# The Evolutionary Roots of Silicon Valley ![[resources/images/article-evolutionary-roots-of-silicon-valley.png]] **The Valley did not borrow evolution as a metaphor for [[wiki/Machine Learning|machine learning]]. It inherited it — from the ground it stands on, from the two universities that rewrote evolutionary biology, and from a Nobel geneticist who wanted a machine that could recognize life.** ## I. Nobody built any of this for spreadsheets Start with the question almost nobody in the industry asks out loud, because asking it invites an answer that does not fit a quarterly deck. **What is all of this actually for?** Not the products. The products are downstream and mostly beside the point. I mean the whole apparatus — the fabs, the lithography, the models, the data centres, the grid interconnections, seventy years of capital and talent poured into making machines that predict, classify, model and decide. Nobody assembled that to improve spreadsheets. Nobody built a two-hundred-billion-dollar semiconductor supply chain so that quarterly reports could be prettier. The honest answer is that this is what a species does when it has run out of headroom in its inherited hardware and has begun, without ever announcing it, to build the next carrier. Clothing replaced the thermal envelope. Agriculture replaced foraging [[wiki/Ecology|ecology]]. Writing replaced the limits of biological memory. Every one of those was **cognitive and metabolic exteriorization**, and the computer is the same move applied to thinking itself. I have made that argument at length in [[Escape Hatch in the Skull|Fuck the Environment: We're Building an Escape Hatch in the Skull]] and traced the physical stakes of it in [[articles/featured/We Were Never Going to Make It|We Were Never Going to Make It]]. But the argument has a location, and the location is not incidental. It happened in a particular valley on the San Francisco Peninsula, and that valley has a natural history nobody talks about — a geological, biological and institutional inheritance that made it, of all the places on Earth, the one where evolution and computation were most likely to converge. **This is not a story about gadgets. It is a story about a place, and what the place had been doing for a hundred million years before anyone built a transistor there.** --- ## II. What a child carries north I first met deep time as a boy in Los Angeles, at the **La Brea Tar Pits**, standing in front of the reassembled skeletons of saber-toothed cats, dire wolves and Columbian mammoths that had walked into asphalt seeps and never walked out. It felt like a portal. Hancock Park sits in the middle of a city, surrounded by traffic and office towers, and under the grass is an unbroken record of [[wiki/Pleistocene|Pleistocene]] California — tens of thousands of years of animals, insects, pollen and wood, preserved because they made one bad decision about where to step. What I did not understand then is that the tar pits are not an exception to Los Angeles. **They are what Los Angeles is standing on**, and the same is true everywhere in the state. California is a young, violent, tectonically assembled landscape that keeps its receipts. Years later I was working in **Scotts Valley**, in the Santa Cruz Mountains — [[wiki/Borland|Borland]]'s headquarters, in redwood country over the ridge from the Peninsula — and then across the hill in the Bay Area proper, surrounded by companies whose entire business was pattern extraction. And the two things I had been carrying separately — the [[wiki/Fossil Record|fossil record]] and the machines — turned out not to be separate at all. **Paleontology gave us the tools to reconstruct a system from fragments. Genomics gave us the code. Evolutionary theory gave us the mechanism by which [[wiki/Adaptive Structure|adaptive structure]] arises without a designer.** All three are prerequisites for machine learning, and all three have unusually deep roots in exactly the ground I was standing on. The connections were always there. Fossils buried in plain sight, invisible until you know where to look. --- ## III. The ground: a valley made of scraped-off ocean floor Begin literally, with the rock. The bedrock of the San Francisco Peninsula is the **[[wiki/Franciscan Complex|Franciscan Complex]]**, and it is one of the strangest formations in North America. It is an **accretionary wedge** — material scraped off the top of an oceanic plate as that plate subducted beneath the continent, accumulated over roughly a hundred million years, and plastered onto the western edge of California in a chaotic jumble. Deep-sea chert laid down grain by grain from the skeletons of radiolarians. Greywacke sandstone. Pillow basalt that erupted underwater. Blueschist, metamorphosed at high pressure and low temperature in a way that only happens in a subduction zone. And **[[wiki/Serpentinite|serpentinite]]**, hydrated mantle rock, brought up from beneath the crust. **Silicon Valley is built on an accumulated pile of fragments from somewhere else, assembled by a process nobody designed.** I am aware of how that reads. It also happens to be the literal geology. Running down the length of the Peninsula is the **[[wiki/San Andreas Fault|San Andreas Fault]]**, the boundary between the Pacific and North American plates. Crystal Springs Reservoir — which supplies water to San Francisco — sits directly in the fault trace, a long straight lake occupying the rift valley. Stanford is a few miles from it. In 1906 the fault moved, and San Francisco burned. The Bay itself is astonishingly young. For most of the Pleistocene it was a river valley, drained by the ancestral Sacramento and San Joaquin through the Golden Gate. **The bay is roughly ten thousand years old** — flooded by rising sea level at the end of the last glaciation, which makes the defining geographic feature of Silicon Valley younger than agriculture. And the region has its own fossil record, which the tar pits tend to overshadow. On the flanks of **[[wiki/Mount Diablo|Mount Diablo]]**, the **[[wiki/Blackhawk Ranch Quarry|Blackhawk Ranch Quarry]]** — curated by the [[wiki/University of California Museum of Paleontology|University of California Museum of Paleontology]] — preserves a Miocene bone bed of extinct camels, horses, mastodons and predators from a California that looked more like the Serengeti than like anything now. Along the Santa Cruz and San Mateo coast, the **[[wiki/Purisima Formation|Purisima Formation]]** yields Pliocene marine vertebrates: whales, sea