3 ms·

Despite very real and impressive progress with LLMs towards general conversation agents there has been near zero progress in understanding and recreating consci

Despite very real and impressive progress with LLMs towards general conversation agents there has been near zero progress in understanding and recreating consciousness and agency which even insects posses.

ilaksh 3y ago
Consciousness (another term that is too vague) sure. LLMs do not have a high-bandwidth stream of subjective experience at all similar to animals. But agency? You just give the LLM an open-ended goal and access to manipulate the environment and run in a loop. Many examples of this. Of course its not the same type of agency that animals have, doesn't have self-preservation instincts etc., but there are enough movies and TV shows demonstrating why that may be a pretty bad idea anyway.
cwkoss 3y ago
Consciousness has not been objectively proven. As far as we know it could just be something philosophers made up.
blovescoffee 3y ago
For the sake of discussion, what do you __mean__ by consciousness?
- rllearneratwork 3y ago
  that's the point. Nobody even has a good enough definition of consciousness and intelligence.
ddingus 3y ago
I want to throw something out here: Insects, US, animals, all have two things in common, and thinking about them may prove very useful in this context: (1) - Ongoing stream of what I will call input experience. What we see, feel, hear, etc... (2) - Internal input experience sufficient for the entity to operate and as part of that operation, differentiate itself from everything else. Number one is easy. Number two may be important to understand where consciousness comes from. And yes, I am saying consciousness may be an emergent thing that will emerge on some level given the entity is sophisticated enough to understand itself. Think about your own body. We have nerves and can and do get input about how that body is interacting with literally everything else! From that comes a very basic "self" ChatGPT is something in a box! It has no real sense. It cannot know it's own self. There simply is no context, no data it can use to understand. And it has no real memory. Say the sense data was there. It would have to be prompted in every time. Here is what I am getting at: Say we give ChatGPT the robust and complex sensations needed for it to differentiate itself from everything else? And then we put it on a loop so it can take input from us in that context. And it can remember. Suddenly, asking it a question interrupts it's stream of experiences! And see where that could lead?

ilaksh 3y ago

Consciousness (another term that is too vague) sure. LLMs do not have a high-bandwidth stream of subjective experience at all similar to animals. But agency? You just give the LLM an open-ended goal and access to manipulate the environment and run in a loop. Many examples of this. Of course its not the same type of agency that animals have, doesn't have self-preservation instincts etc., but there are enough movies and TV shows demonstrating why that may be a pretty bad idea anyway.

cwkoss 3y ago

Consciousness has not been objectively proven. As far as we know it could just be something philosophers made up.

blovescoffee 3y ago

For the sake of discussion, what do you __mean__ by consciousness?

rllearneratwork 3y ago
that's the point. Nobody even has a good enough definition of consciousness and intelligence.

ddingus 3y ago

I want to throw something out here: Insects, US, animals, all have two things in common, and thinking about them may prove very useful in this context: (1) - Ongoing stream of what I will call input experience. What we see, feel, hear, etc... (2) - Internal input experience sufficient for the entity to operate and as part of that operation, differentiate itself from everything else. Number one is easy. Number two may be important to understand where consciousness comes from. And yes, I am saying consciousness may be an emergent thing that will emerge on some level given the entity is sophisticated enough to understand itself. Think about your own body. We have nerves and can and do get input about how that body is interacting with literally everything else! From that comes a very basic "self" ChatGPT is something in a box! It has no real sense. It cannot know it's own self. There simply is no context, no data it can use to understand. And it has no real memory. Say the sense data was there. It would have to be prompted in every time. Here is what I am getting at: Say we give ChatGPT the robust and complex sensations needed for it to differentiate itself from everything else? And then we put it on a loop so it can take input from us in that context. And it can remember. Suddenly, asking it a question interrupts it's stream of experiences! And see where that could lead?