This still hardly seems obvious.

How soon? I don't know of any software I've written for which that could be done. The first program I ever wrote was an implementation of Tic-Tac-Toe, and I can't imagine a set of parameters to produce that which aren't specifically for making games of that nature while also being easier to select than writing the code myself. I'd bet such a system is impossible in general, and that that should be the obvious conclusion.

Now, machine learning to find bugs or adjust variables... yeah. I can see that.

Assuming you could concretely nail down all the vague terms you just used, I'd be willing to bet real money we won't have anything remotely like that within the next 10 years :)

Right now source control applications don’t even detect simple merge conflicts based on meaning.

For example if I remove a global variable from a source file and another branch adds another reference to it, if it does not occur on the same line git will happily merge the file with no conflict.

That is already how you program Prolog, no proof reading or machine learning required. I also think that specifying the problem is one of the harder aspects of software development.

On the contrary, I think machine learning might be more immediately useful for things like low-level optimization.