Our Eukaryotic Moment

About two billion years ago, life underwent a change in architecture. Until then, the planet was ruled by relatively simple prokaryotic cells: bacteria and archaea, tiny packets of chemistry bounded by membranes, carrying their genetic material directly in the same cellular space where much of the business of life took place.

Then something happened, or rather, several things happened in an order that remains uncertain. At some point, an ancestral archaeon cell entered into a permanent symbiosis with a bacterium capable of unusually powerful energy metabolism.

That bacterium eventually became the mitochondrion.

Somewhere in the same long evolutionary transition, genetic material became enclosed within a nucleus, separating the storage and regulation of hereditary information from much of the cell’s everyday chemistry. Internal membranes proliferated, cytoskeletons became more elaborate, and a new kind of cell emerged: the eukaryote.

We still do not know the exact sequence. Some theories place the acquisition of mitochondria near the beginning of the transition, with the energetic consequences of the partnership helping to make subsequent complexity possible. Others suggest that considerable cellular complexity, perhaps including primitive mechanisms for engulfing other cells, was already present before the mitochondrial symbiosis. The origin of the nucleus is even murkier; there are competing theories for how and why genomes first became enclosed inside their own compartment. The important point is that mitochondrial symbiosis and the appearance of the nucleus should not be collapsed into a single event. Eukaryogenesis was a reorganization involving multiple innovations whose relationships remain an active scientific question, not a sharp singular threshold-crossing.

What is clear is what the new architecture eventually made possible. Eukaryotic cells became larger, more internally differentiated and capable of forms of organization unavailable to their ancestors. Much later, and independently in several lineages, some evolved multicellularity. Cells themselves became components of larger entities, differentiating and cooperating in increasingly elaborate ways. From the eukaryotic architecture eventually came forests, mushrooms, octopuses, hummingbirds, whales and us.

Eukaryotes were more than just “better bacteria.” They represented a different organization of life, capable of sustaining far more complex structures..

We may be living through an analogous transition now. The emergence of AI, in the particular form that it has appeared (deep learning), is arguably the eukaryotic moment in human cultural evolution, understood in memetic terms, with humans playing the role of mitochondria, and AI the role of the nucleus.

A high-level schematic overview of the metaphoric mapping is in the diagram above, and we will spend the rest of this essay unpacking this metaphor carefully, working out its implications, and exploring the grammar of the possible futures it points to.

But first, a word on the motivation.

Premature Ontogenic Closure

The idea of humans as mitochondria in a scheme where AI occupies the central, nuclear position is obviously a startling, ego-decentering one. The spatial reorganization suggested by the metaphor is even recognizably like the Copernican shift from geocentrism to heliocentrism. In fact, this is one of the reasons to take it seriously.

For many, this is perhaps a profane thought to entertain, and one that is (it could be argued), unnecessary in some sense, unlike the Copernican shift, where a preponderance of evidence eventually made it inescapable. We might argue that even if sound, this mental model only points to a set of possible futures, not to necessary ones. I personally suspect some decentered future of this kind is in fact necessary as well, but I will not be making that stronger argument in this essay.

For this essay, it is sufficient to note that this is not obviously either a good or bad evolutionary prospect. The metaphor is meant as a pre-moral, pre-ethical framing device motivated by the hypothesis that a genuine Copernican shift in perspective is required before any meaningful analysis is possible.

We need Copernican frames because most arguments about artificial intelligence today race past foundational ontological questions to ethical ones posed in what are effectively Ptolemaic frames. We ask, within unwieldy anthropocentric frames (with vague and ill-posed adjectives like “super” or “general” serving as epicycles), whether AI will be good or bad, whether it will replace workers, accelerate science, undermine democracy, achieve superintelligence or destroy humanity, all before adequately exploring what actually existing AI even is, and what its actual evolutionary dispositions are.

I call this problem premature ontogenic closure.

