I think the path forward is to automate more and more of the research. There is just too much to do, and it is too important not to do. Incidentally, Zeroth-Order approaches lend themselves very well…