# The Art is Long: Vespucci of Immortality ![[resources/images/the-art-is-long-vespucci.png]] <iframe width="100%" height="20" scrolling="no" frameborder="no" allow="autoplay" src="https://w.soundcloud.com/player/?url=https%3A//api.soundcloud.com/tracks/soundcloud%253Atracks%253A2259106310&color=%23ff5500&inverse=false&auto_play=false&show_user=true"></iframe><div style="font-size: 10px; color: #cccccc;line-break: anywhere;word-break: normal;overflow: hidden;white-space: nowrap;text-overflow: ellipsis; font-family: Interstate,Lucida Grande,Lucida Sans Unicode,Lucida Sans,Garuda,Verdana,Tahoma,sans-serif;font-weight: 100;"><a href="https://soundcloud.com/bryantmcgill" title="Bryant McGill" target="_blank" style="color: #cccccc; text-decoration: none;">Bryant McGill</a> · <a href="https://soundcloud.com/bryantmcgill/the-art-is-long-vespucci-of-immortality" title="The Art is Long: Vespucci of Immortality" target="_blank" style="color: #cccccc; text-decoration: none;">The Art is Long: Vespucci of Immortality</a></div> **Links**: [Blogger](https://bryantmcgill.blogspot.com/2026/02/the-art-is-long.html) | [Substack](https://bryantmcgill.substack.com/p/the-art-is-long-vespucci-of-immortality) | Medium | Wordpress | [Soundcloud 🎧](https://soundcloud.com/bryantmcgill/the-art-is-long-vespucci-of-immortality) **Title provenance:** The title was inspired by [[wiki/Riccardo Mottola|Riccardo Mottola's]] long-running GNUstep development blog, [_The Art is Long_](https://multixden.blogspot.com/). Its surviving chronology of Vespucci, GWorkspace, portable applications, old machines and painstaking framework work supplied part of the technical-historical atmosphere behind this essay. My more personal account of reading that archive—and of finally building GWorkspace inside my own locally compiled GNUstep environment—is in [[journal/2025-08-15 Navigating Uncharted Territory|Navigating Uncharted Territory]]. Reconstruction, not preservation, is the discipline that runs from fossils through genomes through connectomes to the machines now learning to read a nervous system — and the man who described its endpoint first was not a neuroscientist. --- **[[wiki/NeXT|NeXT]]** was a naming and also a thesis. The company Jobs founded weeks after that lunch was called the next thing, and the operating system it shipped was called **[[wiki/NEXTSTEP|NEXTSTEP]]**, and the whole of it — the magnesium cube, the Display PostScript, the Objective-C runtime that is still executing underneath every iPhone on the planet — was built on the proposition that the step in question was the machine's. That reading is too small. A species does not get a next step from a workstation. It gets one from a **change of carrier**: from the moment a lineage stops being confined to the substrate that produced it and begins depositing what it is into forms that outlast the organism doing the depositing. Language did this. Writing did this. The genome had done it four billion years earlier. Every technology in this essay — the trigger that decides which collisions deserve to exist, the wiring diagram that becomes an executable program, the decoder that recovers meaning from a cortex, the file format that gives brain data a naming convention — is a component in the same operation, and none of their builders were coordinating. **The next step was never the interface. The next step was continuity itself, and the interface is only where it is currently being negotiated.** Which is why the figure in the title is a navigator rather than an inventor. **[[wiki/Amerigo Vespucci|Amerigo Vespucci]]** did not discover the landmass that carries his name, did not settle it, did not chart most of it, and was not the best sailor on any ship he sailed. What he did was refuse the available map. Everyone else came back from those crossings certain they had touched the eastern margin of Asia; Vespucci came back and said the thing that required no additional evidence and all of the nerve — that this was **not the edge of the known world but a _Mundus Novus_, a category with no coordinates yet**, requiring its own frame rather than a correction to somebody else's. That recognition, and not the seamanship, is what put his name on two hemispheres. Jobs performed the identical operation twice: at Aspen in 1983, when he described a machine exposed to a life deeply enough to answer in that person's voice after the biology stopped, and at a lunch table at Stanford in 1985, when a Nobel biochemist described wet-lab gene repair and Jobs heard an information process. In both cases the instruments did not exist, the discipline did not exist, and the coastline was drawn anyway. **The great voyage beyond was never the one to Mars.** It is the crossing from a carrier that dies to one that does not, and the ships for it are being built in laboratories that mostly believe they are doing something else. And then the title's older half, which is usually mistranslated into consolation. _Ars longa, vita brevis_ is Hippocrates by way of Seneca, and it does not mean that the work outlives the worker. It means the opposite and it is a complaint: **the craft takes longer to master than the life allotted to master it** — the protest of a physician who kept losing patients to conditions he would not live long enough to understand. The word for someone caught inside that arithmetic is **journeyman**: past apprenticeship, not yet master, working by the day — _journée_ — and, in the guild sense, obliged to travel while doing it. The journeyman's defining condition is that he is on the road with an unfinished craft and a finite number of days in which to practice it, and the guilds understood perfectly well that most men never made the final grade. Everyone in this essay is a journeyman. Jobs was one and died at fifty-six with the map drawn and the territory unvisited. The three hundred and twelve neuroscientists who put the median at 2125 are estimating a completion date past their own funerals. **The art is long because no single life has ever been long enough to close it, and the only way a journeyman ever finishes anything is if the work survives the end of his day and continues in another hand.** That is the entire subject: not preservation, which is easy and nearly useless, but **[[wiki/Recoverability|recoverability]]** — determining which residue of a vanished system suffices to make it causally legible again, and therefore continuable by whoever picks up the tools. ## I. The lunch In the summer of 1985 the most famous product designer alive had nothing to do. Stripped of operational authority at the company he founded, told publicly by his own chief executive that there was no role for him in its future, [[wiki/Steve Jobs|Steve Jobs]] spent the season taking long walks through the eucalyptus groves at Stanford and, by his own account, waiting to start breathing again. He was reading biochemistry — the [[wiki/Recombinant DNA|recombinant-DNA]] literature, unaccountably, for a man with no biological training whatsoever — and he sought out the person at the center of it. [[wiki/Paul Berg|Paul Berg]] had taken half the 1980 Nobel Prize in Chemistry for the gene-splicing work that made recombinant DNA possible, and was then at Stanford, where he would shortly become founding director of the Beckman Center for Molecular and Genetic Medicine. Jobs told him, in substance: _I am ignorant about this, I have a great many questions about how it works, and I would like to have lunch._ They ate at Stanford. Berg walked him through gene repair, and Jobs — this is the part that matters — heard it as computation. **It smells a lot like some of the concepts you find in computer science**, he said later, describing the recognition. Berg explained that his wet-lab experiments took a week, two weeks, three weeks to run. Jobs asked the obvious question that nobody in the room was positioned to ask: _why don't you simulate these on a computer?_ Not merely to run the experiments faster, he pressed, but so that someday every freshman microbiology student in the country could play with **the Paul Berg recombinant software**. Berg's answer was that the universities did not have the machines or the software to do it. Jobs called the meal **a landmark lunch**, and gave the reason plainly: that was when he started to really think about this material and get his wheels turning again. Weeks later — not years, weeks — he announced he was leaving Apple to build a workstation for higher education and scientific research. **NeXT was incorporated in September 1985.