Socio-Technical Machine Learning - Putting the User in Control

Data scientists and users are experts in different domains, with the former possessing expertise in technology and the latter having experience in the problem domain. Current approaches to machine learning (ML) design enable data scientists and technical developers to drive the process. In contrast, this research presents an approach grounded in theory that enables the non-technical user to control ML design, at least initially. We discuss a study that applies a socio-technical systems thinking approach to enable non-technical users to drive machine learning design and ensure their actual needs are met. We extend socio-technical research by introducing a novel type of analysis- a social data system analysis, and provide empirical evidence on the feasibility of non-technical user-driven ML design. We provide practical guidelines for practitioners to enable non-technical user-driven ML design. The study also demonstrates the potential of a socio-technical approach in increasing fairness in training datasets.