I read Stephen Meyer’s Darwin’s Doubt a few years ago and listen to his lectures and interviews occasionally. More recently I read an influential blogger and sci fi writer Vox Day’s Probability Zero, which makes a similar probabilistic argument with more formal machinery. The argument is this. Around 540 million years ago, in a window of roughly 20 million years, essentially every animal body plan we still see today appears in the fossil record. Phyla don’t just diversify; they materialize. The supply of beneficial mutations needed to invent that many body plans in that little time is mathematically implausible. Therefore something other than ordinary neo-Darwinian evolution must be involved. Namely, Intelligent Design.
I wasn’t convinced. But I couldn’t dismiss it either. The math, as presented, wasn’t obviously wrong, and the standard biologists’ answer — “you’re calculating it incorrectly” — felt evasive. Day’s book also had Frank Tipler writing the foreword, which pulled extra weight with me. I’d read Tipler’s The Physics of Immortality years ago, the one where he derives a sort of cosmological resurrection from general relativity, and come away with the feeling that here was a real physicist doing serious math whose conclusions you couldn’t quite take at face value, but couldn’t quite dismiss either. When Tipler endorses a probabilistic argument against evolution, you at least pay attention.
I started talking to ChatGPT about it. This was maybe half a year ago, and the models were noticeably dumber than they are now. I threw darts trying to find a way around the speed of mutations. Maybe it’s epigenetic. Maybe there’s some pre-existing flexible toolkit. Maybe it has to do with how DNA gets wound on a spool, the chromatin layer. Maybe Snowball Earth filtered the survivors for plasticity. Maybe what happened next was polygenic filtering and elimination, then canalization closing the window. The model played along, sharpened some of the ideas, pushed back on others. But it never told me what the actual field was called. (Jumping ahead, evo-devo never came up. Six months later, it would have come up in the first response. The models got dramatically better in that window.)
Then I had a dinner party. I tried to express the half-formed synthesis, and it is always very difficult on the first try. A guy there chewed me up and spat me out the traditional way: nothing to see here, mutations and selection, the math is fine, you don’t understand the field. Well, I don’t. I’m a former physicist and a current software engineer. But it wasn’t an explanation; it was a dismissal in the shape of one. I was no further along.
So I came back to the LLMs very recently. This time the model named names. Davidson and Erwin and the kernel hypothesis. Carroll’s Endless Forms Most Beautiful. West-Eberhard on developmental plasticity. There was a whole field called evolutionary developmental biology — evo-devo — and it had been working on exactly these questions for thirty years. My dart-throwing had landed roughly inside it. I just hadn’t known the field existed.
The path from “the ID math looks scary” to “biology has a clean answer” is one almost nobody walks publicly. Here it is.
A note on intelligent design
Before going further: intelligent design as a metaphysical position is not what I’m arguing against. The simulation hypothesis is a form of intelligent design, and it’s taken seriously by people like Nick Bostrom, David Chalmers, and even very publicly by Elon Musk. Tipler’s Omega Point is in the same neighborhood. The question of whether reality as a whole has a designer is open, and biology can’t settle it.
What Meyer and Day are doing is narrower. They’re claiming that one specific biological event — the Cambrian explosion — could not have happened by ordinary evolutionary mechanisms, and therefore required a designer’s intervention at a specific moment in Earth history. That’s an empirical claim about biology, and biology can address it. You can take the simulation hypothesis seriously and still think Meyer is wrong about the Cambrian. This post is about the second claim, not the first.
What Meyer and Day get right
The Cambrian explosion is real. Darwin was very aware of it. Skeptics like to wave it off as a preservation artifact — biomineralization made hard parts suddenly visible — but the quantitative work doesn’t support that line. Studies of arthropod fossils show morphological evolution in the Cambrian running about four or five times faster than the long-term background rate, and most of the major body plans show up in the rocks within roughly 20 Myr. So there is something to explain.
