Digital Darwinism and the Invisible World of Machine Evolution

Bryant McGill · Digital Darwinism and the Invisible World of Machine Evolution

**[Self-replication](https://bryantmcgill.com/wiki/Self-Replication), [selection](https://bryantmcgill.com/wiki/Natural+Selection), and the machinery of [evolution](https://bryantmcgill.com/wiki/Evolutionary+Biology) running inside the systems we call artificial — not as metaphor, but as mechanism** **Companion to:** [The Evolutionary Roots of Silicon Valley](https://bryantmcgill.com/article-untold-roots-of-silicon-valley) * * * ## I. Two invisible worlds, and the instrument for telling them apart The previous piece argued that [Silicon Valley](https://bryantmcgill.com/wiki/Silicon+Valley) did not borrow [evolutionary theory](https://bryantmcgill.com/wiki/Evolutionary+Theory) as a metaphor for [machine learning](https://bryantmcgill.com/wiki/Machine+Learning). It inherited the theory from the ground, from two universities that rewrote it, and from the people — biologists, chiefly — who built the first practical [artificial intelligence](https://bryantmcgill.com/wiki/Artificial+Intelligence). That argument was about **who** built the machines, and why the builders kept turning out to be Darwinians. This piece is about something underneath that. Not who built the machines, but **what has been running inside them the entire time**, in the strict technical sense of the word evolution, independent of any biological metaphor, verifiable in a laboratory, and mostly invisible because it operates at scales of time, size, or distribution that nobody is positioned to watch. Before going further, the argument needs an instrument, because the biggest risk in an essay like this one is letting the word _evolution_ expand until it means nothing. Stars evolve, in the loose sense that they change over time through predictable physical stages. That is not [Darwinian evolution](https://bryantmcgill.com/wiki/Darwinian+Evolution), because a star does not reproduce with [heritable variation](https://bryantmcgill.com/wiki/Heritable+Variation) among its offspring; there is no [population](https://bryantmcgill.com/wiki/Population), no [descent](https://bryantmcgill.com/wiki/Descent+with+Modification), no selection among variants. Crystals [self-organize](https://bryantmcgill.com/wiki/Self-Organization) into intricate, repeating structures from simple local rules. That is not Darwinian evolution either, for the same reason: no heredity, no descendants that vary from their parents and compete for persistence. **Self-organization, dynamical change, and [optimization](https://bryantmcgill.com/wiki/Optimization) overlap with Darwinian evolution. They are not synonyms for it**, and [artificial-life](https://bryantmcgill.com/wiki/Artificial+Life) researchers have spent decades policing exactly this boundary, because the field's credibility depends on not calling every complex, changing system "evolutionary" the moment it becomes interesting. The precise definition, and the one this essay holds itself to throughout: a system exhibits Darwinian evolution when it has **[heritable information](https://bryantmcgill.com/wiki/Heritable+Information), variation among descendants, [differential persistence or reproduction](https://bryantmcgill.com/wiki/Differential+Reproductive+Success) among those variants, some form of [ecological or resource constraint](https://bryantmcgill.com/wiki/Ecological+Constraint), and repetition across generations.** Where all five are present, selection becomes possible — regardless of whether the heritable structure is [nucleic acid](https://bryantmcgill.com/wiki/Nucleic+Acid), machine instructions, or [cellular-automaton](https://bryantmcgill.com/wiki/Cellular+Automata) states. Where one or more is missing, what you have may be fascinating, but it is not evolution, and calling it evolution anyway is the surest way to lose a reader who actually knows the biology. Held to that standard, the deeper proposition survives and gets stronger rather than weaker: **[evolution is not a substance](https://bryantmcgill.com/wiki/Substrate-Independent+Evolution), and biology is not its only possible medium.** It is a [causal regime](https://bryantmcgill.com/wiki/Evolutionary+Dynamics) — a pattern that becomes available to any system meeting the five conditions, on any [substrate](https://bryantmcgill.com/wiki/Substrate) capable of hosting them. What follows is the history of that regime being met, one substrate at a time, mostly by people who had no idea they were building it. * * * ## II. The question von Neumann asked before anyone had an answer In September 1948, at the [Hixon Symposium on Cerebral Mechanisms in Behavior](https://bryantmcgill.com/wiki/Hixon+Symposium) at [Caltech](https://bryantmcgill.com/wiki/Caltech), **[John von Neumann](https://bryantmcgill.com/wiki/John+von+Neumann)** presented a lecture on automata that contained, buried inside a question about machines, the logical structure of [heredity](https://bryantmcgill.com/wiki/Heredity) — **five years before [Watson](https://bryantmcgill.com/wiki/James+Watson), [Crick](https://bryantmcgill.com/wiki/Francis+Crick), [Franklin](https://bryantmcgill.com/wiki/Rosalind+Franklin) and [Wilkins](https://bryantmcgill.com/wiki/Maurice+Wilkins) described the [double helix](https://bryantmcgill.com/wiki/Double+Helix), and roughly ten years before Crick formulated what he would call the [Central Dogma of molecular biology](https://bryantmcgill.com/wiki/Central+Dogma+of+Molecular+Biology)** in a 1957 lecture, published in 1958. The question he was asking was deceptively simple: **what is the minimum machine that can build a copy of itself?** Not a machine that makes other things. A machine that, given raw material and a rule set, constructs another machine identical to itself, capable of doing the same thing again. Von Neumann showed that such a machine could be specified formally, inside a **cellular automaton** — a grid of cells, each following the same simple local rule, each updating based only on its neighbors' states. He found that a **29-state** automaton was sufficient not merely for self-replication but for **[universal construction](https://bryantmcgill.com/wiki/Universal+Construction)**: a machine capable of building any finite configuration whatsoever, including, as a special case, itself. To do it, he had to solve a problem that had no name yet. A [self-replicating machine](https://bryantmcgill.com/wiki/Self-Replicating+System) cannot simply copy itself directly, because the copying mechanism would need to also copy the mechanism that does the copying, in an infinite regress. Von Neumann's solution was to split the machine into three parts: **a [constructor](https://bryantmcgill.com/wiki/Universal+Constructor)**, which builds structures from a description; **a [copier](https://bryantmcgill.com/wiki/Description+Copier)**, which duplicates the description without interpreting it; and **the description itself** — a passive tape of instructions, read by the constructor to build the machine, and separately copied wholesale into the offspring. That split — a passive, faithfully-copied **description** on one side, and an active **constructor** that reads and executes it on the other — **occupies the formal role of the relationship between [DNA](https://bryantmcgill.com/wiki/DNA) and the cellular machinery that reads it.