Thirty-Six Amino Acids
Let's now see what these numbers mean when applied to widely used assumptions about the hominids — the line that led from the common ancestor of chimpanzees and humans to us. How long would it take to produce something as modest as a single 36-amino-acid stretch of protein-coding DNA?
This isn't meant as a literal reconstruction of how any real protein came about. The setup here, like the Weasel program itself, still assumes one specific, predetermined target has to be hit exactly, which isn't how real evolution has to work. But the pieces going into this calculation — mutation rates, population sizes, generation times, how long a new mutation takes to spread — are standard kinds of figures, and any real evolutionary process has to contend with numbers of this general kind, even if the exact figures in a particular case would differ. So while the result shouldn't be mistaken for a measurement of anything that actually happened, the kind of arithmetic involved is unavoidable in thinking about what evolution can and cannot accomplish.
Start with the target: a stretch of DNA 108 letters long — enough to code for 36 amino acids. The full human genome runs to about 3 billion letters, which works out to roughly 28 million non-overlapping stretches of that size. For simplicity, we'll assume that any mutation landing inside the target stretch changes the amino acid its codon codes for — not strictly true in real biology, where many such changes don't alter the protein at all, but a simplification that favors evolution.
On mutation rate: each human child's DNA carries something like 100 new point mutations, at most, not inherited from either parent. (Measured figures are mostly lower, around 60 or 70; we'll use the generous number.) Across 10,000 births, that's about 1,000,000 new mutations. Spread across roughly 28 million stretches of this size, any one particular 108-letter stretch can expect a hit only about once every 280,000 births.
Suppose that hit occurs — one single child is now born with a point mutation somewhere in that particular stretch. How long before it spreads widely enough that a second, independent mutation in that same stretch has a real chance of landing on top of it?
Assume the hominid line has just split from the chimpanzee line, with a population of 100,000, and that the time from birth to a first child is about 10 years — short for an ape, and so generous again. That's about ten thousand births a year, and one hit to our particular stretch about every twenty-eight years. But the next hit will land in some random member of the population, not in a descendant of the first mutant, unless the first mutation has already spread widely. Working out the actual rate of spread gets complicated fast, and isn't worth pursuing in full detail here. Instead, we'll set every assumption as favorably as possible. J. B. S. Haldane, one of the founders of population genetics, estimated that a population can replace one gene with a better version only about once every three hundred generations — and that's for a change already conferring a real advantage. The single point mutation being discussed here almost certainly isn't beneficial yet; if this is happening in what would otherwise be called junk DNA, it has no reason to be. Without a population that's rapidly expanding, there's no reason for a mutation like this to get any free ride toward higher numbers. It has to spread slowly even under the best of circumstances, and here there's no benefit at all, a population that isn't exploding, and ordinary competition for resources the whole time.
So, generously, we'll assume the whole cycle — spreading from one individual to a meaningful share of the population, and then waiting for a second mutation to land in that very same stretch — takes about 200 years. That's twenty generations, against Haldane's three hundred for a change that actually helps.
But based on the simulation, that second mutation will almost certainly not be the one that brings the stretch any closer to a finished 36-amino-acid target. For a target this size, the simulation takes, on average, about 2,800 separate changes to land on the fully correct result. Multiply that out: 2,800 changes × 200 years each = 560,000 years. At 10 years per generation, that's 56,000 generations — for one single 36-amino-acid stretch of coding DNA, using generous assumptions at every step. (The arithmetic: One Small Stretch.)
Fifty-six thousand generations is, again, considerably more than the forty-odd generations the Weasel program needed.
In theory, real evolution may not need nearly this much time to modify a 108-letter stretch of DNA through point mutation alone. It doesn't necessarily need to land on one exact sequence — several different amino acids may work perfectly well at a given position, not just one. And the change doesn't have to happen in this one particular stretch at all — mutations are occurring across countless stretches of the genome at once; something useful could just as easily arise somewhere else entirely.
But one thing is certain: none of this behaves anything like the Weasel program. In Weasel, many copies mutate every single generation, one generation after another, always checked against the same fixed answer. Nor does every generation in the real world reliably produce one standout mutation that beats all the others — most mutations are neutral, or outright harmful. In the real world, there is no "best mutation wins" scenario, in which the single best result is chosen immediately, prevails completely, and produces the entire next round on its own, while everything else vanishes. In reality, any promising mutation is competing against a great many others at the same time, and the spread of even a genuinely good one doesn't happen quickly at all.
This exercise has its own weaknesses — it isn't trying to model natural selection at work. What it does show is that the path from a single mutation to a finished result is never a straight line, and takes far more steps than forty, even when the process is explicitly steered toward one predetermined outcome. Real evolution isn't steered toward any particular result, but it still has to pass through the same kind of process this exercise goes through. Wherever it eventually ends up, there must be wasted attempts, wrong turns, and dead ends along the way. A random walk through a sample space — the kind of walk that feeds natural selection — is blind to past, present, and future alike. It cannot travel in a straight line toward any final goal. Selection can only work with whatever the walk happens to hand it, and most of the time, what it's handed is nothing useful at all.
In the end, if every single change takes time, and spreading that change through a population takes still more time, evolution cannot be in a hurry. It has to move along the same convoluted, chance-driven path this exercise walks through by hand. And the resulting numbers cannot come out small. They have to land in the same general range as the numbers produced here — not identical in their specifics, but not different in kind either.
Last revised 2026-11-19