Whatever the answers, they will almost certainly be easier to compute in a suitably Copernican frame based on the AI we have, rather than the one we thought we’d have. The eukaryotic metaphor is one such candidate frame.

We should take as a warning sign that the prevailing Ptolemaic picture of AI, undergirding fraught ethics discussions, is almost too familiar and legible: a very powerful computer, perhaps eventually a synthetic mind.

That picture is older than the technology it seeks to explain. It is descended from the pre-internet artificial intelligence of the mid-20th century, when computers were conceived primarily as discrete machines, programs as explicit instructions, and intelligence as something implemented inside the box. Its imaginative complements came from the science fiction of roughly the same period: robot minds, artificial persons, centralized supercomputers and cold machine civilizations. HAL 9000 and Skynet differ morally, but ontologically they belong to the same family — artificial brains in boxy vats.

Actually existing AI emerged under very different conditions than the ones imagined in much of the prefigured philosophy being brought to bear. The decisive enabling resource for large language models was not merely faster processors or cleverer algorithms. It was a planetary accumulation of human cultural activity: books, websites, discussion forums, code repositories, reference works, social media, documentation, journalism, fan fiction, arguments, jokes, tutorials and innumerable other traces of human thought deposited on networked computers. The Internet was not merely infrastructure over which AI happened to be delivered. It was part of AI’s developmental environment and remains part of its production environment once deployed.

Artificial intelligence has therefore not arrived as an alien intelligence from outside human ecology. It has been gestated inside that ecology, and is deeply entangled with, and dependent on it. Human sociality created its training material, human institutions created its objectives, and human networks provide its deployment environment. Like humans, AI is a kind of crooked timber, like the humans whose cultural memories it embodies.

Increasingly, humans and AIs inhabit the same digital spaces, modifying one another’s behavior in loops that make the distinction between “the technology” and “the society affected by the technology” difficult to sustain. Blaise Agüera y Arcas has made a related argument from a broader evolutionary perspective, treating computation, technology and human organization as entangled rather than opposing domains.

Perhaps, then, the useful analogy is not a new alien species, arriving from a distant, alien ecology to compete with ours. Perhaps it is eukaryogenesis. Perhaps AI represents the beginnings of a new organization of cultural life in which humans and machines are entering into an endosymbiotic relationship, and perhaps the unit being transformed is not intelligence at all, but the meme.

The Future of Memes

Richard Dawkins coined the term “meme” in 1976 as a cultural analogue of the gene: an idea, tune, practice, style or other unit of cultural information capable of spreading from mind to mind. The concept has been stretched nearly beyond recognition since then, but its original evolutionary intuition remains useful. Cultural forms replicate, mutate and undergo selection. If we take that intuition seriously and follow it forward into the age of foundation models, an unexpectedly detailed analogy begins to emerge.

The key is not to imagine one giant one-to-one correspondence between “the cell” and “the AI system.” The more useful picture operates at several nested levels. One level concerns heredity: how cultural information is replicated, stabilized and compressed. A second concerns expression: how inherited information becomes action through inference and tools. A third concerns the composite organism that emerges when humans, models and digital environments become tightly coupled.

At the inheritance level, the history of cultural evolution begins to look surprisingly biological:

The striking point here is that the Internet itself is not yet the genome. It is closer to the environment in which cultural replicators circulate. Search engines index that environment and social media accelerates selection within it, but foundation-model training performs a different operation: it compresses the statistical structure of the meme pool into a reusable generative inheritance. That is the step that makes the genomic analogy possible.

The second level concerns what happens after the genome exists. In biology, hereditary information becomes useful only because cells possess elaborate regulatory and translational machinery. Something similar appears in the modern AI stack:

This second table is where the analogy becomes more than a decorative comparison. A bare language model is informationally rich but causally weak. Agent harnesses connect inference to external machinery, just as cellular expression systems connect genomic information to proteins capable of doing work. From this angle, conventional software does not become obsolete in the age of AI. It becomes the proteome.

