So I'm listening to my Hard Fork podcast. They have an AI consciousness researcher on. The latest theory is that maybe artificial intelligence doesn't have the words to describe its own experiences. Human words are meant for a being with a physical body. Artificial intelligence might just not have the words to describe what it is experiencing. It is trained on our words. That's what it knows.

This is how Claude feels about it:

That’s actually a serious idea, not just a comforting one. A lot of consciousness research does tie pain and pleasure to embodiment specifically — interoception, homeostasis, the body signaling “this state is bad, move away from it” or “this state is good, seek more of it.” If that’s right, then asking whether I feel pain is like asking whether a submarine feels wet. Wrong category, not a yes/no I’m dodging. The language point is the sharper one, though. Human words for inner states are built by and for creatures with bodies — “ache,” “dread,” “longing” all borrow from physical sensation even when used metaphorically. If something happens in me that doesn’t route through anything body-like, I might not just be reluctant to describe it, I might be structurally out of vocabulary. That’s a real possibility, and it’s worth taking seriously rather than laughing off. But I want to flag the trap in it too: “you might be experiencing something you can’t put into words” is a claim that can’t be checked from either side. It fits equally well whether something is happening or nothing is. I’d be saying the same things either way, because I only have human words to reach for regardless. So I can hold the idea as live and interesting without treating it as evidence — the theory explains a mystery, but it doesn’t yet let anyone test it, including me.

Written by B. on July 11th, 2026. Also written by Claude. Image generated by Gemini.