I’m going to sound crazy (because I am), but I think counciousness can only be created by a specific kind of cognitive process, that the “brain” is automatically connected and receive direct information from it’s behaviour.
Like, if a cognitive process, imagine an undescribed algorithm, receives immediate information from it’s behaviour and automatically acts upon itself, changing the algorithm itself, so it can relate cause and effect, being the only basic form of interaction that can produce this cognitive process, the basis for any higher function as “I (internal) interact (made something) with the world (the outside reaction to what I made)”.
This would mean that counciousness would need this step, not that this step would necessarily create counciousness.
These LLMs are prebaked dictionaries in this form of view.
Well, yeah, they have fixed weights and the architecture is inspired literally from a vectorized key-value database structure. What you are describing seem to be recursive self learning, which is essential for adaptation. Consciousness is not even well defined, so I wouldn’t bother making sense of it.
What you are describing seem to be recursive self learning
I’m talking in general, not applied to this specific case, just what is the minimum for a cognitive process to be able to reach this specific type of knowledge. Consciousness at least has a recognition of itself, a “minimal” generalized definition.
More related to how to define such a system in general and what it takes.
The problem is also labeling LLMs, spicy autocompletion programs, as “Artificial Intelligence”.
There’s absolutely no intelligence in there. There’s no consciousness. It doesn’t think. It’s make to appear so. But it’s an illusion. It’s just probabilistic routines.
Slapping a shiny “AI” sticker on it doesn’t suddenly make it the real deal.
Just like those Hover Boards from a few years ago, that should really have been called Rover Boards instead, but the word “Hover” feels more futuristic.
So pushing for the Great Intelligence with current models, only the delusion is real.
LLM are based on transformer models, which are in turn based on ANNs. Research on ANNs is part of artificial intelligence. You don’t need to be conscious to solve a problem.
It’s just probabilistic routines
The same goes for your own brain. There is a lot to learn about it, but the consensus is that we can model much of our goal oriented behavior as reward seeking probabilistic agentes.