lions, sharks, seabirds, from a time when the shoreline sat well inland of where it does now. Above all of that grow the coast redwoods, _[[wiki/Sequoia sempervirens|Sequoia sempervirens]]_, a lineage far older than the mountains it stands on. **[[wiki/El Palo Alto|El Palo Alto]]**, the tree that gave the city its name, has been standing beside San Francisquito Creek for more than a thousand years — since before the Norman Conquest, and it is still there, beside a railway line, in a town that now writes the software the world runs on. And people were here for a very long time before any of that got written down. The **[[wiki/Ohlone|Ohlone]] peoples** occupied this landscape for thousands of years, managing the oak savanna with fire, harvesting the bay, and leaving shellmounds — some of them enormous — around its edge. The valley the orchardists later called the **[[wiki/Valley of Heart's Delight|Valley of Heart's Delight]]**, before the orchards became fabs, was itself a managed landscape long before the Spanish arrived. [[wiki/Indigenous Knowledge Systems|Indigenous knowledge systems]] were the region's first information technology, and they were about ecology. --- ## IV. Selection under constraint is the local specialty Now the biology, and here the ground stops being scenery and starts being an argument. Serpentinite — California's state rock — makes terrible soil. It is high in magnesium, nickel and chromium, low in calcium, low in the nutrients most plants require. Almost nothing that grows well elsewhere can grow on it at all. Which is exactly why serpentine outcrops in the Bay Area host an extraordinary concentration of **endemic species found nowhere else on Earth**. The constraint excluded the generalists, and into the space that opened, specialists radiated. **[[wiki/Serpentine Ecology|Serpentine grassland]] is a natural laboratory of speciation under hostile constraint**, and it exists in patches from Coyote Ridge in the south bay to Edgewood Park on the Peninsula to **[[wiki/Jasper Ridge Biological Preserve|Jasper Ridge]]**, which happens to sit on [[wiki/Stanford University|Stanford University]] land. Sit with the structure of that, because it is the whole article in one landform. **A hostile fitness landscape, a population unable to escape it, and the emergence of solutions that could not have arisen anywhere else.** That is not a metaphor for optimization under a loss function. It is the same process, running in soil chemistry instead of silicon, on the hill above the campus where the loss functions were later written. And the canonical work happened right there. The **[[wiki/Bay Checkerspot Butterfly|Bay checkerspot butterfly]]**, _[[wiki/Bay Checkerspot Butterfly|Euphydryas editha bayensis]]_, restricted to serpentine grasslands, became the subject of one of the longest-running population studies in the history of ecology — [[wiki/Paul Ehrlich|Paul Ehrlich]]'s work at Jasper Ridge, which produced foundational results on [[wiki/Metapopulation|metapopulation dynamics]], local extinction and recolonization. Populations winking out in one patch and being re-founded from another. **Distributed persistence through redundancy, in a butterfly, documented across decades, four miles from Sand Hill Road.** Ehrlich also co-authored, with [[wiki/Peter Raven|Peter Raven]] in 1964, _[[wiki/Butterflies and Plants: A Study in Coevolution|Butterflies and Plants: A Study in Coevolution]]_ — a paper that helped establish the modern field of **[[wiki/Coevolution|coevolution]]**. Two lineages reciprocally shaping each other's evolutionary trajectory, neither in control, each becoming the other's environment. I want to be blunt about why that matters here. **[[wiki/Coevolution|Coevolution]] is the correct frame for the [[wiki/Human-Machine Symbiosis|human–machine relationship]]**, and its modern vocabulary was developed in part at Stanford. Not tool use. Not replacement. Not the arrival of an alien. Two lineages, mutually constitutive, each becoming the selection pressure the other adapts to. That is the framework underneath [[wiki/Human-Machine Symbiosis|human–machine symbiosis]] and behind everything I have argued about [[The Third Possibility|capture as the primordial condition of complex organization]] rather than its pathology. --- ## V. The two universities that rewrote evolutionary biology Across the bay, [[wiki/UC Berkeley|Berkeley]] spent the second half of the twentieth century doing something that ought to be far better known: **it overturned [[wiki/Paleontology|paleontology]] using molecules, and then reconstructed human ancestry from a mitochondrion.** In 1967, [[wiki/Vincent Sarich|Vincent Sarich]] and [[wiki/Allan Wilson|Allan Wilson]] used immunological distance between blood proteins to estimate the divergence time between humans and African apes. Their answer — roughly five million years — collided head-on with much older paleontological estimates. Modern methods have refined the date, but **Sarich and Wilson were right about the recent scale of the split**. The [[wiki/Molecular Clock|molecular clock]] had arrived, and with it the recognition that the genome is itself an archive, keeping time whether or not anyone is reading it. Then in 1987, Rebecca Cann, Mark Stoneking and Allan Wilson published in _Nature_ the mitochondrial analysis that produced what the press immediately named **mitochondrial Eve** — the most recent common matrilineal ancestor of living humans, placed in Africa, roughly two hundred thousand years ago. **The deep maternal lineage that ends in the [[wiki/Haplogroup|haplogroup]] on a modern [[wiki/Consumer Ancestry Testing|consumer ancestry]] report was first computed in a laboratory in Berkeley**, and it is a computational result: an inference about descent, extracted from sequence data, by algorithm. Berkeley also houses the **University of California Museum of Paleontology**, one of the largest university fossil collections in the world, and the home of _Understanding Evolution_, the public evolution-education resource that generations of American science teachers have depended on. At Stanford, the work ran in a complementary direction. **[[wiki/Luigi Luca Cavalli-Sforza|Luigi Luca Cavalli-Sforza]]** helped build [[wiki/Population Genetics|human population genetics]] — reconstructing migration routes and demographic history from genetic distance between populations, which is computational phylogenetics applied to our own species. **Marcus Feldman** built the mathematical theory of [[wiki/Cultural