And then, this "smooth path" idea relates neatly with topological notions of computability.
Evolution is much the same, but most people are worse at it because they make the mistake of prioritizing human-values as evolutionary advantages for all species.
For example, cows get slaughtered by the billion by humans. But they're an evolutionary success story from a Darwinian perspective: by being easy to domesticate by early tribes and delicious, they've proliferated far beyond anything which would've happened un-aided by humans.
Of course, there are simple and trivial programs that can be specifically analyzed, but we're definitely not to that level on any of the stuff under discussion here. The analogy really is pretty apt.
But that's the nature of the problem: you know it halts because just executed it till it halts.
while(i>0) { i * 2; }
I think you would agree that this clearly never halts for any i > 0, and I think it also clear that we did not determine that by running it until it did not halt (because there is no such point).Turing proved that for some programs there is nothing we can do but run them and see. That doesn't mean that's all we can ever do.
"Surely these things would offer greater Darwinian fitness" is as basic a mistake as "Surely we can figure out whether this program will halt without running it".
EDIT: In fact, as I think about it, there is a deep connection between the thought "Knowing the genes of this organism, we may predict its fitness" and "Knowing the instructions of this program, we may predict its halting". I'd love to see a paper on that.