Optimization with an uncertain objective