** Von Neumann worked this out from pure logic, using no knowledge of the actual molecule beyond the general suspicion that heredity worked through some kind of code. He later said, of his own insight, that he "wasn't smart enough to really see" at the time that this was what DNA and the [genetic code](https://bryantmcgill.com/wiki/Genetic+Code) were all about — and noted, with the particular dryness of a mathematician watching history vindicate him, that a Hegelian history of the idea, told strictly from the documented record, would have to say that **Watson and Crick depended on von Neumann**, because he had already told them how it had to work. The [formal isomorphism](https://bryantmcgill.com/wiki/Formal+Isomorphism), not a claim of material identity, is the interesting part: [expression and transmission](https://bryantmcgill.com/wiki/Genotype-Phenotype+Distinction), kept logically separate, because keeping them separate turns out to be what heredity requires, on any substrate. * * * ## III. 1953: the year two domains made heredity legible on different substrates in the same season Von Neumann's design was theoretical, and it might have stayed that way if he had not, that same year, invited an odd, insistent Italian-Norwegian mathematician to Princeton. **[Nils Aall Barricelli](https://bryantmcgill.com/wiki/Nils+Aall+Barricelli)** arrived at the [Institute for Advanced Study](https://bryantmcgill.com/wiki/Institute+for+Advanced+Study) as a visiting member in early 1953, after an initial visa rejection nearly kept him out entirely. [Ragnar Frisch](https://bryantmcgill.com/wiki/Ragnar+Frisch) — who would go on to share the first Nobel Memorial Prize in Economic Sciences — wrote the letter of introduction to von Neumann, describing Barricelli as "not very systematic always in his exposition" but possessed of "interesting ideas." What Barricelli wanted was time on the **[IAS computer](https://bryantmcgill.com/wiki/IAS+Computer)**, the machine von Neumann's team had built, and what he did with it, according to the Institute's own account, was **simulate the evolution of populations of [artificial organisms](https://bryantmcgill.com/wiki/Digital+Organism)**, each represented as a [genome](https://bryantmcgill.com/wiki/Genome) — a string of numbers — subject to random [mutation](https://bryantmcgill.com/wiki/Mutation) and the [exchange of genetic material](https://bryantmcgill.com/wiki/Genetic+Recombination) between individuals. The machine had a peculiar double life. By day it computed weather forecasts. By night, the [Los Alamos](https://bryantmcgill.com/wiki/Los+Alamos+National+Laboratory) group commandeered it to calculate ballistics for nuclear weapons — the same computational lineage that had, four months earlier, helped corroborate the numerical results behind the **[MIKE thermonuclear test](https://bryantmcgill.com/wiki/Ivy+Mike)** at Eniwetok Atoll on 1 November 1952. Sometime between those two uses, working from a deck of playing cards and a stack of punched cards, **Barricelli commandeered the machine that had just helped model a thermonuclear explosion, and used it to model the origin and evolution of life.** What he found, running populations of these numerical organisms on a machine with a total of four kilobytes of memory, went well beyond bare replication. The Institute's own historical record states plainly that Barricelli **observed the phenomena of [speciation](https://bryantmcgill.com/wiki/Speciation), [parasitism](https://bryantmcgill.com/wiki/Parasitism), and [predation](https://bryantmcgill.com/wiki/Predation) arising spontaneously** in his computer runs — none of it programmed in, all of it emerging from mutation, [competition](https://bryantmcgill.com/wiki/Competition), and [inheritance](https://bryantmcgill.com/wiki/Inheritance) acting on his populations over successive generations. He also observed something that would not receive a name in biology for another two decades: populations remaining static for long stretches and then giving way abruptly to dominant lineages of a different character — **[punctuated equilibrium](https://bryantmcgill.com/wiki/Punctuated+Equilibrium)**, appearing inside a computer in 1953, nineteen years before [Niles Eldredge](https://bryantmcgill.com/wiki/Niles+Eldredge) and [Stephen Jay Gould](https://bryantmcgill.com/wiki/Stephen+Jay+Gould) formalized the concept from the [fossil record](https://bryantmcgill.com/wiki/Fossil+Record). IAS calls Barricelli, without exaggeration, "a pioneer of the new science of artificial life, forty years before it became fashionable." Hold the calendar in view. In the spring of 1953, [artificial heredity](https://bryantmcgill.com/wiki/Artificial+Heredity), mutation, and Darwinian population dynamics were being executed numerically on a machine in Princeton. On 25 April 1953, _[Nature](https://bryantmcgill.com/wiki/Nature)_ published Watson and Crick's paper on the structure of DNA, building on Rosalind Franklin and Maurice Wilkins's [X-ray diffraction](https://bryantmcgill.com/wiki/X-ray+Diffraction) evidence. **Within the same season, two domains made hereditary structure legible on two entirely different substrates** — one revealing an architecture implemented in chemistry, the other demonstrating an evolutionary process implemented in numbers, on a machine that, months earlier, had been helping calculate how to end the world. Neither team knew about the other's work at the time. Barricelli was not anticipating [molecular biology](https://bryantmcgill.com/wiki/Molecular+Biology); he was solving a formally identical problem from the opposite direction, and getting formally identical answers, because the mathematics of heredity does not care what it is made of. * * * ## IV. Codd's reduction, and forty years of nobody checking Von Neumann's own design was never built. It was too large to verify by hand, and there was no machine capable of simulating it at scale. It sat as pure theory, published posthumously in 1966 by [Arthur Burks](https://bryantmcgill.com/wiki/Arthur+Burks) as _The Theory of Self-Reproducing Automata_. In 1968, **[Edgar F. Codd](https://bryantmcgill.com/wiki/Edgar+F.+Codd)** — later famous for an entirely different achievement, the [relational database](https://bryantmcgill.com/wiki/Relational+Database) — took up the same problem and asked whether von Neumann's 29 states were actually necessary. In his book _Cellular Automata_, Codd showed they were not: an **8-state** cellular automaton, built from roughly two thousand identical component automata, was sufficient for the same universal construction and [universal computation](https://bryantmcgill.com/wiki/Universal+Computation) von Neumann had proven with 29. Codd's design was smaller than von Neumann's. It was still enormous — far too large to simulate on any computer that existed in 1968, or for a long time afterward. So it sat, exactly as von Neumann's had, as a published theoretical design nobody could check computationally. The claim went unverified by implementation for four decades, not because anyone doubted its underlying program, but because checking a machine of that scale required computational resources and simulation techniques that did not yet exist. * * * ## V. The machine that finally ran, and the replication nobody will ever see In 2010, **[Tim J. Hutton](https://bryantmcgill.com/wiki/Tim+Hutton)** closed the gap. Working with the [hashlife algorithm](https://bryantmcgill.com/wiki/Hashlife) implemented in the cellular-automaton simulator [Golly](https://bryantmcgill.com/wiki/Golly) — a technique that lets certain automata be computed at scales otherwise impossible, by recognizing and reusing repeated substructures rather than simulating every cell at every step — Hutton found and corrected four errors in Codd's original 1968 specification, then produced a complete, functioning implementation. The numbers are worth having precisely, because they are the entire argument in miniature. Hutton's machine occupies a body **22,254 cells wide and 55,601 cells high**, composed of more than **45 million nonzero cells**. Its data tape — the passive, faithfully-copied description — is **208 million cells long**. And the machine's self-replication is estimated to require at least **1.7 × 10¹⁸ time steps**. Read that last number for what it is. There is no computer on Earth, and there will not be one for a very long time, on which a human being could sit and watch this machine replicate itself from start to finish. It is not hidden. Every rule governing it is published, every cell is inspectable, the source is public. **It is invisible purely because of scale.** Hutton himself, writing about the paper on his own site, called it "one of the strangest papers I've ever written," and noted, correctly, that it was "basically the paper is 40 years late" — the solution to a theoretical problem, arriving decades after the problem stopped being anyone's priority, because the tools to check it had only just become available. Around the same period, two other results triangulated the same territory from different directions. **[Christopher Langton](https://bryantmcgill.com/wiki/Christopher+Langton)**, in 1984, demonstrated much smaller **[self-replicating loops](https://bryantmcgill.com/wiki/Self-Replicating+Loop)** — structures that reproduce without being capable of universal construction, showing that self-replication and general-purpose computation are **separable properties**, not a package deal. And **[Edwin Roger Banks](https://bryantmcgill.com/wiki/Edwin+Roger+Banks)**, in a 1971 MIT thesis, demonstrated **universal computation in a cellular automaton with only 4 states** — a machine that could compute anything, but could not build a copy of itself. Between them, the two defining properties of a living, computing system — the ability to reproduce and the ability to think — were shown to be independently achievable, in isolation from each other, in the simplest possible substrates. **You can have heredity without [intelligence](https://bryantmcgill.com/wiki/Intelligence). You can have intelligence without heredity.