The aim of this workshop is to explore theoretical foundations of optimally combining human and statistical judgments. Complementarity, referring to the superior performance of a human paired with a statistical model over either alone, is a goal when deploying predictive models to support decision-making in high-stakes domains like medicine or criminal justice. However, considerable empirical evidence suggests that complementarity is difficult to design for and achieve in practice, even when experts are assumed to have access to information that a model may not. This workshop considers how to rigorously define, design for, and evaluate human-AI complementarity.