Except to a human familiar with the Minesweeper game, those squares are pictures of mines, albeit highly abstracted ones. This 
isn’t a literal picture of a mine, either. It’s an abstracted icon that we humans recognize as representing the concept of a “land mine” in some contexts. The blank squares which must be concealed mines given the rules of this specific puzzle and of captchas is at a higher level of abstraction, but they are still “pictures of mines”.
Yup. What Chronos said.
I’ve decided to go supernerd on this and create a program in R to generate and solve these … Wish me luck!
Let us know when they’re ready.
And thus fail the test. The meta joke is that the game involves not clicking the mines, yet this test tells you to do the opposite.
If a test to detect humans is set up such that humans, and only humans, fail the test, then the test is defective, not the humans.
But I think we’re at the point that there are enough interpretations, counter-interpretations, and meta-interpretations that the ambiguity is itself the joke.
If “humans, and only humans, fail the test”, then the test has detected humans and is not defective. 
I agree. That was my point. A test that humans would fail is a valid means for detecting humans.
Turns out algorithmically solving even a small puzzle like this is darn hard, and to check that a randomly generated one is solvable the program needs to be able to solve at least a subset of puzzles. 
Thank you for explanation. As a human unfamiliar with the Minesweeper game, I was entirely confused, both by the xkcd and by the conversation here. (I mean, it was obvious to me that there was something I wasn’t getting and that that was probably due to my ignorance of a reference, but I had no idea what the reference was to.)
If you’re ever confused by an xkcd comic, try checking https://www.explainxkcd.com/wiki/index.php/Main_Page.
That site looks like fun.
as real minefields usually do not have large numbers indicating which of the surrounding land contains mines[[ citation needed
– I tend to assume, when I don’t get xkcd, that it’s a programmer joke of some sort.
IIRC, given a pattern of numbers, determining whether there is any possible arrangement of mines consistent with that pattern of numbers is an NP-complete problem.
But I had hope that you might be able to pull it off for sufficiently-small grids.
Yeah, I initially had high hopes that with that small a grid I could do something iterative and do one clue at a time in combination with neighbors, but it turned out not to work.
I’m going to want to see some truth tables
I assume that the norx gate, when presented with a true input, returns any two same outputs, and when presented with a false input, returns any two different outputs, with which of the two possible outputs it returns being unspecified. In practice, of course, this would be easiest to implement by keeping one output permanently true, and the other just a straight feed-through of the input. But of course, in XKCD land, it’d be programmed for maximum perverseness.
Hehe
Is the alt-text referencing a specific linguist?
Wasn’t the guy really called (or rather, took the alias) Dan Cooper, there was no “D.B.Cooper”?
Ah, yup!