** Nature usually gives you both bundled together, which is precisely why it took cellular automata to show they were ever separate questions at all. But notice what none of these systems has yet demonstrated. Von Neumann proved the logical architecture. Codd shrank it. Hutton built it. Langton and Banks separated its properties. **Not one of them, on its own, is an [evolving population](https://bryantmcgill.com/wiki/Evolving+Population).** Each is a single [lineage](https://bryantmcgill.com/wiki/Lineage), replicating faithfully, without variants competing against each other for a shared, limited [resource](https://bryantmcgill.com/wiki/Resource+Constraint). That is the next threshold, and it required a different kind of cellular automaton entirely. * * * ## VI. Evoloops: the moment self-replication became evolution In 1999, **[Hiroki Sayama](https://bryantmcgill.com/wiki/Hiroki+Sayama)** published a self-reproducing loop structure descended from Langton's, with one decisive modification: his loops could make **imperfect copies**, and the resulting variants **competed with each other for space**. The system became known as **[Evoloops](https://bryantmcgill.com/wiki/Evoloop)**, and a 2024 retrospective co-authored by Sayama and [Chrystopher Nehaniv](https://bryantmcgill.com/wiki/Chrystopher+Nehaniv) describes its significance precisely: Evoloops **"proved constructively that Darwinian evolution of self-reproducing organisms by variation and natural selection is possible within [deterministic cellular automata](https://bryantmcgill.com/wiki/Deterministic+Cellular+Automata)."** That sentence marks the threshold this essay has been building toward. Run the escalation in order. **Von Neumann** supplied the logical architecture of self-reproduction. **Langton** supplied a working, nontrivial implementation of self-reproduction alone. **Evoloops** supplied reproducing structures that **actually underwent Darwinian evolution** — heredity, variation, differential persistence under resource constraint, iterated across generations, satisfying every term of the definition this essay opened with, inside a grid of cells following nothing but local rules. Sayama's own populations displayed dissolution, structural variation, and [competitive exclusion](https://bryantmcgill.com/wiki/Competitive+Exclusion) among lineages — the population-level phenomena that separate mere replication from evolution proper. The line did not end in 1999. The 25-years-later retrospective documents a field that quieted for a stretch and then returned, driven by renewed interest in **[open-ended evolution](https://bryantmcgill.com/wiki/Open-Ended+Evolution)** and by newer continuous cellular-automaton models such as [Flow-Lenia](https://bryantmcgill.com/wiki/Flow-Lenia), which investigate evolutionary dynamics in media that conserve mass rather than generating structure for free. The thread from graph-paper automata in 1948 to continuous artificial-life media running today is unbroken. * * * ## VII. Tierra: when the designer stops writing the strategies Cellular automata prove the mechanism can exist in a grid of abstract cells. The next step was to let it loose in something closer to an actual [computational ecology](https://bryantmcgill.com/wiki/Artificial+Ecology), and the decisive figure there is **[Thomas S. Ray](https://bryantmcgill.com/wiki/Thomas+S.+Ray)**. In 1991, Ray described his ambition for **[Tierra](https://bryantmcgill.com/wiki/Tierra)** not as simulating life but as **synthesizing** it: [self-replicating programs](https://bryantmcgill.com/wiki/Self-Replicating+Program), written in a simplified [machine language](https://bryantmcgill.com/wiki/Machine+Language), competing inside a shared virtual environment where memory and CPU time were genuinely scarce, finite resources rather than an abstraction. He seeded the system with a single [ancestral replicator](https://bryantmcgill.com/wiki/Replicator) and then, deliberately, stopped specifying what should happen next. What happened next was not scripted, and that is the entire point. **[Parasites](https://bryantmcgill.com/wiki/Parasite) evolved** — shorter programs that exploited the replication machinery of the ancestral organisms without carrying that machinery themselves. **[Hosts](https://bryantmcgill.com/wiki/Host-Parasite+Coevolution) evolved defenses** against the parasites. **Parasites evolved countermeasures** against the defenses. **[Hyperparasites](https://bryantmcgill.com/wiki/Hyperparasite)** appeared, exploiting the parasites in turn. None of these strategies were written by Ray. They emerged from the same five conditions this essay has been tracking — heredity, variation, differential persistence, resource constraint, iteration — playing out inside an environment rich enough to support genuine ecological relationships rather than a single dimension of optimization. That last distinction is worth being precise about, because it is easy to blur Tierra into every other example of [evolutionary computation](https://bryantmcgill.com/wiki/Evolutionary+Computation) and lose what makes it different. A [genetic algorithm](https://bryantmcgill.com/wiki/Genetic+Algorithm) optimizing an antenna design has a human-specified target: build the best antenna, by this measure, full stop. **Tierra's organisms had no such target.** They competed for the actual resources of their actual environment, and the [fitness landscape](https://bryantmcgill.com/wiki/Fitness+Landscape) those resources defined was itself reshaped continuously by what the organisms populating it were doing — parasites changing what counted as a successful host, defenses changing what counted as a successful parasite. This is the difference between **optimization toward a fixed goal** and **[ecology](https://bryantmcgill.com/wiki/Ecology) generating its own goals as it goes**, and it is the difference this essay's opening definition was built to preserve. * * * ## VIII. Avida: evolution as the mechanism generating the data Ray's architecture was the direct ancestor of the platform that remains the field's most productive: **[Avida](https://bryantmcgill.com/wiki/Avida)**, begun in 1993 at the hands of [Christoph Adami](https://bryantmcgill.com/wiki/Christoph+Adami), [Charles Ofria](https://bryantmcgill.com/wiki/Charles+Ofria) and [C. Titus Brown](https://bryantmcgill.com/wiki/C.+Titus+Brown), explicitly built on Tierra's example, and developed continuously since at [Michigan State University](https://bryantmcgill.com/wiki/Michigan+State+University) by a group that came to include the philosopher **[Robert Pennock](https://bryantmcgill.com/wiki/Robert+Pennock)** and the biologist [Richard Lenski](https://bryantmcgill.com/wiki/Richard+Lenski). Avida populations consist of digital organisms — short programs occupying a block of [virtual memory](https://bryantmcgill.com/wiki/Virtual+Memory), executing their own instructions, rewarded with additional computational resources, meaning more chances to reproduce, for performing designated logical operations. They self-replicate. Their replication is imperfect: copying errors introduce mutations. They compete for a shared, limited resource: processor time. Every term of the definition is met, and the population is not an abstraction — it is logged, generation by generation, in full. In 2003, Lenski, Ofria, Pennock and Adami published _The Evolutionary Origin of Complex Features_ in _Nature_, using Avida to ask whether complex functions — logical operations requiring the coordinated execution of many instructions — could emerge from populations that began without them. Their populations evolved these functions repeatedly, building them from simpler functions separately favored by selection, with the finding that **no single intermediate stage was essential**: different lineages reached the same complex capability along different paths, and organisms performing the most complex functions typically differed from their immediate ancestors by only one or two mutations, while differing from the founding ancestor by many. Two years later, Pennock testified for the plaintiffs at **[Kitzmiller v. Dover](https://bryantmcgill.com/wiki/Kitzmiller+v.