Chapter 7: Junk DNA and a Demolition Tool
Word Ladders
There's a language version of the asymmetry point mutations show in working genes, and while it isn't the same phenomenon as what's been shown about proteins — language has no chemistry, no folding, no metabolism to sustain — it's close enough in shape to be worth borrowing as a way of thinking about it.
Take a coherent English sentence — say, twelve words long — and change one letter at random, anywhere in it. The overwhelming likelihood is that you've just broken it: a misspelled word, a word turned into a different, wrong word, or no word at all. Or put it another way: take any sentence in front of you right now, and pick any letter in it. Of the other twenty-five letters of the alphabet, how many could take its place without doing harm, and how many would break the meaning? Now ask the reverse question: starting from twelve words of gibberish, change one letter at random, and what are the odds you've just produced a coherent sentence? Nobody needs to run the numbers to feel the answer in their gut, but the numbers back the intuition up anyway. English has a few hundred thousand words, and a twelve-word sentence occupies one exact point in a space of possible word combinations so large it dwarfs anything a single random step could plausibly land on. Breaking a working sentence takes one careless letter. Building one from noise takes something close to a miracle. The direction of the asymmetry, if nothing else, is exactly the one this book keeps finding in biology.
But there's a more precise version of this idea, one that gets closer to the actual mechanics of point mutation rather than just the before-and-after snapshot — a simple word game that turns out to have a good deal to say about it.
There's an old word game, sometimes called a word ladder, that some readers will already know and some won't. The rules are simple: start with one word, change one letter at a time, and after each change the result has to be a real word too. The goal is to work your way, one valid word at a time, from a starting word to some target word. "COLD" to "WARM" is the classic textbook example, solved one link at a time — COLD, CORD, CARD, WARD, WARM.
We're going to play a version of this game at random, and it's going to give us some further insight into the kinds of obstacles evolution actually faces. To be clear, no claim is being made that this game, or the numbers that come out of it, directly tells us anything about evolution. If nothing else, it produces numbers, and it gives us something real to think about.
Here's how I set it up. I downloaded a list of 12,920 five-letter English words — I have no particular reason to think this list is complete, exhaustive, or even accurate, but it's a large, ordinary sample. I then sorted every word into "islands": a word belongs to an island together with every other word reachable from it by the word-ladder rule, one letter changed at a time, with every intermediate step also required to be a word on the list. Altogether, this produced 990 separate islands. A great many of them hold only a single word, with no neighbors reachable at all. The largest island holds 11,651 words. The second-largest holds just 11. After that, it's a steady drop straight down to islands of one.
This raises an obvious question, and the best answer is: nobody knows for sure. Do proteins work the same way — organized into isolated islands with no bridge from one to another — or is the space of functional proteins fully connected, so that, given enough time, anything could in principle mutate into anything else? If bridges always exist, one protein could theoretically transform into an entirely different one, however long it took. If they don't, a protein could only ever mutate into something else sitting on its own island — figuratively speaking, since most of the proteins on that hypothetical island might not actually exist yet.
The game itself works like this. The computer picks two random words. If they land on the same island, the game proceeds; if not, it tries again with two new words. Then, one letter and one position are chosen at random. If placing that letter in that position produces a real word that exists on the current island, the substitution is made. If not, the attempt is discarded and another random letter-and-position pair is tried.
In a loose sense, this mimics randomness paired with selection: the "mutations" are generated at random, and "selection" consists of nothing more than requiring the result to be a real word. A hit is treated as beneficial and kept; a miss simply doesn't count, and the game tries again. This shouldn't be taken too literally — there are important differences between this and actual biology — but it's worth seeing what happens.
Out of 100 random word pairs, 17 landed on different islands and the attempt was abandoned; the remaining 83 pairs landed on the same island — and in every one of those 83 cases, that island turned out to be the main one. On average, it took 1,005 randomly chosen letters to get from one word to the other. All but three of the 83 successful pairs could be solved in five steps (the minimum possible between two five-letter words that differ in every position); three needed six. The quickest success was CLOCK to PILED, solved after only 47 letters tried. The slowest was MOHUA to CHERE, which took 7,422 letters tried before making it to the end — though a shorter, six-step route existed and simply wasn't the one chance stumbled onto. Along the way to that result, 112 of the attempted letters happened to land on other real words entirely, off the direct path to the actual target, on a longer path that eventually made it.
These results are for five-letter words. Try the same game with something much larger — pick any twenty-letter string, in any language, that makes sense, and try to walk it, one letter at a time, into some other twenty-letter string that also makes sense, never once passing through a stretch of nonsense along the way. You'll probably be able to nudge small parts of it here and there. But actually completing the walk — changing every position, arriving at a wholly different sensible string, and never once leaving the island of "makes sense" along the way — starts to look extremely difficult, quite possibly impossible in practice, for strings of any real size.
Moreover, one condition matters a great deal: these results only count paths that never leave the island — every single intermediate step has to be a real word too. That restriction is exactly what guarantees the game can be won eventually. Loosen it even slightly — allow a step or two off the island, into a nonsense combination of letters — and there's no guarantee of ever finding a way back onto any island, let alone the one holding the target word.
Who's to say biology has an easier time of it? Remember, in any language there are very many more ways to break a sentence than there are ways to change it and have it remain viable. The data earlier in this chapter suggests biology behaves the same way: a small number of mutations tends to destroy a functional protein, not improve it, or even keep it stable. So it's time for a reality check — an experiment that gives us more data to mull over.
Last revised 2026-11-19