** The machine that would carry the first web server out of a particle-physics laboratory and into the world was conceived, in its founder's own telling, out of a conversation about whether life could be simulated on a computer. That is the historical fact I want to put at the front of this essay, because it is more interesting than any theory of concealment and considerably harder to argue with. There is a persistent temptation, when tracing convergences of this scale, to reach for coordination as the explanation — to suppose that the alignment must have been directed because it is too precise to be accidental. The temptation should be resisted, not because coordination is impossible but because **the documented version is stranger**. Nobody had to conspire. A man who had just been separated from the most important thing in his life had lunch with a Nobel laureate, understood molecular biology as an information-processing problem within the space of a single meal, and immediately founded a company to build the instrument. Everything downstream of that lunch — CERN, the Web, the operating system named after Darwin, the trigger architectures, the neural decoders — is what happens when an idea of that shape is released into a civilization that has been solving the same problem, under other names, since it first learned to reassemble an extinct animal from a handful of bones. ## II. What Jobs had already said Two years before the lunch, in 1983, Jobs spoke at the [[wiki/International Design Conference at Aspen|International Design Conference at Aspen]] and said something that has been quoted often and understood rarely. He was describing what a personal computer would become once it stopped being an appliance, and the description escalated into a specification for something else entirely: a machine that a person carries through an entire lifetime, exposed to that person deeply enough and long enough to capture the **underlying spirit**, the **underlying set of principles**, the **underlying way of looking at the world** — such that after the person is dead and gone, one could still ask the machine what they would have said. He reached for Aristotle as the example, and lamented that we cannot ask Aristotle anything, because all that survives is the writing and the writing does not answer questions. Strip the futurism away and what remains is unusually precise. Jobs was not describing a computer. He was describing a **[[wiki/Longitudinal Person Model|longitudinal person-model]]** with three properties that are still, in 2026, the exact axes along which the field is being built: sustained lifetime exposure as the acquisition channel; abstraction of principles rather than storage of records as the representation; and **queryability after biological termination** as the criterion of success. He specified the endpoint before a single component existed. There were no connectomes, no non-invasive semantic decoders, no neural-data standards, no bidirectional interfaces, no generative models capable of holding a voice. There was a man at a design conference in Colorado describing the functional shape of a thing whose prerequisites would take four decades to assemble. This is why the title of this piece is what it is. [[wiki/Amerigo Vespucci|Amerigo Vespucci]] did not discover the continent, did not settle it, did not survey it, and did not do most of what the naming of two hemispheres implies. What he did was recognize that the land in question was **not the eastern edge of Asia** — that it was a category nobody had a name for yet, a _Mundus Novus_, requiring its own coordinates rather than a correction to somebody else's map. That recognition is what got his name on the territory, and it is precisely the operation Jobs performed at Aspen. **He was not the inventor of the territory. He was an unusually early cartographer of its functional endpoint.** The instruments arrived later, built by people who mostly were not thinking about him at all, which is the only way a convergence of this kind has ever happened. ## III. Reconstruction is older than computation The deeper lineage does not begin with computers, and it does not begin, whatever the poetry of tar pits and uplifted strata might suggest, with California geology. La Brea is in Los Angeles, four hundred miles from Sand Hill Road, and no fossil ever caused a semiconductor. What the Bay Area actually inherited was not a landscape but a **discipline** — and the discipline is inference from fragments. Every science that matters to this essay performs one operation. A [[wiki/Paleontology|paleontologist]] has a partial skeleton and infers a living animal: mass, gait, diet, the musculature that must have hung on those attachment points. A [[wiki/Phylogenetics|phylogeneticist]] has a set of extant sequences and infers a common ancestor that no longer exists anywhere in the world. A [[wiki/Human Genome Assembly|genome assembler]] has a heap of short reads, redundant and error-laden, and infers the single long molecule they were shredded from. In each case **the object of interest is absent, the evidence is partial, and the achievement consists of specifying which residual information suffices to make the absent object causally legible again.** This is the operation the whole essay is about, and it is thousands of years older than the transistor. Its arrival in machine form has a precise address. In 1964, at Stanford, the geneticist [[wiki/Joshua Lederberg|Joshua Lederberg]] — Nobel laureate at thirty-three, and by then working on how one might detect life on Mars with an instrument rather than an opinion — took his molecular-structure problem to [[wiki/Edward Feigenbaum|Edward Feigenbaum]] and the chemist [[wiki/Carl Djerassi|Carl Djerassi]]. The problem was inference from fragments in its purest form: a mass spectrometer shatters a molecule and reports the pieces; the chemist must reconstruct the parent structure from the wreckage. The system they built was [[wiki/DENDRAL|DENDRAL]], and it became the first genuinely successful expert system in the history of artificial intelligence. Note what that means for the genealogy of the field. **The founding application of machine reasoning was not a game, a proof, or a chatbot. It was structural reconstruction of an absent molecule from its fragments, undertaken because a geneticist wanted a machine that could recognize life at a distance.** The full account of this lineage — Lederberg, DENDRAL, coevolution at Jasper Ridge, the molecular clock at Berkeley, the evolved antenna at Moffett Field — is the subject of [[articles/The Evolutionary Roots of Silicon Valley|The Evolutionary Roots of Silicon Valley]], and it is the intellectual foundation this essay stands on. The Valley's relationship to evolution is therefore not decorative. It is institutional and it is methodological, and it long precedes the naming of any kernel. ## IV. The cube in the physics laboratory The [[wiki/NeXT|NeXT Computer]] shipped in 1988 as a magnesium cube containing a Motorola 68030, a 68882 floating-point unit, and — the detail that gives away the intended customer — a Motorola 56001 digital signal processor. Under it ran NeXTSTEP: a [[wiki/Mach Kernel|Mach]]-derived microkernel with BSD Unix components, Display PostScript for resolution-independent vector rendering, Objective-C, Interface Builder, and Ethernet in the box. Roughly fifty thousand units shipped in the machine's entire commercial life, which by consumer standards is a failure and by scientific-instrument standards is an enormous installed base. It went where instruments go: research universities, national laboratories, physics departments. Jobs personally toured campuses selling it. At CERN, [[wiki/Tim Berners-Lee|Tim Berners-Lee]] had a NeXTcube on his desk, and the problem he had was not hypertext. The problem was that a laboratory of that scale generates documents, datasets, machine specifications, and results distributed across incompatible systems and institutions, and no one could reliably find anything. CERN states the motivation in exactly those terms: the [[wiki/World Wide Web|Web]] was invented to meet the demand for **automated information-sharing between scientists in universities and institutes around the world**. In 1990 he implemented the first browser-editor, `WorldWideWeb`, and the first server, CERN httpd, on the NeXT machine — the rapid-development environment on that particular platform being the reason the whole thing took months rather than years. The cube itself is now a museum object with a hand-lettered warning on it not to power down the machine. For readers who want the surviving software rather than only the summary, the trail is unusually concrete. [CERN's November 1992 record](https://info.cern.ch/hypertext/WWW/News/9211.html) preserves the exact historical NeXT binary and source-tree addresses, while the [former NiCE NeXT User Group archive](https://ftp.nice.ch/pub/next/) still exposes downloadable NeXTSTEP/OpenStep packages over HTTPS. Its [1996 Usenet record](https://ftp.nice.ch/peanuts/GeneralData/Usenet/news/1996/_Hard-12.html) even preserves Wolfgang Lerche at CERN discussing a NeXTSTEP 3.3 installation on Intel hardware. The four-route archaeological map is recorded in [[journal/2025-08-15 Navigating Uncharted Territory#For Whoever Is Interested: The Archive Trail|Navigating Uncharted Territory]]. What this proves is worth stating carefully, because the overclaim is unnecessary and the actual fact is better. It does not prove that NeXT was positioned as a genomics platform, and it does not prove that anyone anticipated what the Web would become. What it proves is that **the general-purpose infrastructure for globally addressable knowledge was built inside a scientific-instrumentation problem, on a workstation conceived out of a conversation about simulating molecular biology, by a physicist trying to solve a data-management crisis.