Chapter 6: The Monkeys That Didn't
Writing the Library of Congress
The absurdity of the claim becomes obvious the moment you run the same method on something else. So let's run it on the entire Library of Congress.
The Library of Congress holds some forty-one million books and other printed volumes. Using rough figures — a few hundred pages a book, a couple thousand letters a page — that's something in the neighborhood of twenty-five trillion letters, total, across every book in the collection. In ordinary English text, the letter E shows up more often than any other letter, at a rate of a little over twelve percent. So the entire library needs, in round numbers, around three trillion E's to exist somewhere within it.
Now have a computer generate letters completely at random — each of the twenty-six letters equally likely, about one time in twenty-six. Keep generating until you've produced enough E's to match that three-trillion target. At that rate, hitting the E target takes roughly eighty trillion random letters overall. Since E is the most common letter in English text but gets no special treatment from a machine generating letters uniformly at random, by the time you've generated enough E's to cover the library's need, you will, automatically and without even trying, have generated more than enough of every other letter too. T, A, O, all the rest — every one of them appears less often than E in real English, so every one of them will have cleared its own, smaller quota well before E does.
Run that on a single ordinary computer, at a million letters a second, and it would take about two and a half years. One fast modern machine, at a billion letters a second, would finish in less than a day — far quicker than Anderson's own project took him. (The arithmetic: Writing the Library of Congress.) By Anderson's own logic, you will have "written" every book in the Library of Congress. Hand me the collection in digital form, and I'll happily confirm that every letter output along the way actually exists somewhere in the massive pile of books currently sitting in the library. Come to think of it, if I let this run long enough, I'll have "written" every book that ever was or ever will be. Not bad for a simple computer program.
I hope you're looking at me right now with some mixture of incredulity, astonishment, and plain disbelief.
Good. Nobody would accept that as having written the Library of Congress. It's a pile of randomly ordered letters that happens to contain, buried and scattered throughout it, every letter any of those books would ever need — in numbers matching real English usage, purely because we deliberately arranged for that to be true. Calling that "writing the library" is absurd on its face. And it is, exactly and precisely, the same claim the BBC ran under the headline "Virtual monkeys write Shakespeare."
Because what Anderson did isn't meaningfully different from what I just did. He simply worked with larger pieces at a time. At no point did his monkeys put even two nine-letter fragments in the right order; whatever order there was came entirely from the copy of Shakespeare they were being checked against. It never produced Hamlet, or even a single connected line of Hamlet. What it produced was nine-letter combinations, generated in random order, checked one by one against Shakespeare's collected works to see which of them happened to show up somewhere inside it. That doesn't resemble the writing of anything, except nine-letter strings — which any computer can produce at random with a trivial program.
This isn't a demonstration of chance producing something extraordinary. It's a demonstration that computers are fast. You would get the identical result — every single checklist item found, every time — by simply having the computer generate all 5.4 trillion possible nine-letter combinations in plain alphabetical order, with no randomness involved at all. (The count: Nine at a Time.) Nothing about doing it randomly added anything. The outcome was never in doubt.
Last revised 2026-11-17