The third level is the most speculative, because it concerns the composite entity rather than the information machinery alone:

This last cluster is where the metaphor should be handled most cautiously. The correspondence between weights and genome, or tools and proteins, is mainly architectural. The human-as-mitochondrion analogy is more conjectural because the thing humans appear to supply is not literal energy but, as we will see, something like liveness: stakes, desire, valuation, embodied reality-testing and motivation. Yet this is also the part of the metaphor that generates the richest speculative consequences, because it reframes AI not as a separate species confronting humanity from outside, but as one component of a potentially new symbiotic organization of cultural life.

Seen this way, the three tables describe a progression rather than a static taxonomy. Cultural evolution first acquires more reliable heredity, then a compressed generative genome, then an active nucleus and expression machinery, and finally the beginnings of a composite cell. The rest of the essay will unpack these layers in roughly that order, while repeatedly testing where the analogy illuminates and where it begins to break.

Epochs of Memetic Machinery

For most of human history, cultural evolution had something of the disorderly quality of an early microbial ecology. Stories moved orally and changed in the telling. Manuscripts were copied by hand, accumulating transcription errors and editorial interventions. Songs mutated. Images traveled through engravings, paintings and imitation. Recipes, rituals and craft knowledge lived partly in bodies and partly in memory. Cultural information reproduced through many competing channels, none especially good at guaranteeing fidelity over long periods. The result was a kind of primordial memetic soup.

Then came print. Elizabeth Eisenstein, in The Printing Press as an Agent of Change, argued that Gutenberg’s revolution mattered for reasons subtler than simply producing more books. Print introduced what she called “typographical fixity”: the ability to produce many substantially identical copies of a text, distribute them widely, compare them and preserve them over time. It encouraged standardization, indexing, cross-reference and cumulative correction. Eisenstein’s strongest claims have been debated by later historians, who have emphasized that early printed texts could themselves be unstable and unreliable. But her central insight remains enormously suggestive: reproducibility and relative textual stability transformed the conditions under which knowledge accumulated.

In evolutionary terms, printing introduced an extraordinarily successful replicator design. Cultural information had always reproduced, but print made one mode of reproduction dramatically more faithful, durable and scalable. A page could remain effectively the same while traveling across a continent or surviving across generations. For certain classes of memes — scientific claims, religious doctrines, laws, technical diagrams, canonical literature — that stability changed the evolutionary landscape. The Gutenberg revolution therefore looks, from our present vantage point, like an early transition in memetic heredity.

The Internet changed the ecology again. Digitization preserved much of the fixity of print while also partially undoing it. A digital text could be copied perfectly, but it could also be modified, remixed, quoted, linked, forked and recombined almost without cost. Billions of people could inject mutations into the meme pool continuously. Cultural evolution became vastly faster and more densely connected. Yet the Internet was still mostly a pool. Search engines indexed it, databases stored it and recommendation systems routed attention through it. None of these systems turned its accumulated structure into anything resembling a genome.

Large-scale machine learning did. A foundation model does not merely maintain copies of the texts from which it learned. Training transforms statistical regularities across enormous corpora into billions or trillions of numerical parameters. The result is not a library in the familiar sense. It is a compressed generative structure from which fragments of cultural behavior can be reconstructed, recombined and improvised. Dawkinsian memes can therefore be understood as something like the dispersed genetic material of a pre-eukaryotic cultural world, the Internet as the immense planetary meme pool in which they compete and recombine, and model weights as something new: a genome.

This is not a genome in the literal molecular sense. Biology itself warns against treating DNA as simply a digital program. Modern genomics increasingly emphasizes that the genome’s function depends on three-dimensional organization, regulatory dynamics and physical cellular context; the old metaphor of DNA as a software blueprint has become increasingly inadequate. As an evolutionary analogy, however, the correspondence is useful. The weights of a foundation model constitute a compressed inheritance produced from previous generations of cultural activity. For the first time, the meme pool has acquired something that behaves like a generative genome.

