So, this means a) strong superhuman AI (good luck), b) deciding from an ambiguous input to one of possibly mistaken actions (good luck mapping all possible correct states), or c) drool-proof interface ("It looks you're trying to shut down S3, would you like some help with that?").
TL;DR: yes, but it's a cure worse than the disease.
I don't know. I was suggesting it wasn't realistic to do that, and therefore it wasn't realistic to implement a UI that prevents you making mistakes.