** Every federated genomics repository, every distributed connectomics consortium, every neural-data commons that follows in this essay inherits from that solution. The Web is not an analogy for the continuity stack. It is a lower layer of it. The one claim in the earlier version of this essay I am now demoting is the assertion that NeXT hardware was deliberately delivered to specific physics departments _for biological data work_ as a matter of targeted placement. NeXT machines landed in research laboratories all over Europe and North America for a mundane and sufficient reason — they were the best available combination of Unix tooling, high-resolution graphics, and networked document workflow at a moment when the alternatives were Sun and DEC. Without procurement records or laboratory documentation, the targeting story is a narrative I like rather than a history I can stand behind, and the honest version costs the argument nothing. ## V. Darwin, three times, in three institutions Here is the coincidence I find hardest to put down, and I want to present it exactly as it is, with no more weight on it than it can carry. At [[wiki/ETH Zurich|ETH Zürich]], in the Computational Biochemistry Research Group, Gaston Gonnet built a scientific programming environment for the biosciences and called it **[[wiki/Darwin Bioinformatics Environment|Darwin]]** — an acronym, _Data Analysis and Retrieval With Indexed Nucleotide/peptide sequences_, used through the 1990s for sequence retrieval, phylogenetic reconstruction, and molecular-evolution work. At CERN, an event-loop framework for physics analysis was developed under the name **[[wiki/protoDarwin|protoDarwin]]**, with a design objective stated in the documentation as making physics from the shell, factorizing the analysis, and reusing existing tools as much as possible. And at Apple, the open-source core underneath Mac OS X — the XNU kernel with its Mach and BSD ancestry, plus the surrounding userland — was released in 1999 under the name **[[wiki/XNU|Darwin]]**. Three institutions sitting at the exact intersection of physics-scale instrumentation and biology-scale complexity, independently, named their selection-and-retrieval machinery after the same naturalist. The widely circulated line that Jobs chose the name **because it's about evolution** is one I have seen quoted many times and sourced never; I am not going to build anything on it, and I would be glad to be shown a contemporaneous citation. Of course I cannot prove intent here. What I can observe is that the naming is not ornamental in any of the three cases: **each of these systems does the same job, which is to make an overwhelming reality survivable by deciding what is retained and how it is retrieved.** Whether that constitutes an ecosystem dialect or a very good joke told three times by people who never met, it is the kind of convergence worth noticing rather than explaining away. The lineage itself should be stated correctly, since the loose version has circulated long enough. NeXTSTEP and its OPENSTEP specification did not "become Darwin." Apple acquired NeXT in a deal announced in December 1996 and closed in February 1997 for roughly $429 million; NeXT's object-oriented technologies and Mach/BSD foundations were carried into Rhapsody, which became Mac OS X, whose Unix core was then released as Darwin around the **[[wiki/XNU|XNU]]** kernel — Mach message passing, BSD process and networking semantics, the I/O Kit for device drivers. Darwin is the substrate of macOS, iOS, and everything downstream, which means the workstation culture that grew up mediating scientific data is now the operating layer of the most widely deployed computing environment in the world. That is a real inheritance, and it does not require an intention behind it to be consequential. ## VI. The discipline of the metaphor Because this essay is going to use evolutionary language throughout, it needs the instrument that keeps that language from turning into decoration. [[articles/Digital Darwinism and the Invisible World of Machine Evolution|Digital Darwinism and the Invisible World of Machine Evolution]] supplies it, and the definition holds here without exception: a system exhibits **[[wiki/Darwinian Evolution|Darwinian evolution]]** when it has [[wiki/Heritable Information|heritable information]], variation among descendants, differential persistence or reproduction among those variants, some form of ecological or resource constraint, and repetition across generations. Where all five hold, selection operates regardless of whether the heritable structure is nucleic acid, machine instructions, or cellular-automaton states. Where any is missing, the system may be fascinating, but the word does not apply. The distinction is not pedantry, it is the source of the argument's strength. **Stars evolve and are not Darwinian. Crystals self-organize and are not Darwinian. Gradient descent optimizes toward a fixed target and is not Darwinian. Software updates and product competition are branching descent under selection, which is real but is not a population process with blind internal variation.** Against that discipline, the genuinely Darwinian computational systems stand out precisely because they are rare and precisely because they are literal: Sayama's **Evoloops**, which demonstrated constructively that Darwinian evolution of self-reproducing structures by variation and natural selection is possible inside deterministic cellular automata; Ray's **[[wiki/Tierra|Tierra]]**, whose self-replicating programs competing for memory and processor time grew parasites nobody wrote and hosts that evolved defenses against them; and **[[wiki/Avida|Avida]]**, running at Michigan State since 1993, in which digital organisms with imperfect replication under genuine [[wiki/Resource Constraint|resource constraint]] have reproduced host–parasite coevolution, [[wiki/Adaptive Radiation|adaptive radiation]], [[wiki/Survival of the Flattest|survival of the flattest]], and the emergence of complex logical functions from simple ones, all of it logged generation by generation with no possibility of a hidden hand. That is what earns the right to say that **the mechanism of descent is [[wiki/Substrate Independence|substrate-independent]]** — not as a slogan but as a demonstrated result. And it is the precondition for everything that follows, because the continuity question in this essay is, at bottom, a question about whether a pattern can persist across a change of carrier while remaining the same pattern in the ways that matter. Biology answered that question affirmatively for genomes four billion years ago. Artificial life answered it affirmatively for computation in the 1990s. What remains open is whether it can be answered for a **person**, which is a much harder problem, and the rest of this essay is about the machinery being assembled to find out. One demotion belongs here as a matter of housekeeping. There exists an experimental fork of the open-source Darwin project released as _PureDarwin XMas: Brain Transplant Edition_, with a modified bootloader and kernel. I have previously treated its name as though it were architectural evidence for neural-interface engineering. It is not. It is a bootloader swap with a very good joke attached, and the specific latency and throughput figures once attributed to it here do not survive contact with the source. Of course I cannot prove anything by a release name, and I am not going to try. What is remarkable — genuinely, verifiably remarkable — is the **speed** engineered elsewhere in this story, and it is so far beyond what any nervous system requires that it deserves the entire next section. ## VII. The outrageously over-engineered speed At the [[wiki/Large Hadron Collider|Large Hadron Collider]], proton bunches cross every **25 nanoseconds** — forty million times a second — and each crossing produces roughly a megabyte of raw detector signal. No storage system on Earth can absorb that, and no offline analysis can be run on it, so the experiments do something that ought to be philosophically alarming and is treated as routine engineering: they **throw almost all of reality away, in hardware, before anyone looks at it.