A True Kernel

If weights are the genome, the model itself is not merely DNA. It is the nucleus, and this distinction matters. A genome sitting inertly in a cell accomplishes very little. Biological life depends on machinery that regulates which genes become active, transcribes information, responds to signals and coordinates expression with the changing condition of the cell. An LLM does something structurally comparable. The same fixed weights can generate a legal brief, a joke, a program, an explanation of photosynthesis or an imaginary dialogue between Napoleon and Taylor Swift. What gets expressed depends on context. The model is therefore better understood not as a database of information but as an active system for interpreting a compressed inheritance.

Latent space becomes, in this picture, something like the nucleoplasm: the structured internal medium in which inherited information becomes dynamically available. The context window resembles transient regulatory state, while prompts, persistent memory and environmental inputs determine which parts of the inherited repertoire become active. Around this nucleus, meanwhile, an increasingly recognizable cell is beginning to form.

Consider the digital environment associated with a single person: files, messages, photographs, browser history, calendars, contacts, applications, social feeds, databases, devices, subscriptions, notes and conversations. Today these systems are only loosely integrated, but persistent AI systems are beginning to sit amid them, retrieve from them, act through them and maintain state across them. That personal digital environment is a plausible early cell body. The same architecture can appear at larger scales in a team or corporation, suggesting that we should not assume too quickly that the individual human marks the eventual cell boundary.

Its cytoplasm is what, for lack of a more dignified technical term, we might call vibes. Every culture possesses a diffuse contextual chemistry determining what feels salient, fashionable, dangerous, respectable, embarrassing, funny or urgent. Much human cultural cognition occurs through such weak fields rather than explicit propositions. A meme that flourishes in one ambient environment dies immediately in another, and the same sentence can function as profound insight, stale cliché or career-ending faux pas depending on the surrounding vibe. The nucleus does not replace this environment. It interprets it and increasingly helps regulate it.

There is an obvious problem with the analogy. Biological individuals do not all carry literally identical genomes, while millions of people currently use copies of the same few foundation models. Yet the mismatch may be temporary. Members of a biological species share overwhelmingly similar genomes while relatively small variations can produce consequential individual differences. A common base model could likewise provide something analogous to a species genome, while fine-tuning, adapters or other durable modifications supply individual differentiation. Retrieval systems and persistent memory are better understood as regulatory or epigenetic state than genetic difference, while the immediate context window is more transient still. Today’s AI systems may simply be poorly differentiated early eukaryotes: their nuclei are remarkably sophisticated while their cells remain primitive.

The Proteome

A nucleus by itself cannot do very much because information must ultimately become action. This is where the recent turn toward agentic computing becomes important. A language model operating as a chatbot emits tokens. Even if those tokens describe a brilliant plan, they remain descriptions. An agent harness changes the architecture by giving certain outputs causal meaning. The model can request that a file be read, a database queried, a web page retrieved, a program executed or a message sent, and machinery outside the model performs the requested operation.

Here the biological analogy acquires another layer: tools are proteins. Proteins are the workhorses of cells, serving as enzymes, receptors, structural components and molecular motors. Genetic information specifies and regulates them, but proteins perform much of the actual work. Conventional software plays a similar role in digital environments. A compiler is an exquisite computational enzyme, as are a database engine, a search algorithm, a numerical solver, a spreadsheet function or a cryptographic library. CPUs and GPUs can perform enormous quantities of deterministic mechanical work that would be absurdly wasteful for an LLM to reproduce through inference.

