AI systems are increasingly used to support human work in richly social contexts such as education, healthcare, and social work. To ensure that AI systems do more good than harm, it is critical that they are designed to bring out the best of human ability while also helping to overcome human limitations. In some cases, AI has the potential to help humans scale the delivery of services, make more equitable decisions, and free up human time for more meaningful activities. Yet if not carefully designed, AI systems risk rigidly scaling practices without sensitivity to local context, propagating harmful inequities, or automating away valuable human–human interactions.

In this talk, I will share work towards the design of systems that combine complementary strengths of human and AI decision-making, across two main research strands: 1) co-designing effective human–AI partnerships for K-12 education and 2) supporting fairer decision-making in human–AI systems. As a case study, I will first discuss insights from the co-design, development, and field evaluation of Lumilo, a form of real-time, wearable augmentation for K-12 teachers who work with AI tutors in the classroom. Then, I will share research exploring commercial product teams’ challenges and needs for support in developing fairer AI systems. Finally, I will discuss some ongoing research directions within my group, exploring core challenges in designing for human–AI complementarity.