I often think of human intelligence as having a pre-conscious thought generator to basically come up with ideas that might be bulls***. This is what we observe in mindfulness meditation as the monkey mind, or what some people have called "babel". But we also have an editor or curator who looks at the stream of babel and selects those ide…
I often think of human intelligence as having a pre-conscious thought generator to basically come up with ideas that might be bulls***. This is what we observe in mindfulness meditation as the monkey mind, or what some people have called "babel". But we also have an editor or curator who looks at the stream of babel and selects those ideas which make sense in light of a broad cognitive model of reality.
It seems to me like the current generative AI models do a pretty good job of brainstorming and remixing stuff to produce some raw idea material but they totally lack the editor/curator functionality of human intelligence. When using Bing chat, the human user is essentially playing the role of editor to the babel produced by the model.
I would like to understand more why the second part is so difficult to build into the systems themselves. It seems these companies take a very ad hoc approach with the alignment stuff, which is essentially a kind of dumb editor that is slapped on after the fact, and is not integrated in an intelligent way.
I know this is an unsolved problem, but I'm curious what are the current best ideas for how we might build such a combination.
"we also have an editor or curator who looks at the [unconscious] stream". Hmmm. So who is watching over that little homunculus? It's mother? Any such supposition remains within the realm of old-school philosophy.
I'm not sure what you mean when you say somebody would need to watch over the editor or the curator? I'm talking about different cognitive functions here, they just do their thing no supervisor is needed. That said, there may be additional layers of metacognition in humans. I'm not claiming that there is exactly two layers. It could well be that there are cognitive functions that monitor and fine tune are basic thought editor. I'm not a cognitive scientist, but I have read a lot of it and this sounds like the sort of thing they talk about.
It seems to me your use of the term homunculus here is entirely rhetorical. You could say that about any theory which posits distinct cognitive modules that do stuff, which is pretty much all of modern cognitive science.
I'm arguing that you're invoking a Cartesian theatre explanation for human consciousness. I'm not a fan of such dualism, preferring emergent theories. See https://en.wikipedia.org/wiki/Homunculus_argument for more details of the fallacy you're trying to use.
Nope. The real world (whatever that might mean) doesn't exist inside our heads. These purely subjective observations are only evidence for the brain being a hallucination/simulation generating machine, from which our consciousness arises as the 'best fit' to the current sensory inputs.
I often think of human intelligence as having a pre-conscious thought generator to basically come up with ideas that might be bulls***. This is what we observe in mindfulness meditation as the monkey mind, or what some people have called "babel". But we also have an editor or curator who looks at the stream of babel and selects those ideas which make sense in light of a broad cognitive model of reality.
It seems to me like the current generative AI models do a pretty good job of brainstorming and remixing stuff to produce some raw idea material but they totally lack the editor/curator functionality of human intelligence. When using Bing chat, the human user is essentially playing the role of editor to the babel produced by the model.
I would like to understand more why the second part is so difficult to build into the systems themselves. It seems these companies take a very ad hoc approach with the alignment stuff, which is essentially a kind of dumb editor that is slapped on after the fact, and is not integrated in an intelligent way.
I know this is an unsolved problem, but I'm curious what are the current best ideas for how we might build such a combination.
"we also have an editor or curator who looks at the [unconscious] stream". Hmmm. So who is watching over that little homunculus? It's mother? Any such supposition remains within the realm of old-school philosophy.
I'm not sure what you mean when you say somebody would need to watch over the editor or the curator? I'm talking about different cognitive functions here, they just do their thing no supervisor is needed. That said, there may be additional layers of metacognition in humans. I'm not claiming that there is exactly two layers. It could well be that there are cognitive functions that monitor and fine tune are basic thought editor. I'm not a cognitive scientist, but I have read a lot of it and this sounds like the sort of thing they talk about.
It seems to me your use of the term homunculus here is entirely rhetorical. You could say that about any theory which posits distinct cognitive modules that do stuff, which is pretty much all of modern cognitive science.
I'm arguing that you're invoking a Cartesian theatre explanation for human consciousness. I'm not a fan of such dualism, preferring emergent theories. See https://en.wikipedia.org/wiki/Homunculus_argument for more details of the fallacy you're trying to use.
Intrusive thoughts and schizophrenia are good evidence something like Avi is describing is actually happening.
Nope. The real world (whatever that might mean) doesn't exist inside our heads. These purely subjective observations are only evidence for the brain being a hallucination/simulation generating machine, from which our consciousness arises as the 'best fit' to the current sensory inputs.