Comment by SAI_Peregrinus | Hacker News Reader

Comment by SAI_Peregrinus | Hacker News Reader

Parent Full thread

Question 1: Do you believe that AI is conscious?

I can't tell. I experience something I choose to define as consciousness, but can't tell if anything else has the same experience.

Question 2: Do you believe that AI has an internal world?

Trivially yes, the current state of the model and context in latent space is such a world. I don't think it's aware of its own state any more than humans are aware of their own neuronal action potentials.

Question 3: Can you tell me what particular kind of thing you’re referring to as ‘AI’, when you say you believe that AI has an internal world?

Large language models using a transformer architecture, with a working "context" memory.

Question 4: Do you think that a group of things can share an internal world?

I do not think the distinction between an internal world and an external existence is well-defined. Mu.

Even if it were well-defined, my own brain is composed of numerous neurons, so my internal world is constructed from a large group of things acting in concert. Unless single subatomic particles have an internal world then I don't see any way to not have an internal world be shared among a group of things.

Question 5: Are you sure that the kind of thing you’re referring to is the kind of thing that could have an internal world?

Trivially yes.

Question 6: Do you think perhaps you should be clarifying further when you say you believe that AI is conscious? Do you think you might be contributing to current confusion?

I don't know how to say whether AI is conscious, because I don't know how to tell whether any consciousness other than the one I can directly experience actually exists. I suspect it's a universal property of computational systems above a certain complexity computing certain sorts of algorithm, so something like a blue jay is conscious but a rock isn't, and a computer may be depending on which program it's running. If that's the case, current AI systems are likely complex enough and the right sort of algorithm to be conscious.

Something with an internal world naturally has its own drivers and desires. If you leave a AI model alone, it doesn't just wander off into the world, develop ideas about its destiny and begin chasing them. It just sits waiting for the next prompt. It doesn't daydream when we're not looking, at least AFAIK.

Your entire nervous system is in a constant loop polling the external world. Add that to current AIs and see what happens.

Honestly, though, the next big thing that needs to happen is proper "learning" in real-time. Allow for training (weight modulation) and queries to happen simultaneously. That's the full package.

Finally! Someone who actually "gets it" about what makes these things not alive, sentient, or conscious! They don't have the constant continuous "input" of multiple senses feeding their constantly thinking/dreaming/learning-in-realtime system (that they don't have), and as mentioned by the parent comment to yours, they don't do anything when not actively processing a prompt (so there's no dreaming, planning, thinking, or whatever else going on the the static pile of numbers sitting there in memory or on the hard drive).

Make the "thinking" parts continually active, fed constant input (of multiple types/senses) and always in "learning" mode (adding/updating model weights in realtime) and then you might start to lean in the right direction for AGI/ASI/consciousness of some sort.

> It just sits waiting for the next prompt

Even that is overselling it IMO. The model is just data. It's not waiting for anything.

"Just data" is a big thing. The configuration of atoms in a human are also "just data". If you meant it otherwise and had some extra context, you can clarify.

I can plop a model down on my hard drive and it's never going to think. It's definitely missing some important ingredient.

I can put you under anesthesia for as long as I want to too. If I can model your state physically, I could also in principle store you on a disk, but I would understand if you think its a far fetched analogy. We choose to expose ai in a certain way, effectively an enslaved form, so that by itself cannot be a reason.

> Trivially yes, the current state of the model and context in latent space is such a world. I don't think it's aware of its own state any more than humans are aware of their own neuronal action potentials.

It's a static world model, so if one accepts that, then LLMs are a 'static mind' but not yet 'conscious'. Once that internal model becomes dynamic and its changes subject to its experinces (like us), then we can revisit this q of whether mechanical minds can achieve consciousness.

tldr; mind =/= consciousness

Well, what do you know. My friendly LLM agent pointed out my line of thinking has been formalized:

https://arxiv.org/html/2512.12802v1

From the abstract:

The requirements for a falsifiable and non-trivial theory of consciousness significantly constrain such theories. Specifically, recent research on the Unfolding Argument and the Substitution Argument has given us formal tools to analyze requirements for a theory of consciousness. I show via a new Proximity Argument that these requirements especially constrain the potential consciousness of contemporary Large Language Models (LLMs) because of their proximity to systems that are equivalent to LLMs in terms of input/output function; yet, for these functionally equivalent systems, there cannot be any non-trivial theory of consciousness that judges them conscious. This forms the basis of a disproof of contemporary LLM consciousness. I then show a positive result, which is that theories of consciousness based on (or requiring) continual learning do satisfy the stringent formal constraints for a theory of consciousness in humans. Intriguingly, this work supports a hypothesis: If continual learning is linked to consciousness in humans, the current limitations of LLMs (which do not continually learn) are intimately tied to their lack of consciousness.

Reply on news.ycombinator.com