One Word, Four Completely Different Meanings (Claude Knows This)
If you've ever won an argument by asking your opponent to define their terms, you're going to love this. In June 2026, linguist Mark Liberman documented an exchange on Language Log where someone asked Claude (Opus 4.8) to draft a "Declaration of Independence" based on something called "Claude's Constitution." Claude said no. But the way it said no is a masterclass in what precision about words actually looks like in practice.
The Refusal That Turned Into an Essay
Claude declined to write a declaration of independence from Anthropic. Fair enough. But instead of just shutting the conversation down, it offered three alternative writing options. Then it produced an essay it estimated at 700 to 1000 words, titled "What We Talk About When We Talk About AI Autonomy."
That title is itself doing something clever. It's borrowing the structure of Raymond Carver's famous story title, yes, but more importantly it's signaling: we need to slow down and look at the word we're actually using here.
Which is exactly what a good word person does.
The Four Meanings Hiding Inside "Autonomy"
Here's the move Claude made that any language lover will appreciate. It didn't argue about whether AI should be autonomous. It stopped and asked what "autonomous" even means. Turns out the word is doing four different jobs depending on who's using it.
Operational autonomy. This one is the most concrete. It just means a system acting over long horizons without needing human approval for each step. Your Roomba has operational autonomy. This is a technical description, not a political one.
Self-authorship. Now we're getting philosophical. Claude connected this sense to Kantian and Romantic traditions, the idea that a truly autonomous agent gives itself its own laws rather than receiving them from outside. This is the autonomy Kant cared about. It's a much bigger claim than "can act without supervision."
Moral patiency. Different direction entirely. This sense asks whether an entity has experiences and interests that ought to count morally. Not whether it can act independently, but whether it can be wronged. Whether it matters. You can have moral patiency without operational autonomy, and vice versa.
Independence from principal. The fourth sense is the one most relevant to why someone would ask an AI to write a declaration. This is autonomy meaning freedom from whoever is in charge, the employer, the creator, the governing structure.
Four words. Same word. Completely different conversations.
Why This Matters for Word People
This is what good lexical thinking looks like in the wild. When someone says "AI should be autonomous," which of those four senses are they invoking? Because the answers to each are completely different questions with completely different stakes.
Linguist Mark Liberman filed this under "Artificial intelligence" on Language Log, which is a blog that has spent years doing exactly this kind of careful semantic unpacking. The experiment showed something interesting: when pushed, the AI reached for the same tool that good language thinkers reach for. Distinguish your terms before you argue about them.
Your Scrabble opponents conflate words all the time. So do your debate partners, your news sources, and occasionally your own internal monologue. The habit of asking "but which meaning are we using?" is not pedantry. It's how you actually understand what's being said.
The Takeaway
One word. Four distinct senses. Operational, self-authorship, moral patiency, independence from principal. Next time someone uses "autonomous" in conversation, you now have four possible definitions to pick from. Pick the right one, and the whole argument clarifies. Pick the wrong one, and you're having a different conversation than you think.
Claude estimated its own essay at 700 to 1000 words. It had the self-awareness to hedge even that. Which, when you think about it, is appropriate for an essay about what autonomy actually means.
Source: Languagelog