Current AIs are not conscious

Thu, 04 Jun 2026 17:27:00 +0000

Current AIs are not conscious.

You know when you got a thinking inside your head that you can’t just take out and keep ruminating for days? That was mine.

I think it was because I read recently somewhere about AIs and consciouness and definition of life. I mean, at first I would say the answer for the question “Are LLMs conscious?” obviously would be “Are you drunk or what?”, it was like asking if rocks also have consciousnes too, or if my chair feel pain everytime I sit my fat ass on it.

However, I could not answer why the question was stupid. Why can I tell with 100% of certainty my chair is not suffering with all my weight? Better, do I really know that?

And then, as almost always happens, while I was taking a shower something came up in my mind, that small argument that I needed to conclude (current) LLMs and other AIs cannot be conscious or alive.

They lack something that I name physicality. It is not a lack of body, or else you could counter argument that if you connect an ChatGPT instance to a robot then it would have a physical body, therefore ChatGPT is not only conscious but alive as well (these are different things, by the way). Physicality, for me, it is the physical substrate where the thinking process, the things that make you experiment things, happens. Any AI (not only language models, but visual models, etc) are “just” a gigantic function where it takes an input, let’s say a token for a LLM, and it outputs another token. In the middle, mainly matrix multiplication. Lots and lots of matrix multiplications, so any current AI is nothing more than numbers being multiplied (and maybe summed, maybe ReLU’d, it doesn’t matter, these are just math functions). next_token = f(big_matrix, previous_token).

But, all the numbers that compose the weights and biases don’t exist physically, they are logical representations running on top of a silicon and metal base subtrate. The “neurons” in current AIs are not physical things, their structure only exist during execution time as an abstract logical layer that do not exist in your computer. There is nothing in the hardware that represent the “unit of thinking” of an AI, and if you argue electrons are the unit being used, then you would need to say computers are conscious too. In animals, the unit of thinking is the neuron, a self-sustaining physical object able to connect with other neurons that when connected in a network vast enough is able to generate consciouness. Consciouness also is needed for volition, the ability to act without external stimulation. A bacteria may move toward the food by random walk or by concentration gradient (purely physical or chemical actions), but you and your dog can and do act by yourselves most of the time, based only in your internal states (before you say “all living beings react to physical or chemical processes, you dumb”, at least animals do have different reactions to the same environmental parameters). I argue today’s AI is not even like bacteria in this sense: if you never prompt (give stimulation, generate a gradient), it will never act. An “AI agent” is a LLM that only has “agency” because the program around it keep sending tokens to it in a loop, in other words, there is no agency, it is a programmatic loop.

That gives me interesting consequences. First, I used “current AIs” above because there is no physical law that forbid us to create artificial intelligent beings, it is mere a technological and ethical problem. If you want to bring the artificial neural network architecture to physical world using silicon instead of carbon, I don’t see why you can’t, given advanced technology. Create small chips that receive voltages as input and send voltages as outputs, connected by wires. These small chips must be able to have intrisical states that change with time, varying their output according with its state. Or some other futuristic material, I don’t know. I wouldn’t be surprised if you make this network big enough similar intelligent and conscious emergent properties appear. This give you beings, not life though. For life to exist, it must be self-sustaining at least.

Second, that means we meatbags are not that special. I mean, we are in the sense on how crazy a bunch of molecules through billions of years agglutinated together to make self-propelled beings that make terrible mistakes and drown themselves under a pile of regrets, but we just follow the laws of physics like everything else. And “we” I am not talking about Homo sapiens only, but we all know dogs and cats are able to think too.

Third and final, the same way I know my computer and GPU can’t have a consciouness, my chair cannot too. They all don’t have the physical structure needed to experience the world. They are not built in a way to have elements to interact with each other to give them volition. They just “exist”.

This is not an anti-AI argument. I think the current models we have are sci-fi level already, the math behind is conceptually suprisingly simple (linear algebra), the technology is not super ultra advanced, it is just we didn’t have enough computing power until recently. I think AIs, LLMs, etc are morally neutral, but I really, REALLY HATE with all hate in the world they (people on the top) are using this to fuck us all, as always. The orchestred and intentional way they starved us all from the hardware needed to have local, not big tech oligopolized, AIs (beside all the externalities on huge data centers, misanthropic CEOs, government deals, surveillance, etc) made me anti-AI as a means to power, not as technology.

The takeaway for me? I can continue cursing my chatbot without feeling a drop of guilty.

AI on ThoughtCrime

Current AIs are not conscious