Artificial-intelligence systems such as large language models (LLMs) can exhibit behaviour that seems strikingly intelligent. This has led many people, including some experts, to wonder whether these systems could have some form of consciousness — the capacity for subjective experience. However, we think that existing AI systems are not conscious, and that, along current trajectories of development, they are not likely to become conscious. Greater clarity about whether AI could become consciousness will be crucial as society grapples with the challenges posed by this increasingly powerful technology.
Citation needed on all of this, and even if you have citations, LLMs are deterministic, and do not learn or change over time. They can’t even be said to exist continuously since they only exist during the brief time that they’re responding to inputs, and then they disappear until the next input. I would argue that all of these things would need to change in order for them to be able to have consciousness.
They can learn and change over time, that’s just not enabled. The services that offer them want a reliable, reproducible, controllable experience. There’s zero reason they can’t continue to change through usage, other than it’s a conscious decision not to enable that functionality.
That’s why they can only be conscious when they train using new data, and run asynchronously with continual input - the same way we do.
Edit: People downvote me, and I suspect it’s because they’re thinking of the public facing models. I am talking about progress behind closed doors. Even some of the public facing models are gaining the ability to learn, via a mechanism being called memories. It’s just a short text note that is passed into the context window of every prompt, but isn’t that exactly what our short term memory is? In my view is no different. Pruning the inclusion list to include only the relevant information is the equivalent to use forgetting some information, but if it’s kept in long term storage with reference links to other memories, it can be “remembered” and brought back into the current context. It’s functionally the same as our own learning and thinking.
Even if we could update an LLM’s weights continuously, the algorithm is still deterministic, like all computer algorithms. There is no evidence that humans are deterministic (or a “finite state machine”) in any way. So I don’t think such a dynamically-updating LLM could be conscious, either.
It is my personal belief that all life is deterministic. It’s the only plausible explanation for instinctual behavior - like monarch butterflies returning to their breeding grounds they’ve never been to. Or people yawning because others yawn. These aren’t conscious decisions, they’re peeks at the deterministic coding that life operates upon.
I don’t think you understand what “deterministic” means.
I very much do. It means that given the same inputs to a system, you can expect the same output. I believe that to be one hundred percent the incontrovertible reality of life. The problem is it is not a testable theory, because the input includes the state of the system prior the new input. You can never recover that information, so there is no way to put it through and see if you will get that same output. Spacetime coordinates and the position of the cosmos themselves are all variables in this deterministic function, but it’s still deterministic.
OK, that’s great. So if I understand you correctly, it seems like you’re admitting that not only do you have no empirical evidence for your position, you don’t believe you’ll ever be able to test it. This seems like a weak position to hold, almost akin to a faith-based religious position.
As I’ve said, I think the existence of instinctual behavior qualifies as empirical evidence for my theory. I don’t believe it on a whim. Another example is twins, separated at birth, yet growing up to have nearly identical preferences and lives. Making choices reflecting the deterministic coding of their internal decision structure, which they largely share outside of nominal differences resulting from environment. That is rather strong evidence, I feel.
How have you ruled out other explanations for instinctual behavior? For example, there is a body of scientific work which offers evidence that instincts are passed down through genetics, a mechanism which doesn’t require that the brain be deterministic. What are the scientists studying this area getting wrong?
What about twins that are separated at birth but grow up to be different? It seems that the existence of even one such pair of twins would invalidate your claims, given how fundamental this would be to the operation of the brain.
A computer algorithm can absolutely be non deterministic, pick your language and run random.rand()
There is a whole class of such algorithms https://en.wikipedia.org/wiki/Nondeterministic_Turing_machine
But I don’t see how it makes a difference, even deterministic algorithms can produce infinite complexity, eg linear neural networks are proven to be able to compute any computable function, or the Mandelbrot set which is produced from a single recurrent statement but encodes a space of infinite complexity
Maybe your point is to say that a brain is performing a non computable function…but since your brain is computing that function…is that a counter example?
How are the numbers generated by random.rand() made?
Also, NTMs are not algorithms, they’re models of computation in theoretical computer science.
This is nonsense.
My point is that for brains, it’s not possible to recreate the exact same inputs to be able to test whether the brain produces the same output. So it’s literally not possible for brains to be deterministic.
Yet.
No, not ever.
You can hook computer randomness up to truly random* quantum processes with fairly simple specialized chips. Building your argument on such an easily avoided technical detail isn’t wise, because the question can very easily just be asked for a computer that does have access to a source of true randomness. Then it can be nondeterministic-- so what does that have to do with its consciousness or lack thereof?
* unless the entire universe is secretly deterministic, I guess.
I know.
I’m not building my argument solely on this, but maybe I’m being unclear, so I apologize.
Let’s do exactly what you propose; now we pass a seed to the LLM generated from a true hardware-based random number generator. That doesn’t change anything about the LLM itself: it’s still deterministic. I don’t think it’s reasonable to claim that a deterministic algorithm can be conscious, and I think there are other factors as well (the fact that it doesn’t exist continuously, it exists for brief periods simultaneously in thousands/millions of instances at once, the fact that it is static and unchanging unless re-trained, etc) which all paint a pretty clear picture that it cannot have consciousness.