> It is possible through the sheer randomness of life for this same genetic change to occur
I think you're misunderstanding the nature of mutation - pretty much *every* individual will possess some mutations. But the typical mutation (the kind that occurs with 10^-5 probability in most multicellular species) affects only a single base pair, and corn DNA has 2.3 billion base pairs which might be mutated, and a single gene (in humans at least) will typically contain ~27,000 base pairs, with some containing as many as two million.
So let's run some numbers shall we? Assuming you want to transform a single typical-sized gene into some specific same-sized alternative you're going to need to generate 27,000 mutations in one single section of DNA (let's just call it 10^4, maybe 2/3 of the base pairs are already correct for the replacement gene), with a probability of 10^-5 per mutation, giving each individual a 10^-20 chance of having the gene spontaneously appear in one generation (the vast majority of most major gene mutations are debilitating if not fatal, which complicates the calculations dramatically, so we'll assume we have to do it in one generation). Assuming 50 million kernels of corn per acre (10^7) that means you're going to need 10^13 acres of corn to get decent odds. But that's going to be a problem because there's only 10^10 acres of land area on Earth. So we're still going to need 10^3 generations to get good odds. So, cover the entire planet with corn for thousands of years and your single desired gene has a decent chance of appearing spontaneously.
Granted, some incomplete mutations changes won't be fatal, so we can probably lob a few orders of magnitude off the problem, through multi-generational mutations but at least now we have a sense of just how unlikely it would be for natural mutation to replace just one single gene with something specific.
In the case of Roundup resistant weeds you're getting something very different: a relatively minor mutation to one or more existing genes which makes the plant more resistant, (probably via a different bio-chemical mechanism). And that's important because the more dramatic the change the greater the chance of unintended side effects. If we actually understood genetics I'd have far fewer objections to GMOs, but right now we're still at the finger-painting page - we've only just come to realize that the non-protein-coding "junk DNA" that comprises the majority of most organisms DNA is in fact functional, and we still haven't the slightest clue what most of it does. Even the simplistic repeating bits that we believed to be genetic "parasites" appear to have their role to play. To say nothing of the fact that we are still only just beginning to develop the crudest predictive understanding of biochemistry and the subtleties of cellular biology.
We're playing with fire here, and we don't have the slightest idea how to build a firetruck. We've already discovered that some of the GMOs we've created are unexpectedly toxic. And GMOs are routinely found interbreeding with their normal cousins - once the gene gets into the wild there's no stuffing the genie back in the bottle.
Personally I think Monsanto had the right idea early on: build in a genetic "kill switch" so the GMOs can't reproduce. They did it for all the wrong reasons, but the basic idea is sound: if we're going to be releasing potentially dangerous organisms into the world, especially ones which will inevitably interbreed with crops our civilization depends on, let's take at least a few decades to assess their unexpected consequences. Then if we discover it's wiping out bees or vital soil microbes or whatever we just stop producing new seed and let it go extinct. If on the other hand it proves safe after extensive real-world testing we can remove the kill switch and begin to treat it as a normal crop. It may still become an invasive organism, or interbreed with other sub-species in unexpectedly dangerous ways, but at least we won't have just released a complete unknown into the biosphere.