And the naive version of the standard answer probably wouldn’t do it. If you model evolution as “wait for hundreds of new genes to drift into existence, then assemble them into a trilobite,” the probability math really is unfriendly. Day’s probabilistic critique and Meyer’s information-theoretic version are different formalisms reaching the same conclusion, and the conclusion is correct given their model of what evolution is doing.
What’s wrong is the model.
The stem cell clue
For a while I was reaching for the word epigenetics without knowing quite what I meant by it. In popular usage the word has drifted to mean something like “your grandmother’s trauma is chemically inherited” - the claim, for example, that Jews are still traumatised because the Holocaust is encoded in their cells. That version is mostly junk. The biological meaning is different and more interesting: epigenetics is not why Jews are still oppressed. It’s why stem cells develop into different tissues.
Your body contains over hundred cell types. Neurons, liver cells, bone cells, sperm, retinal cones. They look and behave nothing like each other, and they all carry identical DNA. The differences come from regulation: which genes get turned on, when, where, how strongly, in what combinations, governed by each cell’s developmental history. The genome isn’t the design. It’s the parts list, plus the rules for which parts to use where.
Take that one level up. How do you get from one ancestral animal lineage to thirty-odd phyla — sponges, mollusks, arthropods, chordates — in 20 million years? Not by inventing thirty genomes. By deploying mostly the same genes through different regulatory programs.
The toolkit was loaded long before
The decisive finding of the last thirty years of comparative genomics is that the developmental toolkit — the genes that pattern animal bodies — is shared at a depth nobody expected.
Sea anemones, sponges, even single-celled choanoflagellates (the closest unicellular relatives of animals) already carry most of the genes once thought to be bilaterian inventions. The gene that builds a fly’s eye, Pax6, also builds yours and octopuses. The Hox cluster that lays out your spine lays out a beetle’s segments. These genes were already in place hundreds of millions of years before the Cambrian, and they have stayed mostly the same ever since.
Molecular clocks place the divergence of the major bilaterian lineages somewhere between 650 and 800 million years ago. That gives 100 to 260 million years of cryptic evolution before the Cambrian even starts. The toolkit got assembled during the Neoproterozoic, in soft-bodied ancestors we can barely see in the rocks. What happened in the Cambrian’s 20 Myr wasn’t the building of the toolkit. It was the deployment of it.
This is where Day’s argument quietly comes apart. His probability calculation is for assembling new functional protein sequences. But the Cambrian doesn’t happen in protein-sequence space. It happens in regulatory configuration space — a much smaller state space with much steeper phenotypic gradients per change. Tipler’s foreword notwithstanding, statistical machinery applied to the wrong configuration space gives the wrong answer.
Snowball Earth as the filter
One more piece I guessed correctly, though working biologists treat it as additive rather than essential.
Between 720 and 635 million years ago, the planet froze twice. The Sturtian and Marinoan glaciations, “Snowball Earth,” were the most severe climate events in our planet’s history. Whatever survived had to be a generalist: stress-tolerant, metabolically flexible, able to ride out conditions no specialized organism could handle.
A bottleneck like that does two things to a population. It selects for flexibility (the rigid specialists die). And it builds up reservoirs of cryptic genetic variation, mutations that don’t normally express phenotypically but get unmasked under stress. There is a nice body of work on heat-shock proteins like Hsp90 acting as evolutionary capacitors: buffering variation under normal conditions, releasing it when the organism is stressed. That’s what the LLM confirmed to me.
So the metazoan lineages that came through the Marinoan were carrying a pre-built toolkit, broad developmental flexibility, and reservoirs of latent variation ready to express. The Ediacaran was the long fuse — oxygen rising, multicellularity getting more sophisticated, ecological space opening up. The Cambrian was the moment all of it converged.