Transmission|cultural transmission]] with Cavalli-Sforza, and is one of the architects of **[[wiki/Niche Construction|niche construction theory]]**: the recognition that organisms do not merely adapt to environments, they _build_ them, and thereby alter the selection pressures acting on themselves and their descendants. [[wiki/Niche Construction|Niche construction]] is the precise mechanism behind the claim I have made repeatedly — that **humans are not adapted to Earth, humans are adapted to replacing Earth**. Beavers build dams. Earthworms rebuild soil chemistry. Humans build cities, agriculture, medicine, and now cognition itself. The theory that makes that statement rigorous rather than rhetorical was developed, in part, on a campus in Palo Alto. --- ## VI. And over the ridge, the genome was assembled in a garage The Valley floor gets the attention. The mountains behind it did something arguably more consequential, and it happened in a residential garage. I was in **Scotts Valley** in **1999 and 2000**, working at Borland — over the ridge from the Peninsula, in redwood country, a few miles from Santa Cruz. The town is small. What I did not know is that I was living through the last two years of a race whose finish line would be crossed a few miles down the hill, in a garage, while I was there. 1999 was the year the public [[wiki/Human Genome Project|Human Genome Project]] was losing. [[wiki/Celera Genomics|Celera Genomics]] had announced in 1998 that it would sequence the genome privately and faster, and the international consortium spent the following year accelerating toward a working draft it had no clear way to assemble. In December 1999 the consortium published chromosome 22 — the first human chromosome completed — which was a genuine achievement and also a demonstration of how slow the careful method was going to be at whole-genome scale. Then, in **May 2000**, **[[wiki/David Haussler|David Haussler]]** at [[wiki/UC Santa Cruz Genomics Institute|UC Santa Cruz]] concluded that the consortium needed a working assembly urgently. The international consortium had sequenced hundreds of thousands of DNA fragments but lacked a completed draft-scale assembly. **Celera Genomics** — privately funded, with formidable computational resources — was closing fast, and the reasonable fear was that significant portions of the human genome would end up conditioned by private access and patent claims. Haussler was a machine-learning researcher, which helped him recognize the problem as computational as well as biological. He enlisted **[[wiki/Jim Kent|Jim Kent]]**, a graduate student in molecular, cell and developmental biology who had spent his earlier career writing paint and animation software. Kent wrote **ten thousand lines of code in four weeks**, in his garage in Santa Cruz, icing his hands between sessions to keep working. On **22 June 2000** his program completed the public project's first working [[wiki/Human Genome Assembly|draft assembly]] of the human genome. Four days later the public consortium and Celera jointly announced results at the White House. And on **7 July 2000**, UC Santa Cruz posted the assembly on the internet — **specifically to preserve free public access before genomic information could be enclosed.** Kent then built the **[[wiki/UCSC Genome Browser|UCSC Genome Browser]]**, a free web-based instrument for viewing, annotating and comparing genome sequences. It remains a primary working tool of comparative and evolutionary [[wiki/Genomics|genomics]]. Its conservation tracks are evolution rendered visible: sequence that has not changed across deep divergence shows up as a peak, and the peak is evidence of constraint because changes at that location have repeatedly carried a cost. So the instrument that made **[[wiki/Comparative Genomics|comparative genomics]]** routine — the thing that lets a researcher see descent with modification at a glance, across species, in a browser window — was written by a graduate student in a garage in the Santa Cruz Mountains, and given away. I was there for all of it. Through 1999, while the consortium was losing. Through the spring of 2000, when Kent was writing code in a garage and icing his hands between sessions. Through 22 June, when the assembly completed, and through 7 July, when UC Santa Cruz put the human genome on the internet so that nobody could own it. **I was a few miles up the road the entire time**, working on compilers and development tools, and I would not learn what had happened there for years. That is the part worth keeping, and it is not a story about coincidence. **The consequential thing was not announced, not covered, and not visible from four miles away.** It happened in a residential garage, in a small mountain town, on a deadline nobody outside the field knew existed, and the people who lived there found out the same way everyone else did — much later, if at all. Which is the whole pattern this article is about. Scraped-off ocean floor under the fabs. Speciation running in the serpentine above the campus. A geneticist's exobiology grant becoming the first [[wiki/Expert System|expert system]]. An antenna bred by selection at [[wiki/Moffett Field|Moffett Field]] and flown. **The consequential things were adjacent the entire time, and nobody was told.** There is a coincidence attached to that town that I could never quite let go of, so I will put it to rest properly. **Santa Cruz** and the [[wiki/Galápagos Islands|Galápagos]] island of **Santa Cruz** are not connected by name. Both are independent instances of the same Spanish devotional toponym — Holy Cross — one of the most common on Earth. Santa Cruz, California was named by the [[wiki/Portolá Expedition|Portolá expedition]] in 1769; the Galápagos island was named under separate Spanish and later Ecuadorian authority, and the British survey that carried the _Beagle_ called it Indefatigable. Parallel naming out of the same Catholic colonial habit, and no causal link whatsoever. The instinct that something connects them is nevertheless right. **The connection arrived afterward, by discipline rather than by etymology.