Learning about Responsible AI On-The-Job: Learning Pathways, Orientations, and Aspirations

Prior work has developed responsible AI (RAI) toolkits and studied how AI practitioners use such resources when practicing RAI. However, AI practitioners may not have the relevant skills or knowledge to effectively use RAI resources—particularly as pre-trained AI models have enabled more people to develop AI-based applications. In this paper, we explore current practices and aspirations for learning about RAI on-the-job, by interviewing 16 AI practitioners and 24 RAI educators across 16 organizations. We identify AI practitioners' learning pathways for RAI, including information foraging and interpersonal learning; the orientations of RAI learning resources towards computational and procedural approaches to RAI; and aspirations for RAI learning, including desires for more sociotechnical approaches to understand potential harms of AI systems—aspirations that can be in tension with organizational priorities. We contribute empirical evidence of what and how AI practitioners are learning about RAI, and we suggest opportunities for the field to better support sociotechnical approaches to learning about RAI on-the-job.