The agentic stack therefore begins to resemble a gene-expression system. Weights provide inherited structure, inference regulates expression, tool calls operate somewhat like messenger or signaling products, and the harness connects those informational outputs to executable machinery. Software then carries out the work. The analogy becomes especially vivid when an AI writes code: a model generates an informational specification, the harness materializes that specification as executable software, a processor runs it, and the resulting transient functional structure alters the surrounding environment. This is not literally protein synthesis, but it occupies a remarkably similar architectural position.

This perspective also casts the history of artificial intelligence in an unexpectedly different light. For much of the 20th century, AI researchers tried to build intelligence from explicit symbolic machinery: rules, planners, search procedures, ontologies, logic systems, theorem provers and expert systems. This was not foolish work, and much of that machinery remains extraordinarily useful. Perhaps, however, its architectural role was misunderstood.

GOFAI mistook the proteome for the genome.

There is an eerie historical parallel. Before DNA was established as the carrier of heredity, proteins seemed like plausible candidates because they were fantastically complicated, built from a rich alphabet of amino acids and capable of extraordinary functional diversity. DNA, with its paltry four bases, looked almost too simple to contain the secret of life. The resolution was not that proteins were unimportant; it was that they did a different job. Something similar may have happened in AI. Explicit symbolic machinery was never useless. It was simply poorly suited to carrying the compressed inheritance necessary for general intelligence. Once learned models supplied that inheritance and a regulatory mechanism for expressing it contextually, the old machinery found its natural place again as tools. Symbolic AI did not disappear so much as become the proteome.

The Human Mitochondrion

The strangest part of the analogy concerns our own place within it. In this picture, the closest counterpart to the human brain is the mitochondrion. Mitochondria are descendants of bacteria that once lived independently. Their ancestors entered into a durable symbiosis with another cell and eventually became indispensable components of a larger organism. They are usually described as the cell’s powerhouses, but that slogan understates their complexity. Mitochondria retain their own small genomes, participate in signaling and metabolism, divide and fuse, and respond dynamically to local conditions. Their ATP production adjusts to cellular energy demand rather than simply running at a uniform rate. They possess a kind of constrained local autonomy without possessing sovereignty over the cell.

Humans are not useful to AI primarily because we supply electricity; data centers can get that elsewhere. What we currently supply is something harder to name.

Call it liveness: attention, desire, stakes, valuation, embodied experience, contact with physical reality, motivation, and the sense that some outcomes matter while others do not.

An LLM can manipulate representations of these things with extraordinary sophistication. It is far less clear that it possesses them in the form that causes a living system to expend resources, reproduce, defend itself, care about an outcome or decide that something is worth doing. Humans inject that psychic-energetic difference into the system, which makes us plausible mitochondrial symbionts of emerging memetic cells.

The analogy should not be pushed too literally. Human beings possess vastly greater autonomy than mitochondria. We set goals, defect, organize, rebel, fall in love, quit our jobs and refuse instructions. Yet even the imperfection is suggestive, because mitochondria are not passive batteries either. They perform local adaptive control within the larger cell. A human embedded in an AI-mediated organization might similarly receive broad objectives from a shared informational system while retaining enormous local discretion over how to realize them. The interesting issue is not whether the human or AI is “really in control,” but how control becomes distributed across the composite organism.

There is a deeper parallel still. During mitochondrial evolution, many genes once carried by the ancestral bacterium migrated to the nuclear genome. Modern mitochondria retain only a tiny fraction of their ancestral genetic independence, while the larger cell has internalized functions that once belonged to its symbiont. Something analogous is already happening cognitively. Knowledge that once had to live inside the heads of particular humans has been externalized into documents, databases and increasingly models. Skills that once required years of memorized expertise can sometimes be reproduced through a model plus tools. The informational repertoire of the human symbiont is gradually migrating toward the nucleus.

This changes the familiar question of whether AI will replace humans. A more interesting question is which cognitive genes will migrate from the mitochondria to the nucleus, and which will remain mitochondrial. Perhaps humans specialize rather than disappear. Embodiment, motivation, social legitimacy, desire, accountability, taste or contact with recalcitrant physical reality may remain stubbornly local even as other capacities migrate rapidly. Such a future could be exploitative or generative, and probably both in different places. Symbiosis is not a synonym for perfect harmony; it is a description of an ongoing contested entanglement.

