> If you corrupt a line of code you don’t get a different program you just get something that won’t compile.

Usually, but not always.

Add to that an astounding number of iterations (random point changes), the fact that changes are guaranteed to be ... well, the metaphor breaks down quickly, but it would be something like "syntactically valid in the language being used, and no larger or smaller than a certain size of diff", and a sort of error correcting/weighted (as opposed to absolute/binary) behavior as a result of each change, and it starts to seem a bit more workable.

But even that is barely scratching the surface of how this stuff works. Code is far from a perfect analogy.