Optimal regimes for algorithm-assisted human decision-making

We introduce optimal regimes for algorithm-assisted human decision-making. Such regimes are decision functions of measured pre-treatment variables and enjoy a "superoptimality" property whereby they are guaranteed to outperform conventional optimal regimes currently considered in the literature. A key feature of these superoptimal regimes is the use of natural treatment values as input to the decision function. Importantly, identification of the superoptimal regime and its value require exactly the same assumptions as identification of conventional optimal regimes in several common settings, including instrumental variable settings. As an illustration, we study superoptimal regimes in an example that has been presented in the optimal regimes literature.