Why so fast, then
The “20 million years is impossible” intuition assumes evolution had to invent body plans from scratch. It didn’t. It redeployed a pre-built toolkit, and the phenotypic leverage of a regulatory change is enormous. One mutation in a non-coding enhancer can move a limb, duplicate a segment, relocate an eye, change a body axis. The same kind of change to a coding sequence usually just breaks the protein.
So when fossil studies show the Cambrian running at five times the background rate of evolution, that doesn’t mean mutations were happening five times more often. Mutation rate is the same. The leverage per mutation is higher, because the mutations are landing on a sophisticated regulatory network rather than on protein-coding genes that don’t yet exist.
The physicist’s way of putting this: Day calculates the diffusion constant in protein-sequence space, but the dynamics live somewhere else. They live in regulatory-configuration space, where the landscape is steeper and the relaxation time completely different. Right particle, wrong potential.
Why it never happened again
Here is the second puzzle, the one that to me clinches the whole picture.
If the Cambrian was so productive, why hasn’t anything comparable happened in the 500+ million years since? Mass extinctions have cleared the decks repeatedly. The dinosaurs died 66 Mya and mammals had a clean shot at radiating into empty ecospace. No new phyla appeared. There has been spectacular diversification within existing body plans — primates, cetaceans, songbirds, flowering plants — but nothing topologically new.
Davidson and Erwin proposed the answer in 2006. I vaguely guessed it 20 years later. The same regulatory networks that enabled the Cambrian became evolutionarily refractory once they had locked in. Body-plan-specifying networks contain deeply recursive “kernels” with redundant inputs. Perturb one and you don’t tweak the body plan; you kill the embryo. The architecture that opened the window also closed it.
Waddington called this canalization. The pattern is general across evolutionary radiations at every taxonomic scale: disparity peaks early, diversity fills in later, design space narrows.
So the full arc is:
Cryogenian loads the substrate (toolkit, plasticity, cryptic variation)
Ediacaran is the long fuse (oxygen, multicellularity, opening ecospace)
Cambrian is the regulatory deployment, analogous to a stem cell differentiating into different tissues
Post-Cambrian is canalization closing the window — kernels lock, ecospace fills, and the same architecture that enabled new phyla now prevents them
What Meyer and Day miss
The mathematical-impossibility argument depends on a model of evolution that wasn’t really mainstream even by the 1980s: protein-coding mutation as the engine of morphological novelty. The model is wrong, and not for theological reasons. It’s wrong because biology has moved on.
Meyer is aware of evo-devo, by the way. I just missed it. Darwin’s Doubt has a chapter on developmental gene regulatory networks, and his response is to push the probability argument up one level. Fine, you say morphology is regulatory rather than coding — but the regulatory networks themselves had to be assembled, and that’s also mathematically impossible. It sounds principled. It’s really the same category error climbing the ladder. The calculation assumes random independent assembly of a specified target, which isn’t what evolution does at any level. Apply the math recursively and you keep getting small numbers. The small numbers are telling you that random independent assembly isn’t the mechanism. They aren’t telling you the mechanism is implausible.
Once you take in that morphological evolution is mostly regulatory, that regulatory evolution scales differently from coding evolution, and that the toolkit was already loaded before the Cambrian started, the math relaxes. You don’t need design. You don’t need impossible mutation rates. You need a loaded substrate, an open ecological window, and ordinary regulatory rewiring with a 5× accelerator from environmental opportunity.
The strongest arguments from the ID camp aren’t the dumb ones, and they aren’t the religiously motivated ones. They’re the ones that point at real gaps in how evolution is publicly understood. Meyer, Day, and Tipler-by-association are doing biology a backhanded favor by pointing at a real puzzle whose public answer is too often “trust us, it’s fine.” It isn’t fine to leave it there. There is a real answer, and most people don’t know it. And who knows, perhaps the pre existing toolkit still requires ID. We still don’t know exactly the origin of life itself.
The Cambrian explosion isn’t a problem for evolution. It’s evolution doing what one of your stem cells does every day: making very different things from the same instructions, by deploying them differently.