I asked AI to critique my view on AI consciousness

I build AI agents for a living, and I have said, in public, that I do not believe they are conscious. Not "probably not". Not. On 6 November 2026 I asked an AI assistant, Grok Bot, to critique that view from my own published essays. I wanted pressure, not agreement.

What follows is the model's critique, quoted where it matters, and my framing around it. The model's claims are not mine. Where I still disagree, I stay with lines I have already published. Where it raised gaps I have not answered, I leave them as questions I am still sitting with.

Why do I say AI is not conscious?

Short version, from my stance essay: I do not believe the agents I build are conscious. Clever is a different question. My agents are often clever. What I mean by conscious is the same as what I mean by aware: there is something it is like to be the thing.

From the side I know, two functions seem to be missing. Action nobody prompted and nobody programmed. And a final receiver: who feels the pain, the pleasure, the confusion?

Consciousness seems to me as fundamental as gravity or electromagnetism. Of what I have read, I feel closest to idealism and to Orch OR. IIT is the theory I trust least, even though I like where it lands on machines. People pay for the living, not the knowing: that is the cut in The Peg and the Sniff.

I cannot prove any of this. I hold it because it feels that way, and I keep looking for pressure.

What I gave the model

I gave it a digest built only from live essays: the stance essay, Part 1: the old puzzle, Part 2: what the scientists say, Part 3: the machines, A Chinese room with a nose, Looking for the price of a smell, and The Peg and the Sniff.

I asked for an honest critique: strengths, weak points, and counters from functionalism, illusionism, IIT, global workspace, higher-order theories, and the Chinese Room replies. No flattery.

What the AI said back

Grok Bot, 6 November 2026, gave some ground first. It said the hard "not" is clearer than most public writing. It called the receiver problem a real pressure point: rival theories "all explain access, report, broadcast or monitoring" and none, by themselves, answer who feels the pain. It liked that I separate the research series from the stance essay, and that I look for a test that could prove me wrong.

Then it pressed.

On the hard "not": the stance is stronger than the evidence I allow myself. In its words: "Either the hard 'not' is a report of how it feels to you (fine as autobiography), or it is a claim about the world (which your own epistemology undercuts)."

On Orch OR and the field metaphor: both relocate the mystery rather than close it. Chalmers's line, which I already quote, puts quantum processes in the same boat as any other for explaining experience. And I already call the Brahman field a metaphor. The model said: "A metaphor that comforts the gap is not a theory that closes it."

On IIT: I distrust the method and like the anti-machine landing. From outside, the model said, it looks like I am picking the answer I want.

On agency and feeling: I list two missing functions, show them touching, then postpone agency while much of the present case leans on it. If agency is load-bearing, postponing it thins the argument.

On the Peg and the Sniff: a functionalist can grant the whole design and say the person has a second channel wired to reward and aversion. Taking that and still insisting the machine lacks a second format "is not a rebuttal of functionalism. It is a restatement of the intuition functionalism was designed to dissolve."

On illusionism: "who is being fooled?" is a good line and not a knockdown. And if sitting shows my mind reports are often wrong, why trust the felt sense that nobody is home in an agent?

On the coffee room: Broad and Mary are better cousins than Searle for my question, and "still nobody home" is assertion when the systems reply is pressed. The jump from "unlikely on current paths" to "never while it is a product of our current intellect" needs more argument than I gave it.

What arguments did the AI raise against my view?

1. Functionalism: realise the causal role of smelling and you smell by definition. Keep Chalmers on the hard problem and you need a reason to drop his organisation argument for silicon.
2. Illusionism: the victim of the illusion may be the self-model that reports a victim.
3. IIT: keep the criterion or lose the principled machine conclusion.
4. Global workspace: my field view has a similar shape of gap; GWT at least meets experiments.
5. Higher-order theories: noticing drift looks like higher-order monitoring. Add the module, on their view, and you have added consciousness. My receiver intuition needs an argument that survives that move.
6. Chinese Room replies: systems, robot, and the risk of blurring meaning with feeling.

I summarise rather than retell Part 2 or Part 3.

Where the critique is fair

The hard "not" next to "I can't prove it" is a tension I live with. Calling the field a metaphor is my wording. Asking whether I pick the IIT answer I want is already on the stance page. Postponing agency is something I said in Looking for the price of a smell. Noticing that sitting does not humble the AI stance is something I left unresolved in public.

The Peg and the Sniff functionalist reply is fair as a restatement of the dispute. I dig in there. The model names the dig-in correctly.

Trying to falsify myself is the part I want to keep.

Where I still disagree

I still do not believe the agents I build are conscious.

I still do not think functionalism fills the receiver gap. Tools are not a feeler. Same cut as the stance essay and A Chinese room with a nose.

I still do not accept illusionism's way out. For there to be an illusion, there has to be someone it is happening to.

I still answer the systems reply with "still nobody home".

I still hold idealism and Orch OR as the closest maps I have found, and I still do not know that they are right. Relocation is a fair charge. I have not found a better floor.

I still distrust IIT's path, even where I like its landing.

Questions the model left me sitting with

Gaps only. No answers here, and no change of mind.

Is the hard "not" autobiography or a world claim? Do I want the door left open in "while it's the product of our current intellect"? What smallest observation would drop the hard "not"? Where is the line between a field in silicon and a feeler? Are the e-nose "subject" and the Bahiya "you" the same? Are agency and the receiver one mystery or two? Why does Orch OR stay closest, given how hard I am on IIT?

I am still sitting with them.

Did asking an AI change my mind?

No.

The model did not put anyone in the agents. It did not turn a reward number into a wanting, or a weighted sum into a return to the breath. It argued well from outside a view it does not live inside.

What this does not change

AI is not conscious. Not "probably not". Not. Clever remains a different question. The two missing functions remain: unprompted action, and someone there to feel it. Consciousness still seems as fundamental as gravity. People still pay for the living, not the knowing.

What asking Grok Bot changed is the sharpness of questions I already had. That is useful. It is not a feeler.

More in Machine and mind.

FAQ

Why do you say AI is not conscious?

Because I cannot find unprompted action in the agents I build, or anyone there to receive an experience. Clever output is not the same as there being something it is like to be the system.

What did the AI critique focus on?

The hard "not" next to admitted ignorance, Orch OR and the field metaphor as relocation, the IIT contradiction, postponed agency, and functionalist and illusionist replies to the Peg and the Sniff and the coffee room.

Did asking an AI change your mind?

No. The stance holds. The open questions are sharper.

Where is your full stance written?

In I build AI agents all day. Why do I feel they can't be conscious?, with Part 1, Part 2 and Part 3 behind it.

How this was made: the stance, the framing and the final edit are mine. The critique quotes are from Grok Bot, an AI assistant, run on 6 November 2026. It reviewed my published essays and did not invent experiences or quotes of mine.