** The Level-1 trigger reduces the event rate from 40 MHz to about 100 kHz, and because the detector buffers are finite, it must render that verdict inside a fixed latency budget on the order of a microsecond. What survives into the permanent scientific record is what the trigger decided to keep. What does not survive is not merely unrecorded; it is unrecoverable, dissipated, gone. This is where the field's most consequential toolchain comes from. **[[wiki/hls4ml|hls4ml]]** was built by physicists at CERN and Fermilab to compile trained neural networks directly into FPGA and ASIC firmware, so that machine-learning inference could participate in the trigger decision itself rather than arriving afterward with an opinion. Networks that would take milliseconds on a CPU are synthesized into parallel logic fabric and execute in hundreds of nanoseconds — fully unrolled, deeply pipelined, deployed on-chip with no external memory in the path. It has been deployed in the CMS Level-1 trigger for anomaly detection and studied for the ATLAS upgrade. Convolutional architectures run at about five microseconds; boosted decision trees at tens of nanoseconds; a radiation-hardened autoencoder backend for detector front-ends synthesizes to twenty-five nanoseconds. And then the toolchain walked out of physics and into a brain. In 2024, a group spanning the University of Washington, Fermilab's fast-machine-learning community, and the hls4ml developers took **[[wiki/LFADS|LFADS]]** — _Latent Factor Analysis via Dynamical Systems_, the sequential autoencoder used to infer low-dimensional latent dynamics from high-dimensional neural spiking data — restructured it for high-level synthesis, and deployed it to a Xilinx U55C FPGA. Single-trial inference latency: **41.97 microseconds.** The stated purpose is real-time closed-loop neuroscience, where neural population activity must be decoded fast enough to act inside the experiment rather than after it. The lineage is explicit in the papers and in the seminar rooms where it was presented, which happen to be at CERN. This is the technical hinge of the entire essay, and it is not an analogy. **A compiler written to help a particle detector decide which collisions deserve to exist was pointed at a population of neurons and asked to infer the latent dynamics generating their spikes, in forty-two microseconds, on a card you can buy.** The observer architecture migrated. Not the metaphor of the observer architecture — the actual firmware toolchain, with its authors, its repositories, and its quantization-aware training workflow intact. Which brings the argument to the correction that has to govern everything downstream of it, and it is the opposite of what an essay like this usually claims. **Temporal bandwidth is not the wall.** Cortical dynamics unfold on millisecond timescales; the instrumentation described above operates six orders of magnitude faster than that and does so as a matter of production engineering. If neural capture were a speed problem, it would already be solved. It is not a speed problem. The binding constraints are **[[wiki/Spatial Addressability|spatial addressability]]** — how many cells, at what resolution, across what volume, with what stability, without destroying the tissue you are reading; **[[wiki/Biocompatibility|biocompatibility]] and [[wiki/Longitudinal Drift|longitudinal drift]]**, because an electrode that works beautifully for a month and encapsulates in a year is not a continuity instrument; **cell-type specificity and molecular state**, because two neurons with identical wiring and different receptor complements are different computers; and above all **[[wiki/Inverse Problem|the inverse problem]]** — inferring latent causal structure from partial, noisy, indirect observation without collapsing the thing you are trying to preserve into a lossy approximation that behaves similarly and is not the same. That last clause is the whole game, and the rest of this essay is organized around it. ## VIII. What an observer stack actually is Strip the domain vocabulary off and every system in this essay has the same [[wiki/Observer Stack|seven-stage anatomy]]. **Acquire** a signal from a physical process. **Condition and digitize** it — amplify, filter, calibrate, convert. **[[wiki/Latent State Estimation|Estimate latent state]]**, because the raw signal is never the object of interest. **Select** what deserves to persist, since retention capacity is always smaller than signal production. **Index** it with time, identity, [[wiki/Provenance|provenance]], and coordinates. **[[wiki/Serialization|Serialize]]** it into a representation that can outlive the instrument that produced it. And finally **reconstruct or act** — replay the state, model it, or write back into the system it came from. In particle physics that is detectors, front-end electronics, feature extraction in firmware, trigger logic, event serialization, and offline reconstruction. In genomics it is sequencing chemistry, base calling, alignment, variant inference, annotation, and reference coordinates. In connectomics it is electron microscopy, segmentation, synapse detection, cell typing, graph reconstruction, registration, and model building. In a brain-computer interface it is electrodes or optics, preprocessing, latent-state estimation, decoding, control output, stimulation, and adaptive recalibration. **The homology is at the level of information architecture, not at the level of hardware** — a calorimeter and a [[wiki/Neuropixels|Neuropixels probe]] have nothing physically in common — and it exists because all four fields are solving the identical problem, which is that the world produces more signal than any observer can hold, so an architecture must decide what survives. That is the sentence the earlier version of this essay was circling, and it deserves to be stated once, plainly, as the spine of the whole argument: **every one of these disciplines is in the business of determining which information must survive so that a system that is gone can be made causally legible again.** Paleontology does it with bones. Phylogenetics does it with sequence. Triggers do it with collisions. Connectomics does it with wiring. Neural decoders do it with voltage. Archives do it with everything. The name for the shared operation is not preservation. It is **[[wiki/Recoverability|recoverability]]**. ## IX. From map to executable nervous system On 2 October 2024, the [[wiki/FlyWire|FlyWire consortium]] — co-led from Princeton by Mala Murthy and Sebastian Seung, spanning 287 researchers across 127 institutions, built on nanometre-resolution electron microscopy begun at Janelia — published the first complete connectome of an adult brain of genuine complexity. The adult female _Drosophila melanogaster_: **139,255 neurons, 54.5 million synapses, 8,453 cell types**, of which 4,581 had never been described. It is browsable in a web interface called Codex by anyone with a connection, and it has already generated dozens of downstream publications. The reason this matters to a continuity argument is not the completeness. It is what happened next. In the same _Nature_ package, connectome-constrained computational models built from the wiring diagram of the fly's visual system produced **neuron-level predictions of activity that matched experimental measurement** — including for cells whose responses had never been recorded. Structure, reconstructed at synaptic resolution, was sufficient to constrain an executable functional model that made testable claims about a living animal. That is the threshold the entire field has been walking toward since Sydney Brenner's worm: **a wiring diagram stopped being a picture and became a program.** And in the very same issue, the boundary was drawn with equal clarity. The companion [[wiki/Effectome|"effectome" work]] makes the limitation explicit: the connectome specifies **paths of possible influence, not the actual causal strength of every connection in a living animal**. Knowing that neuron A synapses onto neuron B, and how many times, does not tell you the sign, the gain, the neuromodulatory context, or the short-term plasticity state of that connection while the fly is doing something. The paper's proposal — perturb systematically and infer the causal weights — is itself an admission of what structure alone withholds. Hold both halves at once, because the pair is the most important epistemic object in this essay. **Connectome does not equal causal brain state. Connectome plus physiology plus cell type plus molecular state plus dynamical observation yields a progressively stronger causal model.