My own take on this is that we need to develop strongly declarative languages. That way there is no information about implementation to get wrong. You specify t

My own take on this is that we need to develop strongly declarative languages. That way there is no information about implementation to get wrong. You specify the "what" not the "how". You still might get the "what" of it wrong but this is no different than the mistake of "proving" the wrong thing. Eventually there will be strong AI and we will be able to pass off the "what" of it to the machines as well.