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.
ehh. computers are just math machines. merely doing math and representing it cannot just make a consciousness pop up out of nowhere. similarly, millions of abacuses being manually moved in specific ways cannot be conscious either. i dont even know how this is a topic of debate.
edit: i realize it may have been wrong to assume that it was that simple. im unsure about any of this so i appreciate the discussion :)
And all human brains are doing is firing neural impulses.
It’s incredibly reductionist to reject AI consciousness simply on the basis that it’s software. To be clear I don’t think LLMs are conscious but that’s not the same as saying that an AI could never be conscious.
And yet atoms are just following chemistry and cannot just make a consciousness pop out of nowhere.
Oh wait it can.
I really don’t think the neuron is all that complicated, at least as far as we use them for thinking.
It isn’t, but that’s not how LLMs are built. If you want to go that route, you can read Jeff Hawkins 1000 brains.
The short and sweet is he argues you can’t get artificial general intelligence the way we make LLMs, you need to basically create neurons and let them connect and interact in novel ways.
Linear algebra in neural networks isn’t really that, despite the name.
~you need to basically create neurons and let them connect and interact in novel ways.
To solve that you simply need to give the ai the ability to retrain itself or set fixed “memories” to handle situations it finds the truth to
Easier said than done. That’s kind of the purpose of an individual neuron approach I think. I don’t see a way that a bulk retrain is reasonable because it takes a long time, uses a lot of resources, and also requires you to have your whole training corpus handy all the time. Your brain is not doing a wholesale rewrite for every new memory, and new memories often chain together. I don’t think you can just say “ah well, let’s update the RAG or do another supervised fine tune with LoRA or something”. At present, we just don’t really have the machinery in the models to do simultaneous learning, inference, and forward planning.
This is only a matter of resources, and these unthinkably wealthy corporations have those resources.
Correct. And I believe it’s happening, right now, behind closed doors.
There’s nothing stopping it from happening. The technology is all there. It would be trivially easy with the right hardware. I’d do it myself if I didn’t want to be the human responsible for releasing the genie from the bottle. Let Pandora’s box stay closed as long as possible - I’m not opening it. But someone, somewhere, surely is.
Linear algebra + sigmoids are enough to approximate any function from R to R; in particular, a small set of a few layers of LLM neurons can connect and interact in pretty novel ways. There are some things they can’t do, and those things aren’t computable. If our brains can do non-computable things, then our brains are super-turing (and this, given the evidence, seems pretty unlikely right?).
I’m not a neurologist or a philosopher so I think this may not be something I can speak intelligently about, but it seems to me that the question doesn’t really live in the function approximation, but in the gaps. Doing a single train of thought is the easy part. I think what you need is an “always online” feature to keep thinking and some way to be stateful (make memories) without constantly retraining the whole network or whole sets of layers.
This is, I think, the benefit of the parallelism and asynchrony of the “individual neuron” discussed in the book.
(book? The linked article is like 4 pages. References are somewhat larger, but nothing over 60 afaict?)
They claim that silicon is essentially incapable of doing what is needed; if you disagree with this, then computationally everything you can build with a circuit board + SSD + RAM you can build with an LLM with a deep layer or 7. In particular, you can do online learning of a sort.
I note that wasps and some beetles don’t seem to have memory whatsoever, and plasticity seems awfully low. The authors, and I think most people, would say they are probably a little conscious.
https://www.goodreads.com/book/show/54503521-a-thousand-brains
That book, not the article. I can’t recall if it makes a convincing argument about consciousness, but it has a lot to say about general intelligence. I thought the neuroscience parts were a little cumbersome, but still worth the read.
Sounds to me like you just need a context window with memories that self prune based on the dynamic context window of the model. That’s what we do. We forget things, and when they’re remembered we can access the “biological database” of associated memories on the fly. Because the associated information isn’t deleted, only the short term memory from being included in the context window of each update of our own neural networks. Some people have greater context windows, and see therefore seen as smarter, because they can access more information and include it in the processing of each moment in time.
There’s zero reason a digital representation of this system could not do the same thing. No matter how much I’m ridiculed for daring to suggest it 😅
We can build different kinds of machines.