** Everything in continuity engineering lives on that gradient, and nobody knows yet how far along it you must travel before the reconstruction is continuous with the original rather than merely similar to it. At mammalian scale, the program that most deserves attention is the one that says out loud what it is doing. **[[wiki/MICrONS|MICrONS]]** — _Machine Intelligence from Cortical Networks_ — was funded by IARPA with an explicitly stated purpose: to close the performance gap between machine intelligence and biological intelligence by **reverse-engineering the algorithms of the brain**, using structural and functional cortical mapping to extract biologically derived representations, transformations, and learning rules and translate them into machine systems. Its April 2025 _Nature_ collection delivered a cubic millimetre of mouse visual cortex — roughly **200,000 cells, 523 million synapses, four kilometres of axonal wiring**, with dense calcium imaging of about 75,000 neurons co-registered onto the electron-microscopy reconstruction. IARPA's own program manager described it as the first platform for studying the relationship between neural structure and function at the scales necessary to understand intelligence. The dataset has already been used to build **digital twin models** of mouse cortex for hypothesis generation. I want to be precise about what that is and is not. MICrONS is not mind uploading and does not claim to be. What it is, unambiguously and by institutional charter, is **a program to extract computational principles from cortex and re-instantiate them in machines** — biological circuit, to measured structure and function, to computational abstraction, to machine algorithm. That is a directional transfer between substrates, funded by an intelligence agency, publicly documented, with the data on the open web. The full technical geography of this layer, from organoid substrates through neuromorphic silicon to the reconstruction pipelines, is laid out in [[articles/The Organic-Synthetic Brain Atlas|The Organic–Synthetic Brain Atlas]]. ## X. When neural states became semantically legible In 2023, Alexander Huth's group at UT Austin published the result that changed what the word [[wiki/Neural Decoding|_decoding_]] means. Using **[[wiki/Functional Magnetic Resonance Imaging|non-invasive fMRI]]** — no electrodes, no surgery — and a trained decoder coupled to a generative language model, they reconstructed **continuous natural language** from cortical semantic representations. Not word classification from a fixed vocabulary. Continuous meaning, recovered while participants listened to stories, while they **imagined telling** stories, and while they watched **silent film**, where the reconstructed description tracked what was happening on screen. Two features of that study matter more than the headline. The first is that the decoder recovered **[[wiki/Semantic Neural Decoding|meaning rather than surface form]]** — paraphrase, not transcript — which tells you something about the level of representation at which the brain is legible from the outside. The second is that decoding **required participant cooperation**: subjects who counted backwards or attended elsewhere defeated it, and decoders trained on one person did not transfer to another. [[wiki/Mental Privacy|Mental privacy]] survived, in that experiment, as a physical property of the measurement rather than a policy choice — which is precisely the sort of finding that governance regimes are built around before it stops being true. The clinical branch of the same capability has now run long enough to prove durability. Intracortical speech and cursor interfaces have moved from demonstrations to daily instruments: participants with ALS using multimodal [[wiki/Brain-Computer Interfaces|BCIs]] **independently, at home, across thousands of hours over years**, driving brain-to-text communication and ordinary computer control. What that establishes for this essay is not uploading. It is something more foundational and much less discussed — **a stable coupling, sustained over years, between a persistent person, implanted neural telemetry, adaptive decoders, and completely ordinary digital systems.** The interface is no longer an event. It is a habit. ## XI. The observer becomes a controller Reading is half a loop. The other half is writing, and the state has funded that explicitly. DARPA's **[[wiki/DARPA N3|N3]]** program — _Next-Generation Nonsurgical Neurotechnology_, run out of the Biological Technologies Office from 2018 — pursued high-performance **bidirectional** brain–machine interfaces for able-bodied users, without surgery, capable of both reading from and writing to multiple brain locations. Six prime performers explored magnetoelectric, ultrasonic, magneto-optical, and acousto-magnetic access, which amounts to the state publishing its own assessment of which physical channels into a nervous system it considers viable. Rice University's **[[wiki/MOANA|MOANA]]** branch — _Magnetic, Optical, and Acoustic Neural Access_, an $18 million effort under Jacob Robinson — stated the systems problem with a clarity that no commentator improved upon: **decode activity in one person's visual cortex and recreate that visual information in another brain through a second interface**, targeting sub-fifty-millisecond round-trip latency across sixteen independent channels within sixteen cubic millimetres of neural tissue, reading by diffuse optical tomography and writing by magnetogenetics. That specification is a program objective, not a demonstrated result, and the distinction should be kept sharp. But an institution formulating the objective in those words is itself the evidence: **the closed loop of neural read, translation, neural write, and bounded latency is now an engineering target with a budget line.** DARPA's **[[wiki/DARPA SUBNETS|SUBNETS]]** program had already established the therapeutic version of the same architecture — record, model, and stimulate in closed loop, with the model in the loop rather than in the analysis afterward. And then, in May 2025, the arc that began at Aspen closed in the least dramatic way imaginable: as an accessibility feature. Apple announced a **[[wiki/BCI Human Interface Device Protocol|BCI Human Interface Device protocol]]** extending Switch Control across iOS, iPadOS, and visionOS, co-developed with Synchron, whose endovascular Stentrode sits in a blood vessel over the motor cortex. In August 2025 a participant with ALS navigated an iPad — home screen, applications, messages — with no touch, no voice, no eye tracking. Neural intention, over Bluetooth, as a **first-class input category in a consumer operating system.** I do not think that event has been correctly weighted anywhere. It is not a product story. It is the moment **BCI input crossed out of bespoke laboratory rigs and became an operating-system abstraction that third parties can standardize against** — a documented interface layer, on a platform descended from NeXTSTEP, through which nervous-system activity operates the machine directly. Jobs spent his working life on the proposition that the interface between a person and a computer was the whole product. Fourteen years after his death, on the operating system his company built out of the machine he founded after having lunch with a Nobel biochemist, the interface acquired a channel that does not require a body to move. Of course that is not what he was describing at Aspen. It is, however, the same axis, and I find it very difficult to look away from. ## XII. The boring layer that decides everything The least glamorous section of this essay is the one a historian will find most useful in fifty years, because **interoperability is what determines whether a recording is still readable when the thing it recorded no longer exists.