In my experimentation with meta-genetic algorithms to optimise a Pareto set, be careful that I don't end up with suboptimality in other areas not under consideration.
A lot of my work is based on some research I did in making an artificial intelligence for anti-missile defences over 15 years ago. I guess I was lucky, inasmuch as I used a rules-based system and not a neural net.
On the other hand, the fault-tolerance and self-healing properties I put into FedSat, and helped the MESSENGER spacecraft team with, have proven to be useful in space exploration. I skipped over Terminator-I and went straight to Terminator-II, or as I once wrote, skipped the Battleship approach, armoured but brittle, in favour of "Beware of the Blob".