For the foreseeable future, human-in-the-loop approaches utilizing ML seem to be necessary for productivity and efficacy. Delegating everything cognitive to a machine does not seem to acknowledge the incredible power of humans' creative faculties.
I always think about Star Trek in these scenarios. The humanoids are the strategists, who tell the computer the goal and any constraints it must work with. The computer is a tactician, who translates those high-level instructions into concrete steps and executes them.
I think things like planning and design could be assisted in limited ways, but it seems foolish to delegate that to machines as we have conceived them. But optimization of architectures, geometries, and efficient material usage would all benefit from delegating to machines, since these are concrete subtasks that are evaluated in quantifiable fashion.
