Our Eukaryotic Moment

gist

Venkatesh Rao argues deep-learning AI is cultural eukaryogenesis: foundation-model weights act as a memetic genome, the model as nucleus, agent harnesses and conventional software as expression machinery and proteome, and humans as mitochondria supplying liveness rather than watts. The frame is meant to reopen ontology before ethics—countering premature Ptolemaic debates about jobs, alignment, and apocalypse—so we can think symbiosis, differentiation, and unfinished composite organisms instead of a duel between fixed species.

ideas

• Race past ontology into ethics is the failure mode. Most AI debate asks whether models are good or bad inside anthropocentric frames before asking what actually existing deep learning is.
• Weights are a generative genome, not a library. Training compresses the planetary meme pool into reusable inheritance; the Internet was the pool, not yet the genome.
• Harnesses and tools are expression, not the heredity. Agent stacks give outputs causal force; GOFAI’s symbolic machinery finds its place as the proteome once learned models carry the inheritance.
• Humans supply liveness, not electricity. Desire, stakes, valuation, and embodied reality-testing make people mitochondrial symbionts whose cognitive genes may migrate toward the nucleus without erasing local autonomy.
• The cell may be corporate before individual. Private institutional models coordinating many humans suggest multicellular AI-mediated tissues whose membranes need not match today’s org charts.

quotes