Language Moved Out of Your Head. It Lives in AI Now.
Dan PetersonSynopsis — AI-drafted from Dan's notes
A 32-minute video essay by the persona channel Count Grey Matter, built on an old metaphor: language as a living thing that evolved inside human heads. The Count traces it from August Schleicher’s 1863 claim that languages are natural organisms, through Darwin’s milder comparison in The Descent of Man, William Burroughs’ word-as-virus, Dawkins’ meme and Nicholas Humphrey’s note that a meme parasitises the brain, to Terrence Deacon’s argument in The Symbolic Species (1997) that languages adapt to children’s guessing far faster than brains adapt to languages. The Count sides with Darwin: the organism is a lens, not a fact, and he uses it as one.
Calling language the tenant and its media houses, he walks a street. Heads first, where language can make new sentences but dies with its speakers (humpback song, spreading east across the South Pacific population by population, is his example of a tenant that never got a second house). Then writing, which stores words like a coffin until someone reads them, with Plato’s Phaedrus on the written word’s one fixed answer; then the addressed page of chapters, concordances and numbered verses; then print, with Luther’s New Testament shaping written German; then the internet, the largest library of still words. Along the way Samuel Butler’s 1863 letter on machines as our successors, McLuhan’s humans as the machine world’s bees, a Weizmann Institute study in which crazy ants solved a load-moving puzzle better in larger groups while silenced humans got worse, and Susan Blackmore’s 2010 prediction of a third replicator native to machines.
The evidence section rests on Elan Barenholtz, a psychologist at Florida Atlantic University, who argues that language models show language is autogenerative: everything needed to produce the next word is already in the relations among words, with no reference required, and that brains may run language the same way. The Count’s reading parts from Blackmore: what moved into the machine is not a newborn replicator but the old tenant, carrying everything it brought. He admits that is his reading of a young theory.
The close is about food. The 2024 Nature study on model collapse found that models trained on their own output lose the rare things first, so the new house can run language but not feed it; human writing has to stay in the mix. Hence a better ending than parasite: a symbiont (an idea he credits to Burroughs and to Morten Christiansen), which still eats at our table.