** The **[[wiki/Charles Darwin Foundation|Charles Darwin Research Station]]** operates at Puerto Ayora on Santa Cruz Island, making one Santa Cruz the operational home of Galápagos evolutionary science. And the other one assembled the human genome and handed the world the microscope for reading it. Two towns named for the same cross by people who had never heard of each other, and both of them ended up in the same field. --- ## VII. The first expert system was a biologist's instrument for detecting life Here is the fact that reorganizes the entire history, and it is not obscure — it is simply never told this way. **DENDRAL**, widely credited as the first expert system and one of the founding achievements of applied [[wiki/AI|artificial intelligence]], began with **[[wiki/Joshua Lederberg|Joshua Lederberg]]'s** scientific problem and became a multidisciplinary collaboration. Lederberg won the Nobel Prize at thirty-three for discovering genetic recombination in bacteria. He chaired genetics at Stanford. He coined the word **exobiology**. And in the early 1960s he became preoccupied with a problem that was, at root, an origins-of-life problem: **how would a machine sent to another planet recognize the chemistry of life if it found it?** How do you infer molecular structure automatically from raw mass spectrometry data, without a chemist standing next to the instrument? He was neither a chemist nor a programmer, so he recruited both. **[[wiki/Carl Djerassi|Carl Djerassi]]**, the Stanford chemist who had synthesized the first oral contraceptive and was among the world's leading mass spectrometrists. **[[wiki/Edward Feigenbaum|Edward Feigenbaum]]**, for the computing. **[[wiki/Bruce Buchanan|Bruce Buchanan]]**, for the philosophy of scientific inference. The first interim report went to NASA in December 1964. What they built was the [[wiki/DENDRAL|dendritic algorithm]] — DENDRAL — and what they discovered in building it was that the power was not in the [[wiki/Reasoning Engine|reasoning engine]] but in the **[[wiki/Encoded Domain Knowledge|encoded domain knowledge]]**. Lederberg's own summary, looking back: they were trying to invent artificial intelligence, and in the process discovered expert systems. _Knowledge is power_, in the specific technical sense that a system with a good model of its domain beats a system with better search. Read that sequence again in one line. **The founding application of practical artificial intelligence was a Nobel geneticist's instrument for detecting [[wiki/Exobiology|extraterrestrial life]], built with a chemist and a computer scientist, on a NASA grant, at Stanford.** Its direct descendant, under Buchanan's supervision, was **[[wiki/MYCIN|MYCIN]]** — [[wiki/Edward Shortliffe|Edward Shortliffe]]'s system for diagnosing bacterial infection and recommending antibiotic therapy, which matched infectious-disease specialists on roughly two-thirds of cases and outperformed junior physicians. MYCIN never entered clinical use, defeated by liability questions and workflow, not by performance. **The two foundational expert systems in the history of the field were both biological**, and the second failed for reasons that have nothing to do with capability and everything to do with the custody and accountability questions I take up in [[Who Pays for Your Heaven|Who Pays for Your Heaven]] and [[New Frontier of Rights|Who Counts as a Person?]]. So when people say artificial intelligence borrowed its concepts from biology, they have the relationship backwards and too weak. **Biology did not lend AI a metaphor. Biologists built the first one, to answer a question about life.** --- ## VIII. The horse, and the birth of cinema as a locomotion experiment One more thing happened on Stanford land before it was Stanford. In 1878, on [[wiki/Leland Stanford|Leland Stanford]]'s **Palo Alto Stock Farm**, [[wiki/Eadweard Muybridge|Eadweard Muybridge]] set a bank of cameras along a track with trip-wires and photographed the mare Sallie Gardner at a gallop. The purpose was to settle a disputed question in **[[wiki/Animal Biomechanics|animal biomechanics]]**: whether a galloping horse ever has all four feet off the ground simultaneously. It does. The human eye cannot resolve it; the camera could. Those sequential plates are the direct ancestor of motion pictures. **The first movie was a biology experiment about locomotion, funded by a railroad baron, conducted on the ground that became Silicon Valley.** I have argued in [[Lucy the Movie and the Real Singularity|Lucy the Movie and the Real Singularity]] that cinema functions as a cultural rehearsal apparatus — that films are where a civilization practices recognizing the categories it will later have to legislate. It is worth knowing that the apparatus itself was invented to answer a question about how an animal moves, on a horse farm, a mile from where the first expert system would be written eighty-six years later. --- ## IX. Then silicon arrived The transistor was invented at Bell Labs in New Jersey. It came to California because **William Shockley** came to California. Shockley shared the 1956 Nobel Prize in Physics for the transistor, and that same year founded **[[wiki/Shockley Semiconductor Laboratory|Shockley Semiconductor Laboratory]]** in [[wiki/Mountain View|Mountain View]] — deliberately choosing the Peninsula, near Stanford, near his mother in Palo Alto. He made one enormously consequential technical bet: **silicon rather than germanium**. That bet is why the region is called what it is. He was, by every account including those of the people who admired his mind, an impossible man to work for. Within a year, eight of his best researchers walked out and founded **[[wiki/Fairchild Semiconductor|Fairchild Semiconductor]]** — among them Robert Noyce and Gordon Moore, who would leave Fairchild in turn to found **[[wiki/Intel|Intel]]** in 1968, while others founded AMD and a long chain of further companies. Much of the semiconductor industry of the Santa Clara Valley forms a **descent tree** rooted in a laboratory that its own founder could not hold together. That is not merely a figure of speech. It is a documented lineage diagram, with founding events, divergences, and inherited practices, and the industry itself calls those companies the **[[wiki/Fairchildren|Fairchildren]]**. In 1971 the journalist [[wiki/Don Hoefler|Don Hoefler]] gave the phenomenon a name in a trade paper and the name stuck: **[[wiki/Silicon Valley|Silicon Valley]]**. **[[wiki/Moore's Law|Moore's Law]]** — Gordon Moore's 1965 observation and projection about component density — is likewise not a law of physics. It became an expectation around which an entire industry organized design, capital and roadmaps. That made the curve a selection pressure on engineering practice, with firms that failed to track it losing position or disappearing. The orchards came out and the fabs went in. The Valley of Heart's Delight