Large AI Models Are Cultural and Social Technologies
A large model is not a mind in the making. It belongs with writing, print, markets and bureaucracies - technologies that let people use what other people know - and it should be studied and governed as one.
Read the original ↗Four pages in Science, with no data and no experiments, arguing that the whole AI debate is using the wrong noun. The cheap version of this argument is that a chatbot is only autocomplete. This is not that argument. It is a stranger one: the thing you are talking to belongs in the same family as a price, a census and a printing press, and every one of those reshaped a society without ever being a mind.
Included with a plan
This discussion is included with a plan. The first three lessons of every track are free, with their boards and videos. A plan unlocks every lesson, board, video, lab and paper for one price.
Related on chalk
The same idea as a board, lab, video or lesson.
- BoardFrom Transformer to LLMa model of the text people wrote - the paper's whole argument is what follows from that sentence
- BoardRAGretrieval makes the library framing literal: the model is the index, the humans wrote the books
- BoardWhat the Model Seeswhat goes into the context window is other people's writing, which is the paper's point about where the knowledge lives
- BoardAgents and Tool Usethe agent framing the paper argues against, drawn as the loop its proponents mean by it
- BoardAuctions and Mechanism Designa market as a machine for pooling what many people know - the paper's closest analogy for what a model does
- BoardCan Machines Thinkthe question the paper says is the wrong one to be asking
- BoardWhat Is Knowledgeknowing something and having access to what others know are different, and the paper lives in that gap
- LabThe second-price auctiona mechanism that gets people to reveal what they know - the market half of the paper's analogy, run
- LabTokens and embeddingsthe compressed map of human writing that the paper says is the whole object
- VideoVector databases and RAG, end to endthe model as an index over documents humans wrote
- VideoWhat AGI means, and who gets to saywhat AGI would mean and who gets to say - the framing this paper rejects
- VideoContext engineering: the window is a budgetcontext engineering, which is choosing which other humans the model hears from
- LessonFrom Transformer to LLMUses the board From Transformer to LLM
- LessonThe Bigram ModelUses the board From Transformer to LLM
- LessonThe GPT BlockUses the board From Transformer to LLM
- LessonSampling and What Scale BuysUses the board From Transformer to LLM
- LessonRAG (Retrieval-Augmented Generation)Uses the board RAG
- LessonRAG Over Your Own NotesUses the board RAG
- LessonWhat the Model Sees (one string, every turn, from the top)Uses the board What the Model Sees
- LessonAgents & Tool UseUses the board Agents and Tool Use
- LessonUnder a Buttonwood Tree - the exchange, the order book and the spreadUses the board Auctions and Mechanism Design
- LessonAuctions & Mechanism DesignUses the board Auctions and Mechanism Design
- LessonWhat They Know That You Don'tUses the board Auctions and Mechanism Design
- LessonThree Friends, One RestaurantUses the board Auctions and Mechanism Design
- LessonCan Machines Think?Uses the board Can Machines Think
- LessonWhat a Word Points AtUses the board Can Machines Think
- LessonWhat Is Knowledge?Uses the board What Is Knowledge
Further reading
Read the original for the full argument and its evidence.
- The paper ↗Science 387(6739), 13 March 2025, four pages, no data. Read the section on social and cultural institutions first: the Hayek and Scott paragraphs are where the argument is made, and the rest applies them.
- AI is a cultural, not social, technology ↗William Cullerne Bown, April 2025. The strongest published critique, and it attacks the half of the analogy the paper leans on hardest: a market yields a shared result and a model yields a private capability.
- AI as Social Technology ↗Farrell and Shalizi, Knight First Amendment Institute, 2026. Two of the authors a year on, with coarse-graining in place of the looser vocabulary, and the distributional question moved to the front where it belongs.
- Imitation versus Innovation: What children can do that large language and language-and-vision models cannot (yet) ↗Yiu, Kosoy and Gopnik, Perspectives on Psychological Science 19, 874 (2024). The empirical work under the paper's central concession, from the author who did it. This is what very far from these human capacities is standing on.