Responsible AI Engineering: The Case of an Inclusive Image Annotation Team

Artificial Intelligence (AI) models require a high quantity of high-quality data to learn from. For tolling applications, traffic cameras can collect massive amounts of images. However, annotating them is expensive, time-consuming, and requires distinct skills. People on the autism spectrum, on the other hand, typically have a hard time finding work, especially fulfilling work that benefits from their skills. Based on six interviews, we present the case of an inclusive traffic image annotation team in a global technology company that bridges the gap between the growing demand for high-quality image annotation and the difficult employment situation of people with autism. In particular, we explore relevant use cases in the pilot project, what requirements can be identified to improve software tool support, and what the enabling factors are in general. The results indicate that this initiative has a positive impact along many dimensions. The overall goal of this paper is to provide researchers and practitioners with a blueprint for creating inclusive AI annotation workplaces–with modifications where necessary.CCS CONCEPTS• Human-centered computing → Human computer interaction (HCI); Participatory design; Accessibility; Collaborative and social computing; • Software and its engineering → Requirements analysis; • Computing methodologies → Artificial intelligence.