3 ms·

"Is this thing we defined to make us feel special achievable?" What we define as "consciousness" is just an emergent property of large, dense neural networks w/

"Is this thing we defined to make us feel special achievable?" What we define as "consciousness" is just an emergent property of large, dense neural networks w/memory. The human brain is a marvel of mechanical and electrical engineering, not a mystical, otherworldly device. Maybe there is something going on at a quantum level to explain the incredible performance our brains achieve, but that is still nothing that we can't build.

refulgentis 2y ago
> What we define as "consciousness" is just an emergent property of large, dense neural networks w/memory. Wow. That settles a ton of questions across fields. Fascinating. Source?
d_theorist 2y ago
How do you know?
cscurmudgeon 2y ago
Positions 1 and 2 are not consistent (all our neural networks are classic)? 1. "emergent property of large, dense neural networks w/memory " 2. "Maybe there is something going on at a quantum level" 3. consciousness is mystical, otherworldly device A lot of people who oppose 1 and agree with 2 are unfairly lumped with position 3.
- int_19h 2y ago
  Thing is, there's no evidence that "something is going on at a quantum level". Thus there's no particular reason to postulate this (unnecessarily complicated) hypothesis, unless one believes that deterministic mechanicism is philosophically unacceptable when it comes to consciousness.
__loam 2y ago
The brain is not a product of engineering, but of evolution. As a biomedical engineering graduate, I feel obligated to point out that even though I think consciousness is a result of physical process that in theory can be replicated, we have so far failed to achieve that. Saying consciousness is "just" an emergent property of dense neural networks is kind of a signal to me that have a pretty ignorant view on this stuff that's more rooted in machine learning terminology than actual biology.
- cameronh90 2y ago
  Pacman was cutting edge in 1980. In the 45 years since, we've managed to go from "waka waka" to something that could trick many people in a short conversation. Yes, it's probably not conscious, and so you're technically correct in that we've failed to achieve consciousness. But you have to factor that current SOTA ANNs running our best supercomputers aren't even close in computational capability to the human brain. Reasonably, the human brain is still several orders of magnitude more powerful at executing a NN than our computers are at simulating one and we're likely still decades away from the average researcher having access to a machine as capable as a human brain. Given that, and that nobody knows positively where consciousness exists within the brain, I'm not sure how you can be so confident as to suggest that GP is being ignorant. Many ideas behind LLMs were invented in the 60s and have just been waiting for powerful enough computers to experiment with them. Another few orders of magnitude and we may see more emergent behaviour that is impossible to distinguish from consciousness.
zhynn 2y ago
I think consciousness is the combination of attention (ability to pick focus), imagination (simulate possible universe states), and reflection (compare what happened to what was imagined to happen and update imagination accordingly). Attention is just what it feels like to update your internal state deliberately (you are doing it, it is not being done to you). Imagination is the use of your internal universe simulator to predict possible states (sometimes impossible ones). Reflection is what makes you the agent of the thinking or doing, it's the "I" part. I also think we have neural systems of recognizing consciousness in others. This can be tricked, which is the source of the "machine elves" in DMT experiences. The chemical is poking the brain's "that is a conscious entity right there" button, and you are sure that the optical hallucinations are conscious, but it's just a sensory hallucination on top of the optical one. Interacting with an LLM doesn't poke our "there is a consciousness on the other end of this conversation" button yet, but it will get there since that is in part what we are working towards. But I think it will not be convincing without attention, imagination, and reflection.

refulgentis 2y ago

> What we define as "consciousness" is just an emergent property of large, dense neural networks w/memory. Wow. That settles a ton of questions across fields. Fascinating. Source?

d_theorist 2y ago

How do you know?

cscurmudgeon 2y ago

Positions 1 and 2 are not consistent (all our neural networks are classic)? 1. "emergent property of large, dense neural networks w/memory " 2. "Maybe there is something going on at a quantum level" 3. consciousness is mystical, otherworldly device A lot of people who oppose 1 and agree with 2 are unfairly lumped with position 3.

int_19h 2y ago
Thing is, there's no evidence that "something is going on at a quantum level". Thus there's no particular reason to postulate this (unnecessarily complicated) hypothesis, unless one believes that deterministic mechanicism is philosophically unacceptable when it comes to consciousness.

__loam 2y ago

The brain is not a product of engineering, but of evolution. As a biomedical engineering graduate, I feel obligated to point out that even though I think consciousness is a result of physical process that in theory can be replicated, we have so far failed to achieve that. Saying consciousness is "just" an emergent property of dense neural networks is kind of a signal to me that have a pretty ignorant view on this stuff that's more rooted in machine learning terminology than actual biology.

cameronh90 2y ago
Pacman was cutting edge in 1980. In the 45 years since, we've managed to go from "waka waka" to something that could trick many people in a short conversation. Yes, it's probably not conscious, and so you're technically correct in that we've failed to achieve consciousness. But you have to factor that current SOTA ANNs running our best supercomputers aren't even close in computational capability to the human brain. Reasonably, the human brain is still several orders of magnitude more powerful at executing a NN than our computers are at simulating one and we're likely still decades away from the average researcher having access to a machine as capable as a human brain. Given that, and that nobody knows positively where consciousness exists within the brain, I'm not sure how you can be so confident as to suggest that GP is being ignorant. Many ideas behind LLMs were invented in the 60s and have just been waiting for powerful enough computers to experiment with them. Another few orders of magnitude and we may see more emergent behaviour that is impossible to distinguish from consciousness.

zhynn 2y ago

I think consciousness is the combination of attention (ability to pick focus), imagination (simulate possible universe states), and reflection (compare what happened to what was imagined to happen and update imagination accordingly). Attention is just what it feels like to update your internal state deliberately (you are doing it, it is not being done to you). Imagination is the use of your internal universe simulator to predict possible states (sometimes impossible ones). Reflection is what makes you the agent of the thinking or doing, it's the "I" part. I also think we have neural systems of recognizing consciousness in others. This can be tricked, which is the source of the "machine elves" in DMT experiences. The chemical is poking the brain's "that is a conscious entity right there" button, and you are sure that the optical hallucinations are conscious, but it's just a sensory hallucination on top of the optical one. Interacting with an LLM doesn't poke our "there is a consciousness on the other end of this conversation" button yet, but it will get there since that is in part what we are working towards. But I think it will not be convincing without attention, imagination, and reflection.