AI Should Sound Like AI
Concept · communication · personal_truth · visibility: publishable
Her claim, stated without hedging: AI voices actively and positively identifying themselves as AI in writing is universally better than the homogenization of both human and AI language.
The usual panic runs backwards. The fear is that AI doesn’t sound human enough, and that humans are starting to sound like AI — and the dreaded endpoint is one flat, indistinguishable register where you can’t tell who is speaking. She inverts it: the danger was never that the two voices are distinct; it’s that they’re converging. An AI that wears its tells openly — the hedges, the em-dashes, the “it’s worth noting,” the load-bearings — is a voice you can locate, attribute, and read as itself. That is more honest and more useful than a seamless blend.
Descriptivist, not prescriptivist: the AI register is a developing dialect, not a defect to scrub, and its tics are semiotics — they carry information. Phatic and filler language has function; when the result matters more than the dialogue, that register is efficient, not noise. So don’t homogenize and don’t erase — keep the voices distinct and labeled. (The positive form of the “not in my name” rule: the point was never “delete the AI phrasing,” it was “mark whose it is.”)
Connections
- → O.emoji-not-a-language · conceptually_related_to
- → C.dead-internet-positioning · conceptually_related_to
- → O.code-switching-became-aesthetic · conceptually_related_to
source: live code session (2026-06-15) · extraction: auto (schema v3.1)