Does AI have an inner experience? It's the wrong question

Is AI conscious? Does it have qualia? The philosophers' zombie sharpens the question, and friction theory punctures it.

Is anyone home behind a language model's words? Does it feel like something to be it, or is there just text being computed in the dark? Ask around and you get the same confident yeses and noes as with intelligence. And just as there, the most interesting suspicion is that the question itself is bent.

The sharpest version: the philosophical zombie

Picture a creature identical to you, atom for atom, function for function. It says the same things, pulls its hand back from the heat, talks about its feelings, exactly like you. But there is no light on inside. Nothing feels like anything. Nobody is home. Philosophers call it a philosophical zombie, and they use it to make the question knife-sharp.

David Chalmers used it to pose what he called the hard problem of consciousness: if such a zombie can even be imagined, then inner experience must be something extra, on top of the machinery, that you could in principle remove without changing a single action. The whole debate about qualia, the private sense of how red looks or how pain hurts, rests on the zombie being possible.

Why the zombie can't be built

This is where friction theory punctures it. To be functionally identical to you, the zombie has to run the same races between competing routes, with the same friction, every time something is at stake and has to be settled. And here is the crux: inner experience is not an extra layer painted on top of that race. It is the system registering its own friction as it resolves it. Liking something is a layer registering that the friction fell; something hurting is a layer registering that it rose.

A creature that does all the computing but registers nothing would have to be entirely without friction at exactly the point where the registering happens. And a real system that runs races cannot be that. The zombie asks for something that runs the races without in any way feeling them run. It can't be done, for the same reason the perfectly rational economic agent or the frictionless market can't: they are all frictionless constructions that require a system with no friction at all, and no such thing can be built.

So what is inner experience?

When the zombie falls, the question falls with it. “Does it have qualia, a separate inner glow?” asks for a fact on top of all the processing, and there is no such extra fact, not even for you. What you call your inner experience is not a little cinema behind the eyes where a self sits and watches the show. It is what the processing is like from the inside as it registers its own friction.

There is no extra, magic ingredient waiting to be switched on. The feeling is the registering, not a room behind it. This is the core of what the philosopher Daniel Dennett calls illusionism. It is a position in a live philosophical fight, with Chalmers on one side and Dennett on the other, but it is the one the friction lens points to.

What that means for AI

So the honest question about AI is not whether it has crossed a line and suddenly acquired qualia, because there is no such line. The machinery that registers its own friction, it already has. The difference lies somewhere more down-to-earth.

A language model is almost born with everything at once. It got all its knowledge in one great cut during training, but it never lived with it: never used it, saw what happened, and kept the memory of that. A person has lived a life and laid down traces along the way. That is what makes something rich enough to be at stake, and what lets old mistakes leave their traces. Today's AI keeps none of those memories from one conversation to the next, so there is only a little to register. That isn't a missing spark. It is that it hasn't lived yet.

This is not a wall that has been built. If we trained models on what they themselves experience, and let them keep their memories, they would probably show many of the same phenomena. Some products have already started, and in one of my experiments I install human field-functions straight into a model. There is one point where this doesn't hold: the social amplifier, the thing that makes being cast out hurt, didn't take when we tried. But the rest is a matter of installing it, not an extra threshold to cross. There is only a life to be lived and remembered.

Stop asking whether the light is on

Stop asking whether the light suddenly switches on at one particular place. Ask instead what the system registers, and how richly. There is no hidden switch. There is only how much there is to feel, and how fully it gets felt.