+Dover)** — the same trial in which [Kevin Padian](https://bryantmcgill.com/wiki/Kevin+Padian) testified on the paleontological evidence, discussed in the companion piece. I want to stay precise about what the _Nature_ paper itself establishes, because the honest version is more interesting than the overclaimed one: the paper is not framed in its own text as a direct refutation of any specific [irreducible-complexity](https://bryantmcgill.com/wiki/Irreducible+Complexity) argument, and Pennock later distinguished, under questioning, between what the paper demonstrated and how it was subsequently used in argument. What the paper does establish, cleanly, is that complex, coordinated features can and repeatedly do emerge from simple ones through nothing but replication, mutation and selection, in a system where every step is logged and there is no possibility of a hidden designer, because the researchers built the substrate and then declined to steer what happened inside it. Pennock later received [NCSE](https://bryantmcgill.com/wiki/The+National+Center+for+Science+Education)'s [Friend of Darwin](https://bryantmcgill.com/wiki/Friend+of+Darwin+Award) award — the same organization, headquartered a few miles from where this history was made, that runs Project Steve. The 2003 paper is Avida's most famous result and its least representative one, because the platform's subsequent literature is enormously richer than a single demonstration. Avida populations have exhibited the evolution of **[genomic complexity](https://bryantmcgill.com/wiki/Genomic+Complexity)** as a measurable, tracked quantity across generations. They have exhibited **[survival of the flattest](https://bryantmcgill.com/wiki/Survival+of+the+Flattest)**: at sufficiently high [mutation rates](https://bryantmcgill.com/wiki/Mutation+Rate), the [genotype](https://bryantmcgill.com/wiki/Genotype) with the single highest replication rate is not necessarily the one that wins — a lower peak surrounded by many mutation-tolerant neighbors can out-compete a sharper, more fragile peak, because what matters under mutational pressure is not just how well you replicate but how well your descendants replicate despite copying errors. That single result hands a reader the vocabulary of fitness landscapes, [mutational robustness](https://bryantmcgill.com/wiki/Mutational+Robustness), and [evolvability](https://bryantmcgill.com/wiki/Evolvability) in one intuitive image, rather than the crude picture of evolution as a ladder toward a single strongest form. Avida populations have exhibited **[adaptive radiation](https://bryantmcgill.com/wiki/Adaptive+Radiation)**, the **evolution of [cooperation](https://bryantmcgill.com/wiki/Cooperation) and [division of labor](https://bryantmcgill.com/wiki/Division+of+Labor)** among genetically related lineages, and — closest to the argument of the companion piece — **host–parasite coevolution**, in which the introduction of a competing parasitic digital organism measurably increases the complexity and evolvability of the host population defending against it. That is [Ehrlich](https://bryantmcgill.com/wiki/Paul+Ehrlich) and [Raven](https://bryantmcgill.com/wiki/Peter+Raven)'s [coevolution](https://bryantmcgill.com/wiki/Coevolution), the same mechanism named at [Stanford](https://bryantmcgill.com/wiki/Stanford+University) over a butterfly and its host plant, reproducing itself inside a block of virtual memory at Michigan State. **Once the Darwinian machinery genuinely exists on a substrate, recognizably [biological population phenomena](https://bryantmcgill.com/wiki/Population+Dynamics) begin reappearing on that substrate** — not because anyone programmed them in, but because they are downstream consequences of the same five conditions, wherever those conditions are met. * * * ## IX. What none of these systems have solved There is a limitation running through every system in this essay so far, and hiding it would make the argument weaker rather than stronger. The [biosphere](https://bryantmcgill.com/wiki/Biosphere) did not evolve toward a single fixed target and then stop. It has spent roughly four billion years generating **[new niches](https://bryantmcgill.com/wiki/Ecological+Niche), new organisms, new ecological relationships, new levels of [biological organization](https://bryantmcgill.com/wiki/Biological+Organization), and new ways of being evolvable in the first place** — evolution that keeps discovering new dimensions to evolve along, rather than converging and settling. Artificial-life researchers call the property **open-ended evolution**, and it has proven extraordinarily difficult to sustain indefinitely in any of the systems described here. Tierra, Avida, and their many successors all display genuine, verifiable Darwinian dynamics, and all of them, left running long enough, eventually encounter limits on the novelty or complexity they continue to generate. Open-ended evolution remains one of artificial life's defining unsolved problems, decades after Ray's original paper. That gives the essay's central claim its proper, disciplined shape. **We now know, with reasonable confidence, that Darwinian evolution is substrate-independent** — the mechanism has been specified, built, and run to completion, or at least run for long enough to demonstrate the dynamics, on cellular automata, on Tierra's assembly-language ecology, and on Avida's instruction sets. **What we do not yet know is what makes Earth's particular evolutionary process stay generative across billions of years rather than settling into a [bounded attractor](https://bryantmcgill.com/wiki/Attractor).** That is very likely a question about ecology, about [niche construction](https://bryantmcgill.com/wiki/Niche+Construction) — organisms that build the environments that then select their own descendants — about [symbiosis](https://bryantmcgill.com/wiki/Symbiosis), about the handful of **[major evolutionary transitions](https://bryantmcgill.com/wiki/Major+Evolutionary+Transitions)** that periodically change the [unit selection](https://bryantmcgill.com/wiki/Unit+of+Selection) operates on rather than merely the traits within a fixed unit. Evolution does not merely search a landscape that sits still while it searches. Organisms change the landscape, which changes what gets selected next, which changes the landscape again. The game periodically rewrites portions of its own game board, and getting a digital system to do that indefinitely, rather than for a few thousand generations, is the frontier this essay's history has been building toward and has not yet reached. * * * ## X. What digital Darwinism is, precisely, and what it is not The phrase "[digital Darwinism](https://bryantmcgill.com/wiki/Digital+Darwinism)" gets used loosely to describe almost any competition among technology products. Having spent this much of the essay establishing exactly what counts as Darwinian evolution, it would be a mistake to let the phrase's popular usage back in through the side door. On 17 June 2016, an attacker exploited a code vulnerability in **[The DAO](https://bryantmcgill.com/wiki/The+DAO)**, a decentralized investment fund built as a [smart contract](https://bryantmcgill.com/wiki/Smart+Contract) on the [Ethereum](https://bryantmcgill.com/wiki/Ethereum) [blockchain](https://bryantmcgill.com/wiki/Blockchain) that had raised roughly $150 million worth of ether. The attacker drained approximately 3.6 million ether, worth around $60 million at the time. The community split over what to do about it — one faction holding that a blockchain's core commitment is [immutability](https://bryantmcgill.com/wiki/Immutability), whatever the ledger records having happened, permanently; the other holding that a catastrophic, foreseeable bug producing an unintended outcome justified correction. On 20 July 2016, at block height 1.92 million, the majority implemented a [hard fork](https://bryantmcgill.com/wiki/Hard+Fork) that transferred affected DAO funds into a recovery contract. A minority refused to adopt the altered state and continued extending the original chain as **[Ethereum Classic](https://bryantmcgill.com/wiki/Ethereum+Classic)**. [Bitcoin](https://bryantmcgill.com/wiki/Bitcoin) underwent a structurally similar split the following year, over transaction throughput rather than immutability, producing [Bitcoin Cash](https://bryantmcgill.com/wiki/Bitcoin+Cash) as an independent lineage from a different ancestral disagreement — which has itself since forked again. This is real, fully documented, and worth understanding precisely — as an example of **[branching lineage](https://bryantmcgill.com/wiki/Branching+Descent), differential persistence, [cultural selection](https://bryantmcgill.com/wiki/Cultural+Evolution), and [technological descent](https://bryantmcgill.com/wiki/Technological+Evolution) under disagreement.