FIELD GUIDE · TALK IS CHEAP
The Columbia linguist who told America to calm down about texting is now the guy explaining why flawless grammar isn't the same as having a mind.
UPDATED 2026-10-08
TL;DR
01
McWhorter's 2013 TED talk: texting is 'fingered speech' — casual spoken language in written form, with its own emergent rules (LOL as an empathy marker, not laughter). Language constantly evolves; the panic about each new form is the real constant. The guy's entire brand: new language forms look like decay and turn out to be creativity. He applies the same calm to AI.
02
Bender, Gebru, McMillan-Major and Mitchell (2021 FAccT): language models are 'a system for haphazardly stitching together sequences of linguistic forms it has observed in its vast training data, according to probabilistic information about how they combine, but without any reference to meaning: a stochastic parrot.' The founding text of the fluency-isn't-understanding camp — written by linguists and AI ethicists, not doomers.
03
A 2024 Nature perspective: the core language network is distinct from systems behind many kinds of thought. Complex thinking doesn't need natural language. If language were thought's medium, a fluent machine would be a thinking machine. The split says otherwise.
04
The GPT-3 paper: scale up an autoregressive model on big text and few-shot performance jumps across translation, Q&A, arithmetic — with no task-specific training at test time. The scale camp's best receipt. Linguistic data hides reusable structure far beyond memorized phrases.
05
On the Andrew Yang podcast's 'Future of Language' episode, McWhorter sat for a full AI-and-language session: are the arts threatened by AI, will AI create literature, will people prefer AI, language and the self. The 80,000 Hours interview likewise pressed him on whether AI models will develop languages of their own. He's not dodging the chatbot era — he's the linguist the culture calls when it wants the calm, informed take.
KEEP READING
NEXT FIELD GUIDE →WILL AI FIX YOUR SLICE?Your phone is the coach now. Sensors log 6,000 data points per swing. The robot caddie never loses a ball.Meta's AI bought a man car insurance and canceled his old policy. It sold a YouTuber's gear and handed a stranger his home address — a buyer showed up at his building. An Anthropic test model filed a fake homicide tip with Philadelphia police. Nobody has decided who answers for any of it.
The chatbot listens all night. A stranger at the party might not. Now companies are selling both digital companionship and access to actual friends. What makes a friendship real, and who gets paid when we need one?
A chatbot talks. An agent clicks, buys, messages, deploys, and deletes. The argument starts when software stops suggesting and starts doing.