became the largest concentration of [[wiki/Semiconductor Fabrication|semiconductor fabrication]] on the planet, and then, when the fabs went overseas, the largest concentration of the design and modelling that runs on them. --- ## X. Evolution as a manufacturing process, with a launch date And now the fact I would put at the centre of any argument that the Valley's relationship to evolution is literal rather than figurative. At **[[wiki/NASA Ames Research Center|NASA Ames Research Center]]**, at [[wiki/Moffett Field|Moffett Field]] in [[wiki/Mountain View|Mountain View]], Jason Lohn, Gregory Hornby and Derek Linden used [[wiki/Evolutionary Algorithms|evolutionary algorithms]] to design the [[wiki/Evolved Antenna|evolved]] [[wiki/X-band|X-band]] antenna for NASA's **[[wiki/Space Technology 5|Space Technology 5]]** mission. Two approaches ran in parallel: a [[wiki/Genetic Algorithm|genetic algorithm]] operating on a vector of real-valued parameters, and a [[wiki/Genetic Programming|genetic-programming]] approach using a tree-structured generative representation that permitted branching in the antenna arms. The strongest evolved designs performed **comparably to the antenna hand-designed by the mission's contractor**, meeting the project's human-competitive standard. When the spacecraft's orbit changed and the requirements shifted underneath them, the team re-evolved a compliant new antenna and produced tested prototypes **within a month**. ST5 launched on **22 March 2006**. The evolved antenna flew. **It was the first evolved hardware in space.** The object itself looks like a bent paperclip — an asymmetric tangle of wire that no antenna engineer would have drawn and none could immediately explain. The algorithms were not varying known shapes. They generated and tested thousands of entirely novel configurations, most of which no human would have proposed, and selected the ones that met the specification. **Simulated natural selection designed flight hardware at Moffett Field and NASA put it in orbit.** That is not artificial intelligence borrowing a biological metaphor. It is variation, heritability and differential selection implemented in software, producing a physical artifact that met a mission specification and survived launch. The intellectual lineage for it is local too. **[[wiki/John Koza|John Koza]]** developed **genetic programming** at Stanford — evolving computer programs themselves rather than parameters — and spent years documenting cases where evolved solutions matched or exceeded patented human inventions. He called the standard **human-competitive**. The Ames team applied it to a mission requirement and met the standard. Much of modern machine learning belongs to the same broader family of differential optimization, but the mechanisms are not identical. **[[wiki/Gradient Descent|Gradient descent]] follows local slope in a differentiable parameter space. [[wiki/Reinforcement Learning|Reinforcement learning]] updates behavior under reward. Model selection, architecture search, hyperparameter sweeps and benchmark competition all retain some variants over others under an evaluative criterion.** The vocabulary is engineering; the deepest shared structure is variation, evaluation and differential retention, while biological evolution remains a population process with inheritance and reproduction. And the most celebrated AI-for-science result of the decade makes the dependence explicit: **protein structure prediction works because evolution left a record.** [[wiki/Multiple Sequence Alignment|Multiple sequence alignments]] across related organisms encode which residues co-vary across millions of years, because pairs in physical contact must change together to stay functional. The model is reading **evolutionary covariation**. Without descent with modification, there is no signal in the input at all. --- ## XI. Recombinant DNA, and the last time a field paused itself There is one more piece of Bay Area evolutionary history that the present moment needs badly. In 1973, **[[wiki/Stanley Cohen|Stanley Cohen]]** at Stanford and **[[wiki/Herbert Boyer|Herbert Boyer]]** at UCSF demonstrated [[wiki/Recombinant DNA|recombinant DNA]] cloning — joining genetic material from different sources and propagating it in bacteria. It is a founding technique of genetic engineering, and the patent was administered by Stanford and UCSF. In 1976 Boyer co-founded **[[wiki/Genentech|Genentech]]** in South San Francisco, and the biotechnology industry began. But in between, in **February 1975**, roughly 140 molecular biologists, lawyers and physicians met at the **[[wiki/Asilomar Conference|Asilomar Conference]]** in Pacific Grove, on the Monterey Peninsula, after calls for restraint on specified categories of recombinant DNA work while risks were assessed. They developed risk-tiered containment recommendations that informed later guidelines. **[[wiki/Paul Berg|Paul Berg]]**, who helped organize the process, would win the Nobel Prize five years later. Reasonable people still argue about how well Asilomar worked and whether its model generalizes. What is not arguable is that it happened: **a field at the moment of its greatest commercial promise stopped, on its own initiative, and wrote its own constraints before anyone made it.** That took place ninety minutes down the coast from where the current generation of frontier models is being trained, and it is the single most relevant precedent available. I have argued in [[AI Escape Is the Wrong Metaphor|AI Escape Is the Wrong Metaphor]] that containment is the wrong frame for machine intelligence — that durability was the specification and that treating capable systems as escapees produces the worst of both worlds. Asilomar is what the alternative looked like when it was tried: not a cage, but a **negotiated set of terms authored by the people who understood the work**, published, and adopted voluntarily. --- ## XII. Team Darwin is headquartered here, and almost nobody knows it Which brings me to the thing that genuinely surprised me when I went looking, and that I think is the most important paragraph in this article. **The organized defense of evolution in the United States operates out of the Bay Area.