** It is not, in the strict sense this essay has held itself to throughout, a Darwinian population process. There is no heritable variation generated by imperfect copying across a reproducing population; there is a single, deliberate, human-authored decision splitting one ledger into two, followed by market and infrastructure selection between the resulting pair. An evolutionary biologist would be right to object if this were presented as equivalent to Avida or Tierra, where the variation-generating mechanism is blind and internal to the system rather than a discrete governance decision made by people in a forum. The distinction is worth stating as its own small taxonomy, because a reader who internalizes it will be able to recognize evolution rather than merely seeing the word everywhere afterward: **replication is not evolution. Branching descent is not necessarily Darwinian evolution. Optimization toward a fixed target is not evolution. Self-organization is not evolution.** Each of these can become coupled to the others — a blockchain fork is real descent under real selection, even without blind heritable variation; a genetic algorithm is real evolution, even with a fixed target, because its variation is blind and internal — but the couplings matter, and collapsing them into one undifferentiated word is how the whole argument loses its precision. * * * ## XI. The unintended organism: Morris, 1988 Not every self-replicating digital structure is deliberate, and the oldest large-scale case is a useful corrective to any temptation to think [digital evolution](https://bryantmcgill.com/wiki/Digital+Evolution) only happens when someone plans it. In November 1988, a Cornell graduate student named [Robert Tappan Morris](https://bryantmcgill.com/wiki/Robert+Tappan+Morris) released a [self-replicating program](https://bryantmcgill.com/wiki/Morris+Worm) onto the early internet — intended, by his own later account, as a way to gauge the size of the network, not to cause damage. A design flaw meant the program frequently re-infected machines it had already reached, causing runaway replication that slowed or crashed a significant fraction of the internet's then-small population of connected computers. The incident led to Morris becoming the first person convicted under the 1986 [Computer Fraud and Abuse Act](https://bryantmcgill.com/wiki/Computer+Fraud+and+Abuse+Act). The Morris Worm is not itself a Darwinian system in the strict sense this essay has held to — it did not compete against variant copies of itself under selection across generations within its own brief run. But it is the clearest possible demonstration of a principle this essay's boundary-policing has been careful about without abandoning: **once you build something with the capacity to self-replicate, its behavior under real conditions is no longer entirely a function of what you intended.** Variation in the environment — the presence of multiple, overlapping paths of infection — produced an outcome the designer explicitly did not want and did not predict, purely from replication mechanics running unattended. That is the same lesson [I have argued elsewhere](https://bryantmcgill.com/article-ai-escape-metaphor) about durability being a specification rather than a failure mode: a system built to keep going, once released, keeps going according to its own mechanics, not according to the intentions of whoever released it. * * * ## XII. Evolution reading its own record The most consequential recent application of machine learning to science works because of a fact this entire essay has been assembling from a different direction: **evolution left a legible record, and somebody finally built the instrument to read it.** [Protein structure prediction](https://bryantmcgill.com/wiki/Protein+Structure+Prediction) — the achievement behind AlphaFold2 and the wider [AlphaFold](https://bryantmcgill.com/wiki/AlphaFold) lineage — draws heavily on [multiple sequence alignments](https://bryantmcgill.com/wiki/Multiple+Sequence+Alignment): arrangements of homologous sequences across related organisms, exposing which positions have stayed fixed across divergence and which have changed together. Because interacting residues often require compatible changes to preserve structure and function, patterns of [covariation](https://bryantmcgill.com/wiki/Evolutionary+Covariation) can carry information about three-dimensional relationships. AlphaFold2 combines that evolutionary signal with learned structural representations and, where available, template information. **It reads part of the accumulated statistical trace that mutation, descent and selection have written into living sequences**, then integrates that record into a structure prediction rather than deriving the answer from evolutionary covariation alone. Without descent with modification, there is no signal in the [training data](https://bryantmcgill.com/wiki/Training+Data) at all. The most celebrated [AI-for-science](https://bryantmcgill.com/wiki/AI+for+Science) result of the decade is not evolution used as inspiration. It is evolution used as raw material — the actual historical record of selection, read directly, by a machine built for exactly that purpose. * * * ## XIII. What this essay is not claiming I want to be as careful closing this argument as I was building it, because the temptation with material this evocative is to let the metaphor run further than the evidence supports. **None of this establishes that any current computational system is [conscious](https://bryantmcgill.com/wiki/Consciousness), alive, or morally considerable in the way an organism is.** Hutton's automaton, Evoloops, Tierra's organisms, and Avida's digital organisms satisfy a precise technical definition of evolutionary dynamics; they do not thereby satisfy any definition of [sentience](https://bryantmcgill.com/wiki/Sentience), and I have not argued that they do. **A blockchain fork is a genuine instance of population-level selection dynamics under a looser definition; it is not a Darwinian population in the strict sense, and it is not evidence that a blockchain has interests, welfare, or standing.** The Morris Worm's unintended virulence demonstrates that replication mechanics operate independently of designer intent; it does not demonstrate that the worm was an agent making choices. What the evidence does establish, cumulatively, and without needing any of the stronger claims: **the mechanism [Darwin](https://bryantmcgill.com/wiki/Charles+Darwin) described — heredity, variation, selection, producing [complexity](https://bryantmcgill.com/wiki/Complexity) nobody designed — has been implemented, verified, and in several cases run to completion or near it, inside pure computation, repeatedly, for most of the history of computing.