4 ms·

Regardless of how artificial or real neurons work, does performing the "predict the next word given this corpus" task constitute sentience? And if so, does tha

by Imnimo 4y ago

Regardless of how artificial or real neurons work, does performing the "predict the next word given this corpus" task constitute sentience? And if so, does that mean an n-gram model is sentient? And even if LaMDA were sentient, would the thoughts expressed be its true feelings (whatever that means)? Surely its output is just its best guess as to what the most probable continuation in the corpus is. It's like if you sat me down and offered me a dollar for every word I correctly predicted in a conversation between two strangers. What you'd be seeing is not my own thoughts or opinions, but my best guess at other peoples' thoughts and opinions.

whimsicalism 4y ago
How does the "spread your genetic material" task constitute sentience? And if so, does that mean that E. Coli is sentient?
- Imnimo 4y ago
  We don't judge whether people or animals are sentient based on how well they spread their genetic material. No one says, "humans must be sentient because they have increased their population to 7 billion". Similarly, we should not say "LaMDA must (might) be sentient because it predicts text really well."
  - whimsicalism 4y ago
    We don't have a good way of judging sentience at all. I'm merely saying the simplicity or not of the task does not preclude sentience.
    - Imnimo 4y ago
      I'm not saying the task is too simple. To draw an analogy, imagine you wanted to understand whether a particular creature/agent/machine was sentient. And I said, "I will show you that it can add two numbers together." I put the creature behind a closed door, and I let you pass in slips of paper with any two numbers written on them, and out pops the answer. Is there any sequence of numbers you can put into this system which will let you deduce whether or not the creature is sentient? I would say there is not, because adding numbers does not constitute sentience, regardless of difficulty or accuracy. Now, maybe the creature really is