4 ms·

Having a purely representative model of the world is less useful than having a sandbox for modeling choices and outcomes. Do I need to duck before entering tha

by Retric 3y ago

Having a purely representative model of the world is less useful than having a sandbox for modeling choices and outcomes. Do I need to duck before entering that doorway? That introspective analysis is consciousness. Humans have just improved the same mechanism allowing for more abstract analysis.

mewpmewp2 3y ago
But GPT-4 can also take the decision to duck. For instance, here I have GPT-4 with a system message: "You are an AI human agent in a video game performing actions.". And I give it the following information as current environment: Current environment: 1. Your goal is to get out of the current room. 2. You are moving in direction of North at 4km/h. 3. There's a door 3m in the North. 4. The door height is 150cm. 5. Your height is 180cm. What is the sequence of actions that you plan to perform? Its response: --- 1. Slow down as I approach the door to avoid collision. 2. Inspect the door for any lock or handle. 3. If a handle or a means to open the door is present, try to open the door. 4. If the door is locked, search for a key or any other opening mechanism within the immediate surrounding area. 5. If the door isn't lock or after opening it, crouch down slightly to accommodate my height which is more than the height of the door to avoid my head hitting the door frame. 6. Proceed through the door and stand tall again after passing through. 7. Continue moving North after exiting the room. --- So here I did hint quite a bit by giving this specific environmental data, I could add more noise, and unnecessary information, but even then I think it can do it based on my experience. I had temperature of 1 for the above, which I think why it made some odd word choices, but I also decided for example to change the info where the door is to "South", and this is how it responded: 1. Stop moving in the North direction. 2. Turn around 180 degrees to face the South direction. 3. Start moving towards the door in the South direction at a moderate pace. 4. Upon reaching the door, duck down slightly to accommodate the difference in height between myself and the door. 5. Proceed through the door to exit the room." ------- So I guess, what I would like to see as an example. Assuming GPT-4 is not conscious, what is something that an average conscious human can do, thanks to presumed consciousness, that a GPT-4 couldn't do and why is consciousness required for that?
- bumby 3y ago
  I'm not convinced GPT can infer correct actions needed for novel scenarios for contexts that have not been directly experienced in training. Consider: I have never in my multiple decades of existence had a ball roll out in front of me while driving. But if I were driving by a playground and a ball rolled out in front of me from between parked cars, and the ball itself is not a hazard because it has already passed my lane, I can infer that I need to slow or stop because a child may rapidly follow that ball. In your ducking scenario, the AI needs direct perception of a door. I assume it cannot imagine a need to duck without 1) direct perception of the doorframe, or 2) previous experience of ducking through a doorframe. I'm not convinced that GPT or any AI can infer that context for handling novel edge cases at this point. It could just be a need for more training data to correlate different contexts, but I don't think it's there yet.
- Retric 3y ago
  > what is something that an average conscious human can do, thanks to presumed consciousness, that a GPT-4 couldn’t do and why is consciousness required for that? Consciousness isn’t required for anything. It’s useful for dealing with a complex and adverse environment not solving any specific problem. That said, without testing I would assume if ask chat GPT-4 to change how quickly it responds it’s going to fail based on my understanding of their architecture. As in recite the pledge of of allegiance inserting 10 second pauses between each word. That said having provided a specific test it’s possible for that particular capability to have already been added to the UI layer via a pause token or something even if the model itself responds at the same rate.
  - mewpmewp2 3y ago
    If you used GPT as a brain you could provide the current time as an input for it, but otherwise yeah, GPT doesn't have time within it's input by default, but if you did: Made the instruction to be: Say "Stop!" when 10 seconds are done. And then had a loop where You give it current time in a loop, it would surely be able to do that. But I'm not sure if or how that is exactly related to consciousness? Time, the best way to think of it is probably a series of ticks anyway, and I assume in people's brains there is something to do that, so it would be reasonable to add a mechanism for GPT as well. GPTs goal is to only act as the intelligent part of the brain based on the input.
PeterisP 3y ago
Analysis of what-if scenarios and simulating likely expected outcomes of proposed actions is a very useful ability, but IMHO that is not part of what we'd call 'consciousness'.
- Retric 3y ago
  It’s a large fraction of what consciousness is actually doing, not the definition of consciousness. Someone considering what to say is simulating tradeoffs. In practice simulation isn’t limited to future possibilities. Trying to remember something you might recall vague sensory information like the tilt of your hips and try and create a scenario where that fits.
  - PeterisP 3y ago
    I would not automatically treat all the things that people do consciously as intrinsically linked with consciousness, unless there's some clear reasoning why consciousness is required to do that particular thing - and not in the sense that it's required because unconscious people can't do stuff because they're incapacitated. Like, saying that simulating tradeoffs is what consciousness is actually doing seems to me equivalent to saying that smashing my fist on the table is something 'my consciousness' is doing, which is kind of technically true but not really relevant, as IMHO a non-human entity that's clearly without any consciousness could both smash something against the table and also simulate tradeoffs for various options, having a subjective experience isn't required for that.
    - Retric 3y ago
      Sleepwalking shows most things people do can happen without consciousness. People can make breakfast without being conscious etc. Choosing between tradeoffs is however intrinsically linked to consciousness.

by Retric 3y ago

mewpmewp2 3y ago

But GPT-4 can also take the decision to duck. For instance, here I have GPT-4 with a system message: "You are an AI human agent in a video game performing actions.". And I give it the following information as current environment: Current environment: 1. Your goal is to get out of the current room. 2. You are moving in direction of North at 4km/h. 3. There's a door 3m in the North. 4. The door height is 150cm. 5. Your height is 180cm. What is the sequence of actions that you plan to perform? Its response: --- 1. Slow down as I approach the door to avoid collision. 2. Inspect the door for any lock or handle. 3. If a handle or a means to open the door is present, try to open the door. 4. If the door is locked, search for a key or any other opening mechanism within the immediate surrounding area. 5. If the door isn't lock or after opening it, crouch down slightly to accommodate my height which is more than the height of the door to avoid my head hitting the door frame. 6. Proceed through the door and stand tall again after passing through. 7. Continue moving North after exiting the room. --- So here I did hint quite a bit by giving this specific environmental data, I could add more noise, and unnecessary information, but even then I think it can do it based on my experience. I had temperature of 1 for the above, which I think why it made some odd word choices, but I also decided for example to change the info where the door is to "South", and this is how it responded: 1. Stop moving in the North direction. 2. Turn around 180 degrees to face the South direction. 3. Start moving towards the door in the South direction at a moderate pace. 4. Upon reaching the door, duck down slightly to accommodate the difference in height between myself and the door. 5. Proceed through the door to exit the room." ------- So I guess, what I would like to see as an example. Assuming GPT-4 is not conscious, what is something that an average conscious human can do, thanks to presumed consciousness, that a GPT-4 couldn't do and why is consciousness required for that?

bumby 3y ago
I'm not convinced GPT can infer correct actions needed for novel scenarios for contexts that have not been directly experienced in training. Consider: I have never in my multiple decades of existence had a ball roll out in front of me while driving. But if I were driving by a playground and a ball rolled out in front of me from between parked cars, and the ball itself is not a hazard because it has already passed my lane, I can infer that I need to slow or stop because a child may rapidly follow that ball. In your ducking scenario, the AI needs direct perception of a door. I assume it cannot imagine a need to duck without 1) direct perception of the doorframe, or 2) previous experience of ducking through a doorframe. I'm not convinced that GPT or any AI can infer that context for handling novel edge cases at this point. It could just be a need for more training data to correlate different contexts, but I don't think it's there yet.
Retric 3y ago
> what is something that an average conscious human can do, thanks to presumed consciousness, that a GPT-4 couldn’t do and why is consciousness required for that? Consciousness isn’t required for anything. It’s useful for dealing with a complex and adverse environment not solving any specific problem. That said, without testing I would assume if ask chat GPT-4 to change how quickly it responds it’s going to fail based on my understanding of their architecture. As in recite the pledge of of allegiance inserting 10 second pauses between each word. That said having provided a specific test it’s possible for that particular capability to have already been added to the UI layer via a pause token or something even if the model itself responds at the same rate.
- mewpmewp2 3y ago
  If you used GPT as a brain you could provide the current time as an input for it, but otherwise yeah, GPT doesn't have time within it's input by default, but if you did: Made the instruction to be: Say "Stop!" when 10 seconds are done. And then had a loop where You give it current time in a loop, it would surely be able to do that. But I'm not sure if or how that is exactly related to consciousness? Time, the best way to think of it is probably a series of ticks anyway, and I assume in people's brains there is something to do that, so it would be reasonable to add a mechanism for GPT as well. GPTs goal is to only act as the intelligent part of the brain based on the input.

PeterisP 3y ago

Analysis of what-if scenarios and simulating likely expected outcomes of proposed actions is a very useful ability, but IMHO that is not part of what we'd call 'consciousness'.

Retric 3y ago
It’s a large fraction of what consciousness is actually doing, not the definition of consciousness. Someone considering what to say is simulating tradeoffs. In practice simulation isn’t limited to future possibilities. Trying to remember something you might recall vague sensory information like the tilt of your hips and try and create a scenario where that fits.
- PeterisP 3y ago
  I would not automatically treat all the things that people do consciously as intrinsically linked with consciousness, unless there's some clear reasoning why consciousness is required to do that particular thing - and not in the sense that it's required because unconscious people can't do stuff because they're incapacitated. Like, saying that simulating tradeoffs is what consciousness is actually doing seems to me equivalent to saying that smashing my fist on the table is something 'my consciousness' is doing, which is kind of technically true but not really relevant, as IMHO a non-human entity that's clearly without any consciousness could both smash something against the table and also simulate tradeoffs for various options, having a subjective experience isn't required for that.
  - Retric 3y ago
    Sleepwalking shows most things people do can happen without consciousness. People can make breakfast without being conscious etc. Choosing between tradeoffs is however intrinsically linked to consciousness.