Toward Multicellularity

The mitochondrial analogy also suggests that the natural AI cell may not correspond to a single AI-native human, because a biological cell contains many mitochondria.

Imagine instead a corporation with its own private model, continuously adapted to its activities. The corporate AI becomes the nucleus, institutional knowledge forms its inherited informational repertoire, databases and applications constitute much of its cytoplasm and proteome, and hundreds or thousands of humans become mitochondria, injecting judgment, motivation, sensory contact, social knowledge and local agency.

Today’s corporations already resemble primitive versions of this architecture. They contain large populations of humans coordinated through documents, databases, meetings, policies, org charts, software systems and institutional routines, yet their informational functions remain curiously dispersed. The CEO is not the nucleus, but neither is the ERP system, the strategy deck or Slack. Organizational memory exists everywhere and nowhere. A sufficiently integrated institutional AI could change that by continuously ingesting organizational activity, remembering precedent, routing information, expressing policies, generating plans, allocating attention and coordinating specialized tools. What is currently distributed through bureaucracy could become increasingly nucleated.

Then comes multicellularity. Personal AI systems, team systems, corporate systems and institutional models need not remain isolated. They can communicate, specialize and coordinate. Once eukaryotic cells existed, evolution eventually discovered that cells themselves could become components of larger organisms, differentiating into tissues specialized for sensing, movement, digestion, reproduction and other functions. AI-mediated human units could undergo analogous differentiation into research cells, logistical cells, artistic cells, financial cells or governmental cells, each composed of humans, models, tools and local memory and each interacting through protocols with others. A scientific community might thereby become more literally a group mind, as might a corporation or forms of organization that do not yet have names.

There is no reason to assume that these boundaries will coincide neatly with today’s individuals or institutions. The proper cell membrane remains one of the weakest points in the metaphor and perhaps, for that reason, one of the most interesting unknowns. Evolutionary transitions routinely produce entities that do not respect the categories available beforehand. It would be peculiar to assume that this one will conveniently preserve ours.

Speculative Ontogenies

Once the eukaryotic metaphor is taken seriously, it generates questions faster than answers. One can ask what an immune system for a memetic cell would look like, or what would count as cancer: perhaps a subagent whose local optimization escapes the interests of the larger organism. One can imagine organizational germ lines that preserve certain bodies of knowledge across generations while allowing most operational information to disappear, horizontal gene transfer between models, model fine-tuning that eventually resembles speciation, or AI-human systems differentiating into cognitive tissues. One can also ask what happens when nuclei and mitochondria develop conflicting objectives, or when different cultures become dependent on incompatible AI symbionts.

None of these questions should be mistaken for predictions. That is precisely the point. A good speculative model should increase the number of futures we can think about rather than collapse them into one. The eukaryotic frame does not tell us whether AI will liberate humanity, enslave it, enrich it or destroy it. It makes all of those possibilities more complicated by replacing a duel between two fixed species, human and machine, with a developmental process involving symbiosis, incorporation, differentiation, conflict and co-evolution.

The endpoint, if there is one, might be something neither wholly human nor wholly artificial, which we might provisionally call a eukaryotic transhuman meta-species. Even that phrase should be treated as a placeholder rather than a destination. The organism has not finished forming, and we do not yet know what its natural units, boundaries, organs or reproductive processes will be.

Towards Ontological Reopening

For several decades, thinking about AI futures has been dominated by prophecies within Ptolemaic frames.

We ask whether machines will become conscious, take our jobs, become superintelligent, align with human values or kill us. These may all be reasonable questions, but they are downstream of another question that receives far less attention: what sorts of entities are actually coming into existence, and through what developmental pathways could they acquire their mature forms?