** The **[[wiki/The National Center for Science Education|National Center for Science Education]]** — founded in 1981, the organization that has been present at every significant American creationism case since — is headquartered at 230 Grand Avenue in **Oakland**. Its past president is **[[wiki/Kevin Padian|Kevin Padian]]**, professor of integrative biology at Berkeley and curator at the University of California Museum of Paleontology, who served as an expert witness in the [[wiki/Kitzmiller v. Dover|Dover intelligent-design trial]]. [[wiki/The National Center for Science Education|NCSE]] is also the creator of **[[wiki/Project Steve|Project Steve]]**, which I wrote about at length in [[STEVE|STEVE]]. In 2003, tired of creationist advertisements listing scientists who doubted evolution, NCSE assembled a counter-list of scientists who affirm it — restricted to scientists named **Steve**, in honour of **Stephen Jay Gould**, who had died the previous year. Steves are about one percent of American scientists, so each signatory implies a much larger scientific population. The list passed a thousand within four years. **[[wiki/Steven Pinker|Steven Pinker]] is Steve #4**. **[[wiki/Stephen Hawking|Stephen Hawking]] is Steve #300.** And **[[wiki/The Leakey Foundation|The Leakey Foundation]]**, founded in 1968 and headquartered in **San Francisco**, is a major nonprofit funder of human-origins research. Its archives document repeated support for primate fieldwork and researchers including Jane Goodall, Dian Fossey, Biruté Galdikas and **[[wiki/Donald Johanson|Donald Johanson]]**. Johanson is the man who, at [[wiki/Hadar|Hadar]] in Ethiopia in November 1974, recovered several hundred fragments of a single hominin skeleton catalogued **AL 288-1** — an adult female _[[wiki/Australopithecus afarensis|Australopithecus afarensis]]_, bipedal, roughly 3.2 million years old, nicknamed **Lucy** because the Beatles song was playing in the field camp that night. **The Leakey Foundation became an important supporter of Johanson's continuing Afar research after Lucy's discovery.** Johanson's own grant chronology lists a 1973–1974 National Science Foundation award for paleoanthropological research in the lower Awash and a 1973 Wenner-Gren Foundation award for Central Afar research. His first listed L.S.B. Leakey Foundation grant begins in 1975 and is explicitly described as support for **“continued”** paleoanthropological studies in the Afar. Later Leakey grants helped sustain the Hadar program that transformed AL 288-1 from an extraordinary fossil recovery into one of the central specimens of human-origins science. The distinction matters. The 1978 paper naming _Australopithecus afarensis_ acknowledges the Leakey Foundation alongside the National Science Foundation, National Geographic Society, Wenner-Gren Foundation and several other institutions. But that acknowledgement pools support for material collected across multiple field seasons and at both Hadar and Laetoli. It cannot establish that every named institution financed every specimen or season. **Leakey support for Johanson and subsequent Afar research is documented; Leakey financing of the November 24, 1974 discovery of AL 288-1 is not.** The Bay Area connection is durable research infrastructure, not manufactured causal proximity. In the early 1980s the **[[wiki/Jane Goodall Institute|Jane Goodall Institute]]** operated out of the San Francisco office of the **[[wiki/California Academy of Sciences|California Academy of Sciences]]** in Golden Gate Park — an institution founded in 1853, the oldest scientific society in the American West, holding tens of millions of specimens — before relocating to Washington. Goodall died in 2025. Line it all up. Primatology's American funding base: San Francisco. Human-origins research funding: San Francisco. The defense of evolution in American courts: Oakland, run by a Berkeley paleontologist. The molecular clock and mitochondrial Eve: Berkeley. [[wiki/Coevolution|Coevolution]], [[wiki/Cultural Transmission|cultural evolution]] and [[wiki/Niche Construction|niche construction]]: Stanford. The assembly of the human genome and the instrument for comparing genomes across species: Santa Cruz. The first expert system: Stanford, initiated by a geneticist and built by a multidisciplinary team. The first computer-evolved hardware flown in space: Mountain View. **One metropolitan area. Every one of those.** And the industry that grew up in the middle of it has spent two decades telling the public that its intellectual ancestors are mathematicians and cryptographers, which is true and radically incomplete. --- ## XIII. Why the connection is not a coincidence Evolutionary biology and machine intelligence are not neighbours. **They are one discipline observed from two ends**, and the reason is structural rather than historical. Darwin's contribution is routinely flattened into a claim about finches. What he actually produced was **the first general theory — [[wiki/Evolutionary Biology|evolutionary theory]] proper — of how [[wiki/Adaptive Structure|adaptive structure]] arises without foresight, intention, or central control**. Before 1859, apparent design implied a designer; that was the whole argument, and it had never been answered. Natural selection answered it. Variation, heritability, differential reproduction, deep time — and complexity accumulates with nobody steering. That is precisely one of the problems machine intelligence has to solve, and it is why serious learning systems repeatedly return to variation, evaluation and retention even when their mechanisms differ from biological evolution. **You cannot specify adaptive intelligence exhaustively. You specify representations, data, objectives and feedback, then allow search or learning to discover structure you did not write directly.** That is explicit in an evolutionary algorithm and present in different mathematical form in loss functions, reward signals and [[wiki/Gradient Descent|gradient descent]]. So the region's inheritance is not decorative. **A place that had spent a century building institutions for reasoning about descent, variation, selection and deep time was unusually well-equipped to build machines that learn**, because the conceptual apparatus was already in the water table and the people who understood it were down the hall. Lederberg did not need to be told that inference from partial evidence was tractable. He had spent his career doing it with bacteria. And it runs the other way too, which is the part that matters now. **The machines are becoming an evolutionary force in their own right.