AI systems. Understanding failures and limitations of the AI is as crucial as understanding its functionality. Some stakeholders need to grasp the AI well enough to explain it to others, extending beyond their personal sense-making needs. About half of the interviewees expressed the need for explanations, leading to the creation of a Playbook categorizing explanation desires, challenges, and cautions for various stakeholder groups. The article addresses the paradoxical nature of these findings by acknowledging the diverse skill sets and goals of different stakeholders. The Playbook's goal is to guide XAI developers in crafting explanations that cater to these distinct roles and responsibilities of stakeholders beyond end-users.

Link to the full article: Frontiers | Explainable AI: roles and stakeholders, desirements and challenges (frontiersin.org)