How about picking the appropriate tools for the job
Rather than picking a solution and attempting to shoehorn it into all the problems?
Yes, machine learning might be applicable for some tasks (although how do you train it for a moon mission?), but good old fashioned software written to do a specific job must surely be the correct choice for many, if not most things. Take the flight control software as an example : Surely this is just a variant on autopilot software.