Human-centered Explainable AI: Expanding Explainable & Responsible AI

If AI systems are going to inform consequential decisions such as deciding whether you should get a loan, they must be explainable to everyone, not just software engineers. While there has been commendable progress in Explainable AI (XAI) around “opening” the AI’s black-box, there is an overlooked insight: who opens the black-box matters just as much as opening it. As a result, many popular XAI interventions are ineffective and even harmful in real-world settings. In this talk, I will address this intellectual blind spot by introducing and operationalizing Human-centered XAI (HCXAI), a holistic sociotechnical paradigm for explainable and responsible AI. With a focus on non-AI experts, the talk will take the audience on a journey involving three “turns”--- to the machine, human, and sociotechnical. I will (a) exhibit how Social Transparency can encode socio-organizational context to augment AI explainability without changing the internal model, (b) demonstrate how we can make Explainable AI systems seamful, and (c) illustrate how algorithmic harms persist in the algorithm’s afterlife and what we should do about it. I will share contributions of my work on (1) expanding the XAI design space, (2) deepening our knowledges of “who” the humans are and their explainability needs, and (3) enabling resourceful ways to do Responsible AI (RAI). I will discuss the impact of my work on informing Responsible AI policies, industry adoption, and pioneering the domain of HCXAI and nurturing a vibrant research community. Finally, I will cover my future research agenda on designing for recourse in AI, minimizing the sociotechnical gap in Large Language Model-powered healthcare technology, and imprint-aware algorithmic impact assessments. The work presented in this talk serves my vision of creating a future where anyone, regardless of their background, can interact with AI systems in an explainable, accountable, and dignified manner.