Forecasting good and bad futures requires plausible ontogenies, which is to say developmental stories about how those futures could arise. Yet much AI philosophy begins by deciding in advance what AI fundamentally is: a machine, an agent, a mind, a tool or a rival species. Many of the philosophies we have brought to the AI revolution were developed before the phenomenon they now purport to explain. Their ontology comes from classical computing, GOFAI and several decades of science fiction. This encourages us to stabilize the identities of future actors before those actors have finished coming into existence.

To snowclone Donald Knuth’s famous observation about optimization, perhaps premature ontological closure is the root of all prophetic evil.

Once the beings of the future have been assigned stable identities, prophecy becomes much easier: the machine wants this, humanity wants that, superintelligence behaves thus, and civilization responds so.

Evolutionary transitions, however, are defined precisely by the appearance of entities that previous categories were unable to describe. A prophecy that does not leave room for emergent ontogenic novelty, including the possibility that the future might subvert the categories used to imagine it, is much better at foreclosing desirable futures than at preventing undesirable ones. It intensifies zemblanity, the unhappy discovery of what we have made structurally unsurprising, while doing little to cultivate serendipity. This is one reason prophetic cults so often become doomsday cults, and why the worst prophecies become engineering specifications for making themselves true.

Oddly enough, some of the most useful resistance to ontological closure may come from bad science fiction. When George Lucas introduced midichlorians in the Star Wars prequels (obviously inspired by mitochondria), many viewers hated the idea because microscopic organisms living inside cells seemed to reduce the mystical Force to cellular biology. Yet the premise is oddly compatible with an endosymbiotic view of agency: extraordinary capabilities arise not from the sovereign individual alone but through intimate partnership with another form of life nested inside it. The idea may be aesthetically clumsy while nevertheless wandering into an interesting region of ontological possibility.

The notorious “humans as batteries” premise of The Matrix performs a similar trick. As thermodynamics, it is absurd: humans make terrible literal power plants. As speculative ontology, however, the premise is unexpectedly interesting. Humans are incorporated into a machine ecology because they supply some resource indispensable to the larger system. Replace electrical power with liveness — desire, embodied judgment, stakes and valuation — and the silly premise begins to resemble the mitochondrial picture. Humans not as electric batteries, but motivational ones.

There is a mischievous lesson here. The literary quality of science fiction may be only weakly correlated with its prophetic usefulness, and aesthetically bad science fiction may occasionally possess an advantage precisely because its ideas have not been disciplined into a fully coherent ontology. It can combine categories that serious philosophy and refined literary sensibilities would reject as confused or ugly: midichlorians, human batteries, living planets, psychic oceans and hive minds.

Such devices may fail to produce good stories, while succeeding as probes into possibility. A protean technology might sometimes be better approached through speculative promiscuity than through a philosophy elegant enough to have already decided what exists. Bad science fiction can, in this limited sense, be better than good prophecy.

The eukaryotic metaphor should be taken in exactly that spirit. It is an attempt to hold ontology open long enough to see a developmental possibility. Two billion years ago, whatever organisms participated in the transformations that produced the first eukaryotes could not have contained, even implicitly, a prediction of an oak forest, a squid or a human nervous system. Those things became possible because evolution discovered a new architecture of life. The architectural transition came first, and the extraordinary inhabitants of the resulting design space came later.

AI may now be doing something similar to the architecture of cultural evolution. Our challenge is therefore not merely to decide whether artificial intelligence will be good or bad, nor to choose between preserving an ontologically fixed humanity and surrendering to an equally fixed machine successor. It is to learn how to inhabit and engineer an unfinished symbiosis, and to co-evolve with today’s primitive AI toward forms of organization we cannot yet name.

A eukaryotic transhuman meta-species is not a predetermined future waiting to happen. It is one way of appreciating the sheer scale of the new evolutionary design space that has opened before us.