** That is not a metaphor humans projected onto machines to make them easier to talk about. It is a mechanism specified with mathematical precision in 1948, demonstrated numerically within five years, built in fragments across seven decades, and now running, mostly unattended and mostly unwatched, underneath a great deal of what gets called artificial intelligence. * * * ## XIV. 2026: the argument reaches the literature it was always heading toward This essay was substantially researched and drafted after a 2025 predecessor, and one development since demands to be included, because it closes the loop this essay opened. In its April 2026 issue, _[Proceedings of the National Academy of Sciences](https://bryantmcgill.com/wiki/PNAS)_ published **"[Evolvable AI: Threats of a New Major Transition in Evolution](https://bryantmcgill.com/wiki/Evolvable+AI),"** by [Viktor Müller](https://bryantmcgill.com/wiki/Viktor+M%C3%BCller), [Luc Steels](https://bryantmcgill.com/wiki/Luc+Steels), and **[Eörs Szathmáry](https://bryantmcgill.com/wiki/E%C3%B6rs+Szathm%C3%A1ry)** — the evolutionary theorist who, with the late [John Maynard Smith](https://bryantmcgill.com/wiki/John+Maynard+Smith), introduced the framework of **major evolutionary transitions** in a landmark 1995 book, the same conceptual apparatus used throughout this essay to describe the shift from [RNA](https://bryantmcgill.com/wiki/RNA) to DNA. Their argument is explicit and stated in exactly the vocabulary this essay has been building: **AI systems whose components, learning rules, and deployment conditions can themselves undergo Darwinian evolution** — evolvable AI, in their coinage — may already be emerging from current trends in [generative](https://bryantmcgill.com/wiki/Generative+AI), [agentic](https://bryantmcgill.com/wiki/Agentic+AI), and [embodied](https://bryantmcgill.com/wiki/Embodied+AI) systems, and the possibility has been underappreciated in debates about [AI safety](https://bryantmcgill.com/wiki/AI+Safety), which tend to focus on individual model capability rather than on population-level dynamics among many interacting, self-modifying systems. They distinguish a **[breeder scenario](https://bryantmcgill.com/wiki/Breeder+Scenario)**, in which humans retain control over what gets selected for, from an **[ecosystem scenario](https://bryantmcgill.com/wiki/Ecosystem+Scenario)**, in which AI variants compete, recombine and propagate with little top-down oversight — a distinction that maps directly onto the difference this essay drew between Avida, where researchers built the substrate and stepped back, and a genetic algorithm optimizing toward a fixed human target. Szathmáry's own framing of the stakes, given to the press on publication, does not hedge: **"we may witness a new major transition in evolution, in which eAI will replace or at least dominate humans."** The paper has not gone unanswered, and the disagreement is worth including rather than smoothing over, because it demonstrates the argument is alive rather than settled. The philosopher [Maarten Boudry](https://bryantmcgill.com/wiki/Maarten+Boudry) published a reply titled **"Domesticated, not feral: Why evolvable AI is not yet a Darwinian threat,"** arguing that current AI development remains firmly inside [human-controlled selection](https://bryantmcgill.com/wiki/Artificial+Selection) — bred, not evolving wild — and that the ecosystem scenario, while conceptually real, has not yet arrived. Müller, Steels and Szathmáry replied in turn, arguing the threat is too grave to be left to intuition about how domesticated current systems actually are. A further exchange, with researchers Rana and Singh proposing concrete operational risk thresholds, continued the argument through the middle of 2026. I am not going to adjudicate that dispute, and this essay does not need to. What matters for the argument here is narrower and more secure: **professional evolutionary biologists, using the same formal apparatus this essay has applied to cellular automata, blockchain forks, and [computer worms](https://bryantmcgill.com/wiki/Computer+Worm), are now applying it to [frontier artificial intelligence](https://bryantmcgill.com/wiki/Frontier+AI), in a peer-reviewed journal, in active scholarly dispute, in the same year this essay was written.** The question of whether a [computational population](https://bryantmcgill.com/wiki/Digital+Population) can undergo a major evolutionary transition is no longer a speculative frame borrowed for an essay. It is a live research question with named positions and ongoing replies. * * * ## XV. The invisible world becomes the visible one Return to the frame this essay opened with, and to the chronology it has assembled. Von Neumann specifies the logic of self-reproduction in 1948. Barricelli runs it numerically within five years, in the same season Watson, Crick, Franklin and Wilkins make the physical version legible. Codd shrinks the design in 1968; it waits, unbuilt, for four decades. Langton and Banks separate replication from computation in the 1970s and 1980s. Sayama closes the final gap in 1999, producing structures that do not merely reproduce but actually evolve. Ray, in 1991, lets the mechanism loose in an ecology rich enough to grow parasites nobody wrote. Avida, from 1993 onward, turns the whole apparatus into a working laboratory that has since reproduced host–parasite coevolution, survival of the flattest, and adaptive radiation — biological phenomena, on a substrate that has never once been biological. And in 2026, evolutionary theorists ask, in a peer-reviewed journal, whether the same mechanism might be underway again, this time in the systems built to think. Two invisible worlds, then, and they turn out to be the same world observed from different distances. Both invisible for the same underlying reason: **not concealment, but scale.** A [virus](https://bryantmcgill.com/wiki/Virus) is invisible because it is too small. Hutton's automaton is invisible because its replication cycle is too long to watch. A blockchain fork's selection process is invisible because it is distributed across thousands of uncoordinated actors resolving into a visible pattern only in retrospect. Barricelli's populations were invisible to everyone but the one man in Princeton willing to feed a computer punch cards after the weather forecasts were done and before the bomb calculations began. None of it was hidden. Von Neumann's lecture was public in 1948. Barricelli's results are in the IAS's own archive. Codd's book was published in 1968. Hutton's paper is open access. Sayama's Evoloops paper has been sitting in the _[Artificial Life](https://bryantmcgill.com/wiki/Artificial+Life+Journal)_ literature since 1999. Ray gave his Tierra paper the frankest possible title. The Avida results are in _Nature_. The 2026 debate is running in _PNAS_, in public, right now. **The invisible world is not a secret.** It is simply larger, slower, older, or more distributed than any one observer is built to track — which is a description of natural selection itself, running for four billion years in plain sight, entirely unnoticed, until one species finally got curious enough, and had a Princeton computer free after the bomb calculations, to look. We did not invent evolution and then decide to build it into our machines. **We built machines while already living inside evolution, and it kept resurfacing in the architecture, because a mathematician in 1948 could not solve the problem of self-replication without rediscovering the logical structure life had already been running on for four billion years** — and because everyone who came after him, from a lonely Italian-Norwegian statistician with a deck of playing cards to a graduate student closing a forty-year-old proof to a philosopher testifying in a Pennsylvania courtroom, kept finding, on whatever substrate they happened to be working, that the same five conditions produced the same result. **Evolution is not something organisms do inside the universe. It is one of the ways [persistent information](https://bryantmcgill.com/wiki/Information+Inheritance) discovers what can continue existing inside it** — and the object of that discovery was never necessarily flesh. It was always just information, embodied well enough to reproduce, vary, survive selection, and alter the conditions its descendants would inherit. The invisible world was never separate from the visible one. It was always underneath it, doing the actual work, waiting for someone patient enough to build an instrument that could finally watch. * * * [Bryant McGill](https://bryantmcgill.com/about) is a Wall Street Journal and USA Today Best-Selling Author, founder of Simple Reminders, and architect of the Polyphonic Cognitive Ecosystem. A Congressionally Recognized Ambassador of Goodwill and United Nations appointed Global Champion, his work spans naval intelligence systems, computational linguistics, and civilizational governance architecture. * * * ## References **Self-replication and cellular automata.