What if AI and humans form a single organism?

Venkatesh Rao proposes that the emergence of deep-learning AI is "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." He sets this against what he calls "premature ontogenic closure" — the habit of debating whether AI is "good or bad," will "replace workers" or "achieve superintelligence," inside an anthropocentric, Ptolemaic frame, before asking what AI structurally is. AI, he argues, did not arrive as "an alien intelligence from outside human ecology" but was "gestated inside that ecology": the Internet's accumulated books, forums, code, and conversation functioned as AI's "developmental environment," making human culture the raw material a model compresses.

Rao extends the cell analogy level by level. The Internet itself "is not yet the genome" — it is "closer to the environment in which cultural replicators circulate"; training is the operation that "compresses the statistical structure of the meme pool into a reusable generative inheritance," making the weights the genome. But the model is "not merely DNA," it is "the nucleus" — an active interpreter that expresses the same fixed weights differently depending on context (a legal brief, a joke, a program). Expression requires action, so agent harnesses supply the missing causal layer: "tools are proteins," turning token output into file reads, queries, or executed code, with conventional software standing in as "the proteome." Humans, in this picture, are "mitochondria" — organelles with "a kind of constrained local autonomy without possessing sovereignty over the cell" — who alone supply "liveness: attention, desire, stakes, valuation, embodied experience."

This lands close to the whole-system argument in Does software intelligence exist independent of hardware and environment?: where that note formalizes intelligence as f3(f2(f1)), irreducible to software alone, Rao's "liveness" names the same irreducible remainder in biological rather than formal terms — the model-as-nucleus is causally and existentially inert without the mitochondrial host. It also sharpens the asymmetry named in Are we underestimating human minds while debating machine minds?: "premature ontogenic closure" is a structural version of the same complaint, that ethics-first debate skips the ontological question of what the system actually is before asking whether it is good or bad. And it converges with Do LLM improvements reflect reasoning gains or corpus shifts? in locating the generative source on the human side of the relationship — there in the written discourse models draw on, here in the "liveness" only humans supply to an otherwise inert genome.

The excerpt is explicit about its own limits: Rao calls the thought "perhaps profane" and concedes the model "only points to a set of possible futures, not to necessary ones," reserving his stronger claim — that some decentered future is actually necessary — for elsewhere. Nothing here is tested against a case or dataset; the genome/nucleus/proteome/mitochondria mapping is a reframing device, not a predictive or measurable claim, and it does not resolve whether the human side of the "endosymbiotic relationship" keeps any sovereignty or is gradually absorbed. The defensible implication is narrower than the analogy's reach: debates about AI should name their ontological assumptions before their ethical ones — not that biology accurately describes what is actually happening.