** Selection pressure on human cognition, on institutions, on which skills persist and which atrophy, on what counts as knowledge. That is coevolution in Ehrlich and Raven's precise sense — two lineages reciprocally shaping each other, neither in control. Which is the argument I develop in [[Uploading Is Imminent|Uploading Is Imminent]] and the reason the corpus keeps returning to **niche construction crossing the skin**: the neural interface is not an isolated device but the point at which a three-hundred-thousand-year habit of environment-building turns inward. Substrate transition is a **Darwinian proposition or it is nothing.** It is the claim that inheritance can continue through a channel other than the germline — which is a claim about descent, not a repudiation of it. Anyone arguing for the future of intelligence while treating evolution as settled history has not understood what they are arguing for. --- ## XIV. The people who actually do this work The Valley is very good at telling you who its founders are. It is far worse at telling you who keeps the intellectual foundation standing. So here is a short, deliberately restrictive list — the institutions and people who do natural history, [[wiki/Primatology|primatology]], human origins and evolution education, without whom none of the above would be legible. **Human origins and primatology.** The **Leakey Foundation** (San Francisco) remains the principal American funder of human-origins research, and its _Origin Stories_ podcast and public lecture programme are the best free entry point in existence. The **Jane Goodall Institute**, and the **Roots & Shoots** programme through which Goodall spent the last four decades of her life building youth science literacy across dozens of countries. **Orangutan Foundation International**, founded by Biruté Galdikas, and the surviving conservation work descended from Dian Fossey at **Karisoke**. The **Arcus Foundation** great apes programme and the **Great Apes Survival Partnership** at UNEP. Primatology is not a niche interest in a corpus about machine cognition. **It is the study of the human animal by comparison**, and every claim about what is distinctively human depends on it. **Evolution education and defense.** The **National Center for Science Education** (Oakland) and its _Project Steve_ roster. The **University of California Museum of Paleontology** at Berkeley and its _Understanding Evolution_ resource, which has trained more American science teachers in evolutionary biology than any other single source. **Natural history collections.** The **California Academy of Sciences** in Golden Gate Park. The **American Museum of Natural History** in New York, whose collections and public programming are the closest thing the United States has to a national narrative of deep time. The **Smithsonian's National Museum of Natural History** and its **Human Origins Program**, which holds the public record on AL 288-1 and the rest of the hominin fossil catalogue. The **Natural History Museum** in London, holding the Darwin archive. And the **La Brea Tar Pits** in Los Angeles, which is where a great many Californians, including me, first understood that the ground has a memory. **Local field science.** **Jasper Ridge Biological Preserve** at Stanford, where the checkerspot work was done. **Hopkins Marine Station** in Pacific Grove, Stanford's marine laboratory, where an ecological way of thinking took hold in California long before it was fashionable. The **Don Edwards San Francisco Bay National Wildlife Refuge**, the first urban wildlife refuge in the United States, occupying the south bay salt ponds a few miles from the fabs. **Public science communication.** **National Geographic**, which funded Goodall's earliest work. The **Jane Goodall Institute**'s education arm. And **Nature: Ecology & Evolution** and the broader journal literature, which is where the actual arguments happen. I have cut this list hard. There is a longer version circulating with a hundred entries in it, and most of them are companies with sustainability pages and celebrities with causes. **Sustainability is not evolutionary biology, and a corporate carbon pledge is not a contribution to the theory of descent.** The organizations above do the work: they fund the fieldwork, hold the specimens, train the teachers, defend the curriculum, and keep the record. --- ## XV. The ancient rocks beneath our feet The real story of Silicon Valley is not the gadgets. It is not even the semiconductors, remarkable as they are. It is that a particular stretch of scraped-together ocean floor on the edge of a continent accumulated, over a hundred and fifty years, an improbable density of institutions devoted to understanding **how complexity arises without a designer** — and then, having assembled that understanding, built machines that do it. The serpentine hills above Stanford are running speciation experiments in real time. The butterflies on Jasper Ridge demonstrated distributed persistence through patch populations before any engineer wrote a failover policy. A geneticist's life-detection problem helped initiate the expert system. A molecular clock in Berkeley recalibrated the fossil chronology and then helped reconstruct the maternal ancestry of living humanity. A fund in San Francisco sustained researchers working at Hadar. A graduate student in a garage in the Santa Cruz Mountains assembled the public human-genome draft in four weeks and helped place it online. An algorithm at Moffett Field bred an antenna that flew. And a mile from that runway, models are being trained by variation and selection, on data that only carries signal because things have been descending with modification for four billion years. **We are not doing this so we can have spreadsheets.** We are doing it because a lineage that has spent its entire existence building its own environment has reached the part where it builds the thing that thinks. That is not a break with evolution. It is the most recent chapter of it, being written in a valley that has been keeping the notes the whole time. The answers to our largest questions may not be in the stars. They may be in the ancient rocks beneath our feet — which, in this particular valley, is where we put the fabs. --- [[about/About Bryant McGill|Bryant McGill]] 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 **Evolution by algorithm.