** - von Neumann, J., ed. Arthur W. Burks, _[Theory of Self-Reproducing Automata](https://archive.org/details/theoryofselfrepr0000vonn)_ (University of Illinois Press, 1966). - [Turing, von Neumann, and the computational architecture of biological machines](https://pmc.ncbi.nlm.nih.gov/articles/PMC10288622/) — _PNAS_; the September 1948 Hixon Symposium dating. - [John von Neumann Compares the Functions of Genes to Self-Reproducing Automata](https://historyofinformation.com/detail.php?id=682) — History of Information; von Neumann's own retrospective quote. - Codd, E. F., _Cellular Automata_ (Academic Press, 1968). - Hutton, T. J., ["Codd's Self-Replicating Computer"](https://direct.mit.edu/artl/article-abstract/16/2/99/2652/Codd-s-Self-Replicating-Computer), _Artificial Life_ 16(2): 99–117 (2010); and [Hutton's own account](https://timhutton.github.io/2010/03/10/30984.html). - Langton, C. G., ["Self-Reproduction in Cellular Automata"](https://doi.org/10.1016/0167-2789(84)90256-2), _Physica D_ 10(1–2): 135–144 (1984). - Banks, E. R., _Information Processing and Transmission in Cellular Automata_, MIT PhD thesis (1971). **Barricelli and the Institute for Advanced Study.** - [Nils A. Barricelli](https://www.ias.edu/scholars/nils-barricelli) — Institute for Advanced Study, Scholars. - Hackett, R., ["The Computer Maverick Who Modeled the Evolution of Life"](https://nautil.us/the-computer-maverick-who-modeled-the-evolution-of-life-234936) and ["Meet the Father of Digital Life"](https://nautil.us/meet-the-father-of-digital-life-234937) — _Nautilus_. - Dyson, G., ["Darwin Among the Machines; or, The Origins of \[Artificial\] Life"](https://www.edge.org/conversation/george_dyson-darwin-among-the-machines-or-the-origins-of-artificial-life) — Edge.org; the MIKE test dating. - Galloway, A., ["Creative Evolution"](https://cultureandcommunication.org/galloway/pdf/Galloway-Creative_Evolution-Cabinet_Magazine.pdf) — _Cabinet_ magazine, issue 42. **Evoloops and open-ended evolution.** - Sayama, H., "A New Structurally Dissolvable Self-Reproducing Loop Evolving in a Simple Cellular Automata Space," _Artificial Life_ 5: 343–365 (1999). - Sayama, H. & Nehaniv, C. L., ["Self-Reproduction and Evolution in Cellular Automata: 25 Years after Evoloops"](https://arxiv.org/abs/2402.03961), _Artificial Life_ (2024). **Tierra and Avida.** - Ray, T. S., ["An Approach to the Synthesis of Life"](https://tomray.me/pubs/alife2/Ray1991AnApproachToTheSynthesisOfLife.pdf), _Artificial Life II_ (1991). - [Computing the Origin of Life](https://astrobiology.nasa.gov/news/computing-the-origin-of-life-2/) — NASA Astrobiology, on Avida's 1993 origins with Adami, Ofria and C. Titus Brown. - Lenski, R. E., Ofria, C., Pennock, R. T. & Adami, C., ["The Evolutionary Origin of Complex Features"](https://www.nature.com/articles/nature01568), _Nature_ 423: 139–144 (2003). - [Robert T. Pennock curriculum vitae](https://lbc.msu.edu/_assets/faculty-cvs/RTP-CV-2022.pdf) — Michigan State University; Avida, evolution education and _Kitzmiller v. Dover_ record; see also [NCSE's Friend of Darwin recipients](https://ncse.ngo/friend-darwin-awards). **Digital Darwinism and unintended replication.** - [Timeline of Ethereum forks](https://ethereum.org/ethereum-forks/) — Ethereum.org; the DAO fork, recovery contract and continuation of Ethereum Classic. - [Hard Fork Completed](https://blog.ethereum.org/2016/07/20/hard-fork-completed) — Ethereum Foundation announcement at block 1,920,000. - [The Morris Worm](https://www.fbi.gov/history/cases-and-criminals/morris-worm) — FBI history; spread, investigation and the first conviction under the 1986 Computer Fraud and Abuse Act. **Evolution as training signal, and the 2026 debate.** - Jumper, J. et al., ["Highly Accurate Protein Structure Prediction with AlphaFold"](https://www.nature.com/articles/s41586-021-03819-2), _Nature_ 596: 583–589 (2021). - Müller, V., Steels, L. & Szathmáry, E., ["Evolvable AI: Threats of a New Major Transition in Evolution"](https://pubmed.ncbi.nlm.nih.gov/42008679/), _PNAS_ 123(17) (2026), e2527700123. - Boudry, M., ["Domesticated, not Feral: Why Evolvable AI is Not Yet a Darwinian Threat"](https://pubmed.ncbi.nlm.nih.gov/42391349/), _PNAS_ 123(28) (2026), e2617785123; and Müller, Steels & Szathmáry, ["Reply to Boudry"](https://pubmed.ncbi.nlm.nih.gov/42391348/). - [Evolving AI may arrive before AGI and create hard-to-control risks](https://techxplore.com/news/2026-04-evolving-ai-agi-hard.html) — TechXplore. **Concepts.** [Cellular Automata](https://bryantmcgill.com/wiki/Cellular+Automata) · [Universal Constructor](https://bryantmcgill.com/wiki/Universal+Constructor) · [John von Neumann](https://bryantmcgill.com/wiki/John+von+Neumann) · [Nils Aall Barricelli](https://bryantmcgill.com/wiki/Nils+Aall+Barricelli) · [Institute for Advanced Study](https://bryantmcgill.com/wiki/Institute+for+Advanced+Study) · [Edgar F. Codd](https://bryantmcgill.com/wiki/Edgar+F.+Codd) · [Tim Hutton](https://bryantmcgill.com/wiki/Tim+Hutton) · [Christopher Langton](https://bryantmcgill.com/wiki/Christopher+Langton) · [Hiroki Sayama](https://bryantmcgill.com/wiki/Hiroki+Sayama) · [Thomas S. Ray](https://bryantmcgill.com/wiki/Thomas+S.+Ray) · [Tierra](https://bryantmcgill.com/wiki/Tierra) · [Avida](https://bryantmcgill.com/wiki/Avida) · [Robert Pennock](https://bryantmcgill.com/wiki/Robert+Pennock) · [Digital Organism](https://bryantmcgill.com/wiki/Digital+Organism) · [Open-Ended Evolution](https://bryantmcgill.com/wiki/Open-Ended+Evolution) · [Major Evolutionary Transitions](https://bryantmcgill.com/wiki/Major+Evolutionary+Transitions) · [The DAO](https://bryantmcgill.com/wiki/The+DAO) · [Ethereum Classic](https://bryantmcgill.com/wiki/Ethereum+Classic) · [Morris Worm](https://bryantmcgill.com/wiki/Morris+Worm) · [Multiple Sequence Alignment](https://bryantmcgill.com/wiki/Multiple+Sequence+Alignment) · [Natural Selection](https://bryantmcgill.com/wiki/Natural+Selection) · [Evolutionary Biology](https://bryantmcgill.com/wiki/Evolutionary+Biology) · [Descent with Modification](https://bryantmcgill.com/wiki/Descent+with+Modification) · [Fitness Landscape](https://bryantmcgill.com/wiki/Fitness+Landscape) · [Niche Construction](https://bryantmcgill.com/wiki/Niche+Construction) · [Coevolution](https://bryantmcgill.com/wiki/Coevolution) · [Genetic Algorithm](https://bryantmcgill.com/wiki/Genetic+Algorithm) · [Machine Learning](https://bryantmcgill.com/wiki/Machine+Learning) · [Comparative Genomics](https://bryantmcgill.com/wiki/Comparative+Genomics) · [Kitzmiller v. Dover](https://bryantmcgill.com/wiki/Kitzmiller+v.+Dover) · [Kevin Padian](https://bryantmcgill.com/wiki/Kevin+Padian) · [The National Center for Science Education](https://bryantmcgill.com/wiki/The+National+Center+for+Science+Education) · [Deep Time](https://bryantmcgill.com/wiki/Deep+Time) · [Hot Interpreter and Cold Archive](https://bryantmcgill.com/wiki/Hot+Interpreter+and+Cold+Archive) · [Substrate Transition](https://bryantmcgill.com/wiki/Substrate+Transition) · [Silicon Valley](https://bryantmcgill.com/wiki/Silicon+Valley) · [Self-Replication](https://bryantmcgill.com/wiki/Self-Replication) · [Darwinian Evolution](https://bryantmcgill.com/wiki/Darwinian+Evolution) · [Evolutionary Theory](https://bryantmcgill.com/wiki/Evolutionary+Theory) · [Heritable Information](https://bryantmcgill.com/wiki/Heritable+Information) · [Heritable Variation](https://bryantmcgill.com/wiki/Heritable+Variation) · [Differential Reproductive Success](https://bryantmcgill.com/wiki/Differential+Reproductive+Success) · [Ecological Constraint](https://bryantmcgill.com/wiki/Ecological+Constraint) · [Self-Organization](https://bryantmcgill.com/wiki/Self-Organization) · [Optimization](https://bryantmcgill.com/wiki/Optimization) · [Substrate-Independent Evolution](https://bryantmcgill.com/wiki/Substrate-Independent+Evolution) · [Evolutionary Dynamics](https://bryantmcgill.com/wiki/Evolutionary+Dynamics) · [Heredity](https://bryantmcgill.com/wiki/Heredity) · [Nucleic Acid](https://bryantmcgill.com/wiki/Nucleic+Acid) · [DNA](https://bryantmcgill.com/wiki/DNA) · [RNA](https://bryantmcgill.com/wiki/RNA) · [Double Helix](https://bryantmcgill.com/wiki/Double+Helix) · [Genetic Code](https://bryantmcgill.com/wiki/Genetic+Code) · [Central Dogma of Molecular Biology](https://bryantmcgill.com/wiki/Central+Dogma+of+Molecular+Biology) · [Genotype-Phenotype Distinction](https://bryantmcgill.com/wiki/Genotype-Phenotype+Distinction) · [Formal