Demystifying Artificial Intelligence for End-Users: Findings from a Participatory Machine Learning Show

Interactive and collaborative approaches have been successfully used in educational scenarios. For machine learning and AI, however, such approaches typically require a fair amount of technical expertise. In order to reach everyday users of AI technologies, we propose and evaluate a new interactive approach to help end-users gain a better understanding of AI: A participatory machine learning show. During the show, participants were able to collectively gather corpus data for a neural network for keyword recognition, and interactively train and test its accuracy. Furthermore, the network’s decisions were explained by using both an established XAI framework (LIME) and a virtual agent. In cooperation with a museum, we ran several prototype shows and interviewed participating and non-participating visitors to gain insights about their attitude towards (X)AI. We could deduce that the virtual agent and the inclusion of XAI visualisations in our edutainment show were generally rated positively by participants, even though the frameworks we used were originally designed for experts. When comparing both groups, we found that participants felt significantly more competent and positive towards technology compared to non-participating visitors. Our findings suggests that the consideration of specific user needs, personal background, and mental models about (X)AI systems should be included in the XAI design for end-users.