3 ms·

Of course, DALL-E can only get away with not being introspective because it has reinforcement learning holding its hand. Bit harder to curate a clean training s

Of course, DALL-E can only get away with not being introspective because it has reinforcement learning holding its hand. Bit harder to curate a clean training set without consciousness. Personally, I suspect consciousness is intrinsically connected with whatever lets humans learn things by looking at and thinking about a single example.

soulofmischief 4y ago
I get what you're saying, but it's important to also consider how evolution provides for us as well by giving us neural structures already tuned for things like recognizing faces and recognizing danger. I do wonder how much realtime danger analysis has to do with consciousness.
- FeepingCreature 4y ago
  I'd doubt it has much of a connection. Consciousness can do many things, but generally not quickly.
  - soulofmischief 4y ago
    Hmm. I'd say subconscious might get you away from immediate danger, but it's often a conscious decision to anticipate future danger based on past events, and this feedback loop is sometimes quite small, such as deciding which path to take while being pursued.
    - salawat 4y ago
      I'm reminded of a meme... >Giant robot pilot looking smug Caption: I am a genius! Picture break: >same robot pilot now looking horrified. >Caption: Oh no! I can tell you right frigging now, the average person is trained through education to build up confidence they know what they are doing, then over the next decade plus face-checked against reality to build an aversion/paranoia matrix. The modern world is essentially orchestrated by people who build self-sustaining support structures of specialists. We've done a great job at building co-dependent specialists, but a terrible one at building generalists.
TrevorJ 4y ago
Reinforcement learning is the thing that integrates the network across time though. In one sense, I think it's doing at least part of the job that consciousness does: mediating between memory(past) and the present.
- FeepingCreature 4y ago
  Yes, it's just that because it's an external process, DALL-E can't go meta on it like we can. So you won't ever get "appearance of consciousness" out of a process that depends on RL, imo, because you've located part of its cognition outside its brain and made it intermittent.

soulofmischief 4y ago

I get what you're saying, but it's important to also consider how evolution provides for us as well by giving us neural structures already tuned for things like recognizing faces and recognizing danger. I do wonder how much realtime danger analysis has to do with consciousness.

FeepingCreature 4y ago
I'd doubt it has much of a connection. Consciousness can do many things, but generally not quickly.
- soulofmischief 4y ago
  Hmm. I'd say subconscious might get you away from immediate danger, but it's often a conscious decision to anticipate future danger based on past events, and this feedback loop is sometimes quite small, such as deciding which path to take while being pursued.
  - salawat 4y ago
    I'm reminded of a meme... >Giant robot pilot looking smug Caption: I am a genius! Picture break: >same robot pilot now looking horrified. >Caption: Oh no! I can tell you right frigging now, the average person is trained through education to build up confidence they know what they are doing, then over the next decade plus face-checked against reality to build an aversion/paranoia matrix. The modern world is essentially orchestrated by people who build self-sustaining support structures of specialists. We've done a great job at building co-dependent specialists, but a terrible one at building generalists.

TrevorJ 4y ago

Reinforcement learning is the thing that integrates the network across time though. In one sense, I think it's doing at least part of the job that consciousness does: mediating between memory(past) and the present.

FeepingCreature 4y ago
Yes, it's just that because it's an external process, DALL-E can't go meta on it like we can. So you won't ever get "appearance of consciousness" out of a process that depends on RL, imo, because you've located part of its cognition outside its brain and made it intermittent.