In geothermal reservoir management, combined simulation-optimization is a practical approach to achieve the optimal well placement and operation that maximizes energy recovery and reservoir longevity. The use of machine learning models is often essential to make simulation-optimization computational feasible. Tools from machine learning can be used to construct data-driven and often physics-free approximations of the numerical model response, with computational times often several orders of magnitude smaller than those required by reservoir numerical models. In this short perspective, we explain the background and current status of machine learning based combined simulation-optimization in geothermal reservoir management, and discuss several key issues that will likely form future directions.