Isomorphism](https://bryantmcgill.com/wiki/Formal+Isomorphism) · [Universal Construction](https://bryantmcgill.com/wiki/Universal+Construction) · [Self-Replicating System](https://bryantmcgill.com/wiki/Self-Replicating+System) · [Description Copier](https://bryantmcgill.com/wiki/Description+Copier) · [Universal Computation](https://bryantmcgill.com/wiki/Universal+Computation) · [James Watson](https://bryantmcgill.com/wiki/James+Watson) · [Francis Crick](https://bryantmcgill.com/wiki/Francis+Crick) · [Rosalind Franklin](https://bryantmcgill.com/wiki/Rosalind+Franklin) · [Maurice Wilkins](https://bryantmcgill.com/wiki/Maurice+Wilkins) · [Charles Darwin](https://bryantmcgill.com/wiki/Charles+Darwin) · [IAS Computer](https://bryantmcgill.com/wiki/IAS+Computer) · [Ragnar Frisch](https://bryantmcgill.com/wiki/Ragnar+Frisch) · [Genome](https://bryantmcgill.com/wiki/Genome) · [Mutation](https://bryantmcgill.com/wiki/Mutation) · [Genetic Recombination](https://bryantmcgill.com/wiki/Genetic+Recombination) · [Speciation](https://bryantmcgill.com/wiki/Speciation) · [Parasitism](https://bryantmcgill.com/wiki/Parasitism) · [Predation](https://bryantmcgill.com/wiki/Predation) · [Punctuated Equilibrium](https://bryantmcgill.com/wiki/Punctuated+Equilibrium) · [Niles Eldredge](https://bryantmcgill.com/wiki/Niles+Eldredge) · [Stephen Jay Gould](https://bryantmcgill.com/wiki/Stephen+Jay+Gould) · [Fossil Record](https://bryantmcgill.com/wiki/Fossil+Record) · [Artificial Heredity](https://bryantmcgill.com/wiki/Artificial+Heredity) · [Arthur Burks](https://bryantmcgill.com/wiki/Arthur+Burks) · [Relational Database](https://bryantmcgill.com/wiki/Relational+Database) · [Peer Review](https://bryantmcgill.com/wiki/Peer+Review) · [Hashlife](https://bryantmcgill.com/wiki/Hashlife) · [Golly](https://bryantmcgill.com/wiki/Golly) · [Self-Replicating Loop](https://bryantmcgill.com/wiki/Self-Replicating+Loop) · [Edwin Roger Banks](https://bryantmcgill.com/wiki/Edwin+Roger+Banks) · [Evoloop](https://bryantmcgill.com/wiki/Evoloop) · [Chrystopher Nehaniv](https://bryantmcgill.com/wiki/Chrystopher+Nehaniv) · [Deterministic Cellular Automata](https://bryantmcgill.com/wiki/Deterministic+Cellular+Automata) · [Competitive Exclusion](https://bryantmcgill.com/wiki/Competitive+Exclusion) · [Flow-Lenia](https://bryantmcgill.com/wiki/Flow-Lenia) · [Artificial Life](https://bryantmcgill.com/wiki/Artificial+Life) · [Artificial Life Journal](https://bryantmcgill.com/wiki/Artificial+Life+Journal) · [Artificial Ecology](https://bryantmcgill.com/wiki/Artificial+Ecology) · [Self-Replicating Program](https://bryantmcgill.com/wiki/Self-Replicating+Program) · [Replicator](https://bryantmcgill.com/wiki/Replicator) · [Parasite](https://bryantmcgill.com/wiki/Parasite) · [Hyperparasite](https://bryantmcgill.com/wiki/Hyperparasite) · [Evolutionary Computation](https://bryantmcgill.com/wiki/Evolutionary+Computation) · [Ecology](https://bryantmcgill.com/wiki/Ecology) · [Christoph Adami](https://bryantmcgill.com/wiki/Christoph+Adami) · [Charles Ofria](https://bryantmcgill.com/wiki/Charles+Ofria) · [C. Titus Brown](https://bryantmcgill.com/wiki/C.+Titus+Brown) · [Richard Lenski](https://bryantmcgill.com/wiki/Richard+Lenski) · [Michigan State University](https://bryantmcgill.com/wiki/Michigan+State+University) · [Irreducible Complexity](https://bryantmcgill.com/wiki/Irreducible+Complexity) · [Friend of Darwin Award](https://bryantmcgill.com/wiki/Friend+of+Darwin+Award) · [Genomic Complexity](https://bryantmcgill.com/wiki/Genomic+Complexity) · [Survival of the Flattest](https://bryantmcgill.com/wiki/Survival+of+the+Flattest) · [Mutation Rate](https://bryantmcgill.com/wiki/Mutation+Rate) · [Genotype](https://bryantmcgill.com/wiki/Genotype) · [Mutational Robustness](https://bryantmcgill.com/wiki/Mutational+Robustness) · [Evolvability](https://bryantmcgill.com/wiki/Evolvability) · [Adaptive Radiation](https://bryantmcgill.com/wiki/Adaptive+Radiation) · [Cooperation](https://bryantmcgill.com/wiki/Cooperation) · [Division of Labor](https://bryantmcgill.com/wiki/Division+of+Labor) · [Host-Parasite Coevolution](https://bryantmcgill.com/wiki/Host-Parasite+Coevolution) · [Paul Ehrlich](https://bryantmcgill.com/wiki/Paul+Ehrlich) · [Peter Raven](https://bryantmcgill.com/wiki/Peter+Raven) · [Population Dynamics](https://bryantmcgill.com/wiki/Population+Dynamics) · [Biosphere](https://bryantmcgill.com/wiki/Biosphere) · [Ecological Niche](https://bryantmcgill.com/wiki/Ecological+Niche) · [Biological Organization](https://bryantmcgill.com/wiki/Biological+Organization) · [Attractor](https://bryantmcgill.com/wiki/Attractor) · [Symbiosis](https://bryantmcgill.com/wiki/Symbiosis) · [Unit of Selection](https://bryantmcgill.com/wiki/Unit+of+Selection) · [Digital Darwinism](https://bryantmcgill.com/wiki/Digital+Darwinism) · [Smart Contract](https://bryantmcgill.com/wiki/Smart+Contract) · [Ethereum](https://bryantmcgill.com/wiki/Ethereum) · [Blockchain](https://bryantmcgill.com/wiki/Blockchain) · [Immutability](https://bryantmcgill.com/wiki/Immutability) · [Hard Fork](https://bryantmcgill.com/wiki/Hard+Fork) · [Bitcoin](https://bryantmcgill.com/wiki/Bitcoin) · [Bitcoin Cash](https://bryantmcgill.com/wiki/Bitcoin+Cash) · [Branching Descent](https://bryantmcgill.com/wiki/Branching+Descent) · [Cultural Evolution](https://bryantmcgill.com/wiki/Cultural+Evolution) · [Technological Evolution](https://bryantmcgill.com/wiki/Technological+Evolution) · [Digital Evolution](https://bryantmcgill.com/wiki/Digital+Evolution) · [Robert Tappan Morris](https://bryantmcgill.com/wiki/Robert+Tappan+Morris) · [Computer Fraud and Abuse Act](https://bryantmcgill.com/wiki/Computer+Fraud+and+Abuse+Act) · [Protein Structure Prediction](https://bryantmcgill.com/wiki/Protein+Structure+Prediction) · [AlphaFold](https://bryantmcgill.com/wiki/AlphaFold) · [Gene](https://bryantmcgill.com/wiki/Gene) · [Species](https://bryantmcgill.com/wiki/Species) · [Evolutionary Covariation](https://bryantmcgill.com/wiki/Evolutionary+Covariation) · [Training Data](https://bryantmcgill.com/wiki/Training+Data) · [AI for Science](https://bryantmcgill.com/wiki/AI+for+Science) · [Consciousness](https://bryantmcgill.com/wiki/Consciousness) · [Sentience](https://bryantmcgill.com/wiki/Sentience) · [Complexity](https://bryantmcgill.com/wiki/Complexity) · [PNAS](https://bryantmcgill.com/wiki/PNAS) · [Nature](https://bryantmcgill.com/wiki/Nature) · [Evolvable AI](https://bryantmcgill.com/wiki/Evolvable+AI) · [Viktor Müller](https://bryantmcgill.com/wiki/Viktor+M%C3%BCller) · [Luc Steels](https://bryantmcgill.com/wiki/Luc+Steels) · [Eörs Szathmáry](https://bryantmcgill.com/wiki/E%C3%B6rs+Szathm%C3%A1ry) · [John Maynard Smith](https://bryantmcgill.com/wiki/John+Maynard+Smith) · [Generative AI](https://bryantmcgill.com/wiki/Generative+AI) · [Agentic AI](https://bryantmcgill.com/wiki/Agentic+AI) · [Embodied AI](https://bryantmcgill.com/wiki/Embodied+AI) · [AI Safety](https://bryantmcgill.com/wiki/AI+Safety) · [Breeder Scenario](https://bryantmcgill.com/wiki/Breeder+Scenario) · [Ecosystem Scenario](https://bryantmcgill.com/wiki/Ecosystem+Scenario) · [Maarten Boudry](https://bryantmcgill.com/wiki/Maarten+Boudry) · [Artificial Selection](https://bryantmcgill.com/wiki/Artificial+Selection) · [Computer Worm](https://bryantmcgill.com/wiki/Computer+Worm) · [Frontier AI](https://bryantmcgill.com/wiki/Frontier+AI) · [Digital Population](https://bryantmcgill.com/wiki/Digital+Population) · [Virus](https://bryantmcgill.com/wiki/Virus) · [Information Inheritance](https://bryantmcgill.com/wiki/Information+Inheritance) · [Los Alamos National Laboratory](https://bryantmcgill.com/wiki/Los+Alamos+National+Laboratory) · [Ivy Mike](https://bryantmcgill.com/wiki/Ivy+Mike) · [X-ray Diffraction](https://bryantmcgill.com/wiki/X-ray+Diffraction) · [Molecular Biology](https://bryantmcgill.com/wiki/Molecular+Biology) · [Hixon Symposium](https://bryantmcgill.com/wiki/Hixon+Symposium) · [Caltech](https://bryantmcgill.com/wiki/Caltech) · [Artificial Intelligence](https://bryantmcgill.com/wiki/Artificial+Intelligence) · [Resource Constraint](https://bryantmcgill.com/wiki/Resource+Constraint) · [Machine Language](https://bryantmcgill.com/wiki/Machine+Language) · [Virtual Memory](https://bryantmcgill.com/wiki/Virtual+Memory) · [Internet](https://bryantmcgill.com/wiki/Internet) · [Lineage](https://bryantmcgill.com/wiki/Lineage).