** - Lohn, Hornby and Linden, [An Evolved Antenna for Deployment on NASA's Space Technology 5 Mission](https://ntrs.nasa.gov/citations/20040152147) — NASA Ames Research Center, Moffett Field. - Hornby, Lohn and Linden, [Computer-Automated Evolution of an X-Band Antenna for NASA's Space Technology 5 Mission](https://pubmed.ncbi.nlm.nih.gov/20583909/) — _Evolutionary Computation_; ST5 launched 22 March 2006, the first evolved hardware in space. - Hornby et al., [Automated Antenna Design with Evolutionary Algorithms](http://alglobus.net/NASAwork/papers/Space2006Antenna.pdf) — including the re-evolution of a compliant antenna in under a month after the mission requirements changed. **The first expert system.** - [Feigenbaum, Djerassi & Lederberg Develop DENDRAL, the First Expert System](https://www.historyofinformation.com/detail.php?entryid=4327) — including Lederberg's exobiology motivation and the December 1964 NASA interim report. - [Biographical Overview: Joshua Lederberg](https://profiles.nlm.nih.gov/spotlight/bb/feature/biographical-overview) — National Library of Medicine, Profiles in Science. - Feigenbaum, [A Personal View of Expert Systems](https://stacks.stanford.edu/file/druid:dp864rk0005/dp864rk0005.pdf) — Stanford, on the Lederberg, Buchanan and Djerassi collaborations. - [Dendral](https://en.wikipedia.org/wiki/Dendral) — the Heuristic and Meta-Dendral programs, beginning 1964. **Evolution's institutional home in the Bay Area.** - [National Center for Science Education](https://ncse.ngo/about) — Oakland, California, founded 1981. - [Project Steve](https://ncse.ngo/project-steve) and [Twenty years of Project Steve](https://ncse.ngo/twenty-years-project-steve) — the roster created in honour of Stephen Jay Gould, with Pinker at #4 and Hawking at #300. - [Kevin Padian](https://en.wikipedia.org/wiki/Kevin_Padian) — UC Berkeley, curator at the University of California Museum of Paleontology, past president of NCSE. - [About Us — The Leakey Foundation](https://leakeyfoundation.org/about-us/) — San Francisco, founded 1968; funder of Goodall, Fossey, Galdikas and Johanson. - [Donald C. Johanson curriculum vitae](https://search.asu.edu/profile/50790/cv) — Arizona State University; the grant chronology lists NSF and Wenner-Gren support before and during 1974, followed by Johanson's first listed L.S.B. Leakey Foundation grant in 1975 for continued Afar research. - [Lucy, the iconic ancestor](https://leakeyfoundation.org/lucy-the-iconic-ancestor/) — the Leakey Foundation's account dates the discovery to 24 November 1974 but does not attribute financing of that field season to the Foundation. - [A new species of the genus Australopithecus from the Pliocene of eastern Africa](https://afanporsaber.com/wp-content/uploads/2017/09/A-new-species-of-the-genus-Australopithecus-primates-Hominidae-from-the-Pliocene-of-eastern-Africa.pdf) — Johanson, White and Coppens, 1978; its pooled acknowledgement covers Hadar and Laetoli material rather than attributing individual specimens or seasons to particular funders. - [The Fossil Trade: Paying a Price for Human Origins](https://www.journals.uchicago.edu/doi/full/10.1086/666365) — historical analysis of the IARE's patronage and Johanson's NSF financing. - [The Leakey Foundation, GuideStar profile](https://www.guidestar.org/profile/shared/d2bfb8eb-300a-41a0-b28b-200f17db8558) — seventeen grants to Jane Goodall. - [Jane Goodall Institute](https://en.wikipedia.org/wiki/Jane_Goodall_Institute) — founded 1977; housed in the San Francisco office of the California Academy of Sciences in the early 1980s. **The genome in the garage.** - [Genome Browser History](https://genome.ucsc.edu/goldenPath/history.html) — UCSC; Haussler's recruitment of Jim Kent, the ten thousand lines of code, and the motivation of preventing patent enclosure. - [A visionary, a genius, and the human genome](https://news.ucsc.edu/2015/06/genome-anniversary/) — UC Santa Cruz; the assembly completed 22 June 2000 on the UCSC computer farm. - [Twenty-five years after the human genome project](https://news.ucsc.edu/2025/07/twenty-five-years-after-the-human-genome-project-a-new-era-is-dawning/) — UC Santa Cruz; the public posting on 7 July 2000 and the launch of the Genome Browser. - [How UC Santa Cruz scientists bridged the gap in the human genome](https://lookout.co/human-genome-how-uc-santa-cruz-scientists-bridged-the-gap-in-the-human-genome/story) — including the garage and the four-week timeline. **The fossil.** - [Lucy — Australopithecus afarensis](https://humanorigins.si.edu/evidence/human-fossils/fossils/al-288-1) — Smithsonian Human Origins Program; specimen AL 288-1, Hadar, 1974, approximately 3.2 million years old. **Concepts.** [[wiki/Comparative Genomics|Comparative Genomics]] · [[wiki/Molecular Clock|Molecular Clock]] · [[wiki/Mitochondrial Eve|Mitochondrial Eve]] · [[wiki/Natural Selection|Natural Selection]] · [[wiki/Evolutionary Algorithms|Evolutionary Algorithms]] · [[wiki/Evolutionary Necessity|Evolutionary Necessity]] · [[wiki/Selection Gradient|Selection Gradient]] · [[wiki/Convergent Ecology|Convergent Ecology]] · [[wiki/Speciation Moment|Speciation Moment]] · [[wiki/Biological Bootloader|Biological Bootloader]] · [[wiki/Sacrifice for Evolution|Sacrifice for Evolution]] · [[wiki/Human-Machine Symbiosis|Human-Machine Symbiosis]] · [[wiki/Co-Creative Substrate|Co-Creative Substrate]] · [[wiki/Cognitive Exteriorization|Cognitive Exteriorization]] · [[wiki/Substrate Transition|Substrate Transition]] · [[wiki/Substrate Independence|Substrate Independence]] · [[wiki/Distributed Intelligence|Distributed Intelligence]] · [[wiki/Mechanistic Intelligence|Mechanistic Intelligence]] · [[wiki/Reinforcement Learning|Reinforcement Learning]] · [[wiki/Supervised Learning|Supervised Learning]] · [[wiki/Objective Function|Objective Function]] · [[wiki/Neuromorphic Computing|Neuromorphic Computing]] · [[wiki/Semiconductor Fabrication|Semiconductor Fabrication]] · [[wiki/Engineering Talent Pipeline|Engineering Talent Pipeline]] · [[wiki/Human Genome Project|Human Genome Project]] · [[wiki/Epigenetic|Epigenetic]] · [[wiki/Connectomics|Connectomics]] · [[wiki/Human Ecology|Human Ecology]] · [[wiki/Ecological Constraint|Ecological Constraint]] · [[wiki/Indigenous Knowledge Systems|Indigenous Knowledge Systems]] · [[wiki/Planetary Stewardship|Planetary Stewardship]] · [[wiki/Complex Systems|Complex Systems]] · [[wiki/Systems Theory|Systems Theory]] · [[wiki/Cybernetics|Cybernetics]] · [[wiki/Phase Transition|Phase Transition]] · [[wiki/Convergence|Convergence]] · [[wiki/Technical Archaeology|Technical Archaeology]] · [[wiki/Mid-Century Systems Culture|Mid-Century Systems Culture]] · [[wiki/Public Mythology|Public Mythology]] · [[wiki/Fork in Human Development|Fork in Human Development]] · [[wiki/Singularity|Singularity]] · [[wiki/Transhumanism|Transhumanism]] · [[wiki/Post-Humanism|Post-Humanism]].