Well, that depends on what we mean by "randomly selects", doesn't it? There's no planning ahead by the evolutionary process - it doesn't "store up" traits on the theory that they might be useful someday. It's all about what's adaptive right now. If mutation is random, then adaptive/dysfunctional/neutral traits will land at random (well, not exactly - dysfunctional mutations are far more common than beneficial ones, but anyway) among creatures, and therefore the creatures that are better adapted are essentially "chosen" at random. Could be robins that inherit a mutation for slightly better eyes, or it could be blue jays just as easily. The only pattern that exists is that creatures in competitive environments that are better adapted tend to survive and flourish better than creatures that are less well adapted. Which particular creatures those are, are essentially "chosen" at random. If it's robins, it's robins, but it could have just as easily been the blue jays.
Of course, some traits will prove to be beneficial in a wide range of environments. Bears survive and flourish by being generalists - they can tolerate a wide range of environments, and can subsist on a wide range of foods, which allows them to adapt to environments as radically different as the jungles of Malaysia and the Arctic circle. But that trait isn't adaptive in all environments - there used to be bears in Africa, but not any more. The competition was too fierce, and all available niches were occupied by specialists. As generalists, that put bears as a competitive disadvantage - no matter where they turned, some more specialized creature had an advantage in that particular niche. They aren't superpredators the way the big cats of Africa are, nor are they as good at finding plant foods as the large African herbivores are. So they died out, because the traits that made them successful elsewhere left them at a disadvantage in Africa.
And not to put too fine of a point on it, but having a large brain is the sort of generally adaptive trait that leads to success in a wide range of environments ;)
So, why not say that in our literary model, some randomly generated texts are--take your pick, either more intelligible, more beautiful, more popular with theater-goers, more profound.
The reason I wanted to avoid things like "beauty" as a criterion is that beauty will tend to be in the eye of the beholder. Much as I like "Hamlet", I prefer "Richard III" or "Macbeth", personally ("Life is a tale told by an idiot, full of sound and fury, signifying nothing"). But if we can come up with some criteria where there's no question about who's a winner and who's a loser, we can make that work instead.
Isn't this the same game as going from "good" to "evil" in one-letter moves, but infinitely harder, no matter how many millions of years intervene? Is this story really believable?
The problem I have is that you're still inserting this requirement that every intermediate step be as "sensible" as the starting and ending products. Unless you can persuade me that this requirement is somehow analogous to how natural selection operates in populations, I don't see the necessity of putting it in there at all. There is no similarly analogous requirement in nature, only that the intermediate form be more fit than its predecessor - "fitness", again, being defined as distance from the end product in this case. If that's how we define fitness as our criteria for selection, what does it give us to add some other criterion, particularly when it appears to me to be completely artificial?