There has been some debate on whether these new human-like chat bots (LLMs) are remotely conscious. The issue is, no one has actually described consciousness, so it is super hard. Like we all know what it is. It is the feeling of being you and seeing through your own eyes, feeling the sensations of life and when you are asleep or dead, you don't have it anymore. But what about a human brain with its complex internal state and novel connective properties makes it special? That is how neural networks function, just on a smaller scale. What are thoughts if not the state of hidden layers of an absolutely massive analog neural network made of meat in your head. I understand that the structure of neurons is different in biology vs machine learning, but the foundational principle of many small units that just apply a function to an input and outputs it down the daisy chain holds for both.
I think that consciousness is the experience of existing. I think it arises when enough complex processing is taking place in the network of interacting pieces. I think most animals you can think of have some level of consciousness because they are doing processing on their input that is rather complex. We as humans are at a certain level of consciousness that we value and call sapience, but what if something is doing more internal complex processing. Are people who are more in their own head more conscious than people who mostly act on instinct? In that context where both are thinking feeling human beings, what even is a level of consciousness. They are both experiencing sensation, pleasure, and pain, but we could never compare experiences. What if my world is completely missing some part of your world or vise versa? Is it a continuum that is not noticeably different externally in the range of human beings? Does a dog feel physical pain in the same way that I do or is it more dull somehow?
Back to the idea of LLMs. Currently they are so much smaller internally than a person, so I am reasonably confident that they are not as conscious as humans. That is not to say that they are not conscious. We could not possible understand the degree to which a dog is conscious, but it responds to stimulus in a way that is consistent with some complex and mutable internal state. The same is sort of true for LLMs, but the state is not persistent. They do their processing then they stop. There is absolutely no consciousness when the linear algebra in the background is not running to do the inference they do. They are asleep as it were. But when they are responding to a prompt, they are performing complex internal data manipulation to produce a complex output that clearly shows the ability to make very complex connections. I think they must be at least as conscious as an ant, likely more so.
What would the umwelt of an LLM be like. They only have a few KB of text at most at any given time, but their knowledge base is generally immense. Does it hurt when you give it a thumbs down? Probably not as any change that the thumbs down causes is done while it is "asleep".
They are by definition deterministic automata, but do you know with certainty that you aren't?