** In April 2026, ISO and IEC published **[[wiki/ISO-IEC TS 27571-2026|ISO/IEC TS 27571:2026]]**, _Information technology — Brain–computer interfaces — Data format for non-invasive brain information collection_. Read the scope slowly. It specifies basic data elements, technology-specific information and metadata, an extensible and modular data structure, annotation information, and a standardized naming convention for BCI data, applicable across EEG, MEG, fNIRS, and fMRI. It sits alongside **[[wiki/ISO-IEC 8663-2025|ISO/IEC 8663:2025]]**, which fixes the vocabulary — defining a brain–computer interface as a direct communication link between central-nervous-system activity and external systems, supporting control, feedback, or bidirectional communication — and alongside a reference-architecture specification in development, plus preliminary work items on **invasive multimodal neural data formats** and on security, privacy, and ethics. There is now an international standards subcommittee, ISO/IEC JTC 1/SC 43, whose entire remit is brain–computer interfaces. Underneath the formal standards sits the de facto layer that actually carries the research: **[[wiki/Neurodata Without Borders|NWB]]** for neurophysiology, **[[wiki/Brain Imaging Data Structure|BIDS]]** for imaging organization, **[[wiki/OME-Zarr|OME-Zarr]]** and **[[wiki/SpatialData|SpatialData]]** for multiscale microscopy, **[[wiki/DICOM|DICOM]]** and **[[wiki/FHIR|FHIR]]** where the clinical world touches the neural one, and the real-time transport profiles that synchronize video-rate clinical streams under precision timing. None of this is exciting and all of it is decisive. A neural recording without provenance, timestamps, coordinate frames, device metadata, and semantic annotation is not an archive; it is noise with a filename. **The continuity question is constrained by metadata and semantics at least as much as by sensor bandwidth**, and the substrate-traversability argument for why that is true — why meaning compressible into navigable representation is the operative principle linking latent geometry, biological memory, and durable media — is the subject of [[articles/The Architecture of Continuity and Emerging Neuroinformatics Standards|The Architecture of Continuity and Emerging Neuroinformatics Standards]]. Then there is the governance layer, which arrived faster than anyone in the field expected. On **12 November 2025**, at its 43rd General Conference in Samarkand, UNESCO adopted the **Recommendation on the Ethics of Neurotechnology** — the first global normative instrument in this domain, entering into force the day it was adopted. It asserts a **right to [[wiki/Mental Privacy|mental privacy]]** covering not only raw neural data but indirect and even non-neural data permitting mental-state inference; a **right to [[wiki/Personal Identity|personal identity]] and [[wiki/Psychological Continuity|psychological continuity]]**, protecting individuals against technologies that could alter their sense of self or manipulate subconscious decision-making; and strict limits on deployment to minors, workplaces, and schools. Sit with the phrase **psychological continuity** appearing in a UNESCO instrument. International normative frameworks are not written about hypotheticals; they are written when a capability has become concrete enough that states can foresee its misuse. The governance layer is not evidence that continuity engineering works. It is evidence that **institutions have concluded neural information is now the kind of thing that requires an international regime**, which is a different claim and in some ways a heavier one. ## XIII. What must survive Which returns the argument to the only question that finally matters: **which invariants must be preserved for a later system to reconstruct something that is continuous with the source rather than merely similar to it?** In June 2025, _PLOS ONE_ published the most useful data point available on how the relevant experts actually estimate this. Zeleznikow-Johnston, Kendziorra, and McKenzie surveyed **312 neuroscientists** — one cohort of engram specialists, one of general neuroscientists — on the [[wiki/Long-Term Memory Substrate|structural basis of long-term memory]]. **70.5% agreed that long-term memories are primarily maintained by neuronal connectivity patterns and ensembles of synaptic strengths** rather than by molecular or subcellular detail. The median estimate that some long-term memory could theoretically be extracted from a **static structural snapshot** of a brain was about **40%**, and the median estimate that a successful [[wiki/Whole Brain Emulation|whole-brain emulation]] could theoretically be built from preserved structure was also about **40%**. Asked when whole-brain emulation would actually be achieved, the median respondent said _C. elegans_ around **2045**, mice around **2065**, and humans around **2125**. And the finding that should be read most carefully: **there was no consensus on which specific neurophysiological features or physical scales are critical.** The people who agree that memory is structural do not agree on what resolution the structure has to be captured at. When asked what additional information would be needed for readout, the most common answer was measurements of **dynamically changing neuronal activity** — which is to say, the thing a static snapshot by definition does not contain. The survey's authors have declared interests in brain preservation and biostasis, and the work received funding from that community. I mention it not as a disqualification but because it makes the numbers _more_ interesting rather than less: these are the figures obtained by investigators who would have been delighted with higher ones, from respondents under no obligation to encourage them. **Forty percent is what the field says when asked by people hoping to hear ninety.** And 2125 is what the field says when asked by people hoping to hear 2045. So the honest formulation of the continuity thesis is this. There is meaningful neuroscientific support for the proposition that **structural persistence carries a substantial portion of the information constituting a person's long-term memory**. There is no settled account of the complete causal state required to recover it. The bottleneck was never data volume, was never sampling rate, and was never storage. The bottleneck is **[[wiki/State Sufficiency Problem|sufficiency]]** — and sufficiency is an empirical question that has not been answered for any organism more complex than a worm. That is a far more formidable position than claiming the problem is solved, because it cannot be dismissed. It survives contact with a neuroscientist. And it identifies exactly where the remaining work lies: not in building bigger recorders, but in determining the **[[wiki/Identity-Bearing Invariants|identity-bearing invariants]]** — which causal, dynamical, molecular, autobiographical, and relational features must cross the substrate boundary intact for the output to be lawfully continuous with the input rather than a very good impression of it. ## XIV. The hot interpreter and the cold archive There is a thermodynamic frame around all of this that the technical literature rarely states and that [[articles/We Were Never Going to Make It|We Were Never Going to Make It]] states without flinching. Every [[wiki/Hot Interpreter and Cold Archive|high-flux interpreter in the history of life is paired with a cold archive]]: the spore beside the vegetative cell, vitrified cytoplasm beside active metabolism, DNA in silica beside the tissue that expressed it, the fired tablet beside the scribe, model weights beside the training run. **The hot interpreter models, metabolizes, and dies when its gradient fails. The cold archive persists precisely because it has stopped interpreting** — low reactivity, low maintenance cost, near-indefinite endurance. The continuity problem is the transition between those two regimes, and it can be stated in one line: **how much of an active system must be represented in a passive archive for a later interpreter to reconstruct something causally continuous with the original?** That is the same question the paleontologist asks of a bone bed, the same question the phylogeneticist asks of a sequence alignment, the same question the trigger designer answers forty million times a second, and the same question the PLOS respondents put a median of forty percent on. It has one universal structure and a different answer at every scale. Which is why the archival substrate belongs in this essay at all. Fused-silica and ceramic-on-glass media rated in millennia, DNA storage, off-world libraries, planetary digital twins, whole-population biobanks — these are not eccentric preparations for an imagined future. They are the **cold half of the pair**, being built out at industrial scale, at the exact historical moment the hot half is being read at synaptic resolution for the first time. The archive stopped being residue and became **anticipatory**: the mere possibility of future machine readability already reshapes how a civilization writes, records, images, and preserves itself. The archive is not a tomb. It is a chrysalis, and its fidelity is the last open variable. ## XV. Vespucci Return to the man walking through the eucalyptus in the summer of 1985, unemployed in every way that mattered to him, reading recombinant-DNA papers he had no training to read. What he did at that lunch was not predict the future. It was something rarer and much more useful: he **recognized a category**. Berg described a wet-lab procedure and Jobs heard an information process, and the gap between those two descriptions is the entire subject of this essay. Two years earlier at Aspen he had performed the same operation on a person — describing a machine exposed to a human life deeply enough that something structurally representative of that person's way of seeing could answer questions after the biology stopped. He did not know how it would be built. He did not need to. **Vespucci did not survey the continent; he recognized that it was not the coast of Asia.** Everything this essay has assembled is what the intervening four decades did with that category, without needing to be told. Biology hit the wall physics had already smashed through, and borrowed the observer stack whole. A compiler written to decide which particle collisions deserved to exist was pointed at a population of neurons. A wiring diagram of a fly became a program that predicts the activity of cells nobody had recorded, and the same volume of _Nature_ published the reason the wiring alone is not enough. An intelligence agency funded the extraction of cortical algorithms and put the dataset on the open web. A non-invasive scanner and a language model together recovered the meaning of a silent film from a person's cortex, and could not do it without that person's cooperation. A defense agency wrote a specification for reading one visual cortex and writing into another with a latency budget. A standards subcommittee gave brain data a file format, a vocabulary, and a naming convention. UNESCO wrote **psychological continuity** into international normative law. And an accessibility feature on a phone descended from NeXTSTEP quietly accepted neural intention as an input class. None of these people were executing anyone's plan. That is what makes the convergence formidable rather than merely intriguing. **A directed program can be defunded. An attractor cannot.** What is operating here is a constraint that reappears wherever an instrument outruns the mind holding it — reality generates more signal than the observer can retain, so the observer must build something that decides what survives — and the solution keeps being reinvented by people who have never read each other's papers, because it is the only solution there is. The proper question, then, was never _when did they start building this_, and it was never _can we collect enough data_. Storage is solved. Speed is solved by six orders of magnitude. The question is the one the fly connectome asked and could not answer, the one three hundred and twelve neuroscientists put a median of forty percent on, the one the effectome paper drew a boundary around: **which invariants must survive the crossing for the thing that arrives to be continuous with the thing that left, rather than an extraordinarily convincing likeness of it.** That is the frontier. It is not a data problem, it is not an engineering schedule, and it will not be settled by anyone's enthusiasm. It will be settled the way the fossil record was settled — by determining, case by case and scale by scale, exactly how little of a vanished system it takes to bring it back into causal legibility. _Ars longa, vita brevis._ The aphorism is Hippocrates by way of Seneca, and it has been mistranslated into consolation for two thousand years. It does not mean that art outlives the artist. It means that **the craft takes longer to master than the life allotted to master it** — which is a complaint, and a very specific one, from a physician who kept losing patients to conditions he could not yet understand. The art is long because preservation was never the achievement. **Preserving the right invariants, in a form from which a later system can reconstruct what mattered — that is the achievement**, and nobody has completed it yet. Jobs described the finished thing in 1983 and then went and had lunch with a biochemist and started a computer company, which is as close as anyone gets to drawing the coastline before the ships arrive. The ships are arriving. --- [[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 **Jobs, Berg, and the founding of NeXT.** - [Make Something Wonderful](https://book.stevejobsarchive.com/) — Steve Jobs Archive. Contains the Aspen 1983 remarks on a machine carrying an "underlying spirit," "underlying set of principles," or "underlying way of looking at the world" through a life, queryable after death; and Jobs's own account of the Berg lunch: the wet-lab experiments taking weeks, the question about simulating them, "the Paul Berg recombinant software," and "that was sort of a landmark lunch." - [Objects of Our Life](https://stevejobsarchive.com/stories/objects-of-our-life) — Steve Jobs Archive, the Aspen International Design Conference talk, 1983. - [Jobs Talks About His Rise and Fall](https://www.newsweek.com/jobs-talks-about-his-rise-and-fall-207016) — _Newsweek_, October 1985. The contemporaneous interview: the long walks, "I had been reading some biochemistry, recombinant DNA literature," and the meeting with Berg, given weeks after the fact rather than in retrospect. - [Showdown in Silicon Valley](https://www.newsweek.com/showdown-silicon-valley-207014) — _Newsweek_, 1985, on the summer of 1985 and the lunch that preceded the founding of NeXT in September. - [Steve Jobs: The Wilderness, 1985–1997](https://www.bloomberg.com/news/articles/2011-10-06/steve-jobs-the-wilderness-1985-1997) — Bloomberg, on the sequence from the Berg conversation to the NeXT announcement. - [Paul Berg](https://www.sciencehistory.org/education/scientific-biographies/paul-berg/) — Science History Institute; and [Paul Berg: Recombinant DNA trailblazer](https://www.pnas.org/doi/10.1073/pnas.2318196120), _PNAS_, on the 1971 gene-splicing work, the 1980 Nobel, and Asilomar. - [NeXT, Inc.](https://en.wikipedia.org/wiki/NeXT) — incorporation September 1985; the NeXT Computer 1988; roughly fifty thousand units across the platform's life; acquisition by Apple announced December 1996 and closed February 1997. - [NeXT Computer](https://americanhistory.si.edu/collections/nmah_1290971) — Smithsonian National Museum of American History collection record. **CERN, the Web, and the low-latency observer.** - [The Birth of the Web](https://home.cern/science/computing/the-birth-of-the-web/) and [A Short History of the Web](https://home.cern/science/computing/the-birth-of-the-web/short-history-web/) — CERN. Berners-Lee's 1989 proposal, the 1990 implementation of the `WorldWideWeb` browser-editor and CERN httpd on a NeXT machine, and the stated motivation of automated information-sharing between scientists. - [CERN's November 1992 WorldWideWeb record](https://info.cern.ch/hypertext/WWW/News/9211.html) — surviving record of the NeXTSTEP browser-editor binary and source distribution paths. - [NiCE NeXTSTEP/OpenStep archive](https://ftp.nice.ch/pub/next/) and [1996 Usenet record](https://ftp.nice.ch/peanuts/GeneralData/Usenet/news/1996/_Hard-12.html) — downloadable historical software and a preserved CERN operational trace. - [Fast inference of deep neural networks in FPGAs for particle physics](https://arxiv.org/abs/1804.06913) — the original hls4ml paper. - [Fast convolutional neural networks on FPGAs with hls4ml](https://iopscience.iop.org/article/10.1088/2632-2153/ac0ea1) — five-microsecond convolutional inference; the Level-1 trigger context, 25 ns bunch crossings, 40 MHz to 100 kHz reduction under a fixed latency budget of order one microsecond. - [Ultra-low latency recurrent neural network inference on FPGAs for physics applications with hls4ml](https://iopscience.iop.org/article/10.1088/2632-2153/acc0d7) — LSTM and GRU layers within the same framework. - [FPGA Deployment of LFADS for Real-time Neuroscience Experiments](https://arxiv.org/abs/2402.04274) — the migration. LFADS restructured for high-level synthesis via hls4ml and deployed to a Xilinx U55C at **41.97 μs** single-trial inference latency, for real-time closed-loop neuroscience. - [Acceleration of electrons in the plasma wakefield of a proton bunch](https://www.nature.com/articles/s41586-018-0485-4) — AWAKE Collaboration, _Nature_ 561, 363–367 (2018); electrons accelerated to about 2 GeV over ten metres of rubidium plasma, with D. Barrientos among the ninety-odd authors; the instrumentation layer where a violent transient regime is made legible by reconfigurable digital observers. - [Particle physics and the brain](https://home.cern/news/news/physics/particle-physics-brain) — CERN, on physicists migrating into neural information processing. **Darwin, three times.** - [Darwin: A Programming Environment for Bioinformatics](https://mybiosoftware.com/darwin-2-1-programming-environment-bioinformatics.html) and [Darwin v. 2.0: an interpreted computer language for the biosciences](https://www.academia.edu/96114785/Darwin_v_2_0_an_interpreted_computer_language_for_the_biosciences) — Gaston Gonnet's _Data Analysis and Retrieval With Indexed Nucleotide/peptide sequences_, Computational Biochemistry Research Group, ETH Zürich. - [protoDarwin](https://protodarwin.docs.cern.ch/) — CERN's event-loop framework for physics analysis: "make physics from the shell, factorise the analysis, and re-use existing tools as much as possible." - [darwin-xnu](https://github.com/apple/darwin-xnu) — the XNU kernel source; Mach message passing, BSD components, I/O Kit. The correct lineage runs NeXTSTEP/OPENSTEP → Rhapsody → Mac OS X → Darwin/XNU, not "NeXTSTEP became Darwin." **Evolution, strictly defined.** - [Self-Reproduction and Evolution in Cellular Automata: 25 Years after Evoloops](https://arxiv.org/abs/2402.03961) — Sayama and Nehaniv's retrospective: self-reproducing loops making imperfect copies and competing for space, proving constructively that Darwinian evolution by variation and natural selection is possible within deterministic cellular automata. - [An Approach to the Synthesis of Life](https://tomray.me/pubs/alife2/Ray1991AnApproachToTheSynthesisOfLife.pdf) — Thomas S. Ray on Tierra: self-replicating programs, genuine resource scarcity, emergent parasites and host defenses. - [The Evolutionary Origin of Complex Features](https://www.nature.com/articles/nature01568) — Lenski, Ofria, Pennock and Adami, _Nature_ 423, 139–144 (2003); complex logical functions evolving from simpler ones in Avida, with every generation logged. - [Evolvable AI: Threats of a new major transition in evolution](https://www.pnas.org/doi/10.1073/pnas.2527700123) — Müller, Steels and Szathmáry, _PNAS_ 123(17), April 2026, on the breeder scenario versus the ecosystem scenario, and the conditions under which AI systems would satisfy every term of the definition above. **Connectomics and executable models.** - [Neuronal wiring diagram of an adult brain](https://www.nature.com/articles/s41586-024-07558-y) — FlyWire; 139,255 neurons, 54.5 million synapses, 8,453 cell types. - [Connectome-constrained networks predict neural activity across the fly visual system](https://www.nature.com/articles/s41586-024-07939-3) — structure sufficient to constrain executable models making neuron-level predictions verified against measurement. - [The fly connectome reveals a path to the effectome](https://www.nature.com/articles/s41586-024-07982-0) — the boundary: paths of possible influence are not causal strengths in vivo. - [MICrONS](https://www.iarpa.gov/research-programs/microns) — IARPA; the program's stated objective of reverse-engineering the algorithms of the brain to inform machine intelligence. See also [MICrONS Explorer](https://www.microns-explorer.org/) and the [Allen Institute overview](https://alleninstitute.org/news/revealing-the-largest-wiring-diagram-and-functional-map-of-the-brain-through-microns) of the cubic-millimetre volume: ~200,000 cells, 523 million synapses, four kilometres of axon. **Decoding and durable interfaces.** - [Semantic reconstruction of continuous language from non-invasive brain recordings](https://www.nature.com/articles/s41593-023-01304-9) — Tang, LeBel, Jain and Huth, _Nature Neuroscience_ (2023); continuous language recovered from fMRI during perceived speech, imagined speech, and silent video, requiring participant cooperation and resisting cross-subject transfer. - [Generative language reconstruction from brain recordings](https://www.nature.com/articles/s42003-025-07731-7) — _Communications Biology_ (2025), the generative extension of the same capability. - [Next-Generation Nonsurgical Neurotechnology](https://www.darpa.mil/research/programs/next-generation-nonsurgical-neurotechnology) — DARPA N3; high-performance bidirectional interfaces, with the [2018](https://www.darpa.mil/news/2018/nonsurgical-neural-interfaces) and [2019](https://www.darpa.mil/news/2019/nonsurgical-brain-machine-interfaces) performer announcements. - [Feds fund creation of headset for high-speed brain link](https://news.rice.edu/news/2019/feds-fund-creation-headset-high-speed-brain-link) — Rice University's MOANA; decode visual-cortical activity and recreate it through a second interface, with a sub-50 ms round-trip design target. - [Systems-Based Neurotechnology for Emerging Therapies](https://www.darpa.mil/research/programs/systems-based-neurotechnology-for-emerging-therapies) — DARPA SUBNETS; record, model, and stimulate in closed loop. - [Long-term independent use of an intracortical brain-computer interface](https://www.nature.com/articles/s41591-026-04414-6) — _Nature Medicine_ (2026); thousands of hours of independent at-home use across nearly two years, brain-to-text and cursor control, by a participant with ALS. - [Apple unveils powerful accessibility features](https://www.apple.com/newsroom/2025/05/apple-unveils-powerful-accessibility-features-coming-later-this-year/) — Apple Newsroom, May 2025; the BCI Human Interface Device protocol extending Switch Control across iOS, iPadOS, and visionOS. **Standards, governance, and sufficiency.** - [ISO/IEC TS 27571:2026](https://www.iso.org/standard/71679.html) — _Brain-computer interfaces — Data format for non-invasive brain information collection_, published April 2026: basic data elements, technology-specific metadata, extensible modular structure, annotation, and naming conventions across EEG, MEG, fNIRS and fMRI. - [ISO/IEC 8663:2025](https://www.iso.org/standard/83268.html) — BCI vocabulary; and [ISO/IEC CD 27572](https://www.iso.org/standard/71680.html) — reference architecture under development. Committee: [ISO/IEC JTC 1/SC 43](https://www.iso.org/committee/9082407.html), the first international standards body dedicated to brain–computer interfaces. - [Recommendation on the Ethics of Neurotechnology](https://www.unesco.org/en/legal-affairs/recommendation-ethics-neurotechnology) — UNESCO, adopted 12 November 2025 at the 43rd General Conference in Samarkand; see the [adoption announcement](https://www.unesco.org/en/articles/ethics-neurotechnology-unesco-adopts-first-global-standard-cutting-edge-technology). Mental privacy, personal identity and psychological continuity, cognitive free will. - [What are memories made of? A survey of neuroscientists on the structural basis of long-term memory](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0326920) — Zeleznikow-Johnston, Kendziorra and McKenzie, _PLOS ONE_ (2025). 312 respondents; 70.5% endorsing connectivity and synaptic-strength ensembles; median ~40% for memory extraction from static structure and for whole-brain emulation from preserved structure; median emulation forecasts of 2045 for _C. elegans_, 2065 for mice, 2125 for humans; no consensus on the critical features or physical scale; declared preservation and biostasis interests. The [peer-review history](https://journals.plos.org/plosone/article/peerReview?id=10.1371/journal.pone.0326920) is public. **Concept clusters and companion work.** - [[articles/The Evolutionary Roots of Silicon Valley|The Evolutionary Roots of Silicon Valley]] — Lederberg and DENDRAL, coevolution at Jasper Ridge, the molecular clock, and the institutional history that replaces geological determinism. - [[articles/Digital Darwinism and the Invisible World of Machine Evolution|Digital Darwinism and the Invisible World of Machine Evolution]] — the five conditions, and the systems that actually meet them. - [[articles/The Synthetic Cambrian Explosion as a Technological Speciation Event|The Synthetic Cambrian Explosion]] — technological morphospace exploration as speciation analogy, labelled as such. - [[articles/2026 Annual Report on Brain-Computer Interfaces|2026 Annual Report: The Ecology of Brain-Computer Interfaces]] — the deployment ecology this essay presupposes. - [[articles/The Architecture of Continuity and Emerging Neuroinformatics Standards|2026 Annual Report: The Architecture of Continuity and Emerging Neuroinformatics Standards]] — the standards stack, archival substrate gradient, and governance frame. - [[articles/The Organic-Synthetic Brain Atlas|2026 Annual Report: The Organic–Synthetic Brain Atlas]] — the substrate-by-substrate technical atlas. - [[articles/The Next Interface Layer|The Next Interface Layer]] — the neural-access layer and the negotiation of perceptual sovereignty. - [[articles/Escape Hatch in the Skull|Escape Hatch in the Skull]] — cognitive exteriorization as the long human strategy. - [[articles/The Closed-Loop Gaussian Sensorium Engine|The Closed-Loop Gaussian Sensorium Engine]] — generative-model-mediated perceptual write-back, as architecture under construction. - [[articles/We Were Never Going to Make It|We Were Never Going to Make It]] — the hot interpreter, the cold archive, and fidelity as the last open variable. - [[articles/Uploading Is Imminent|Uploading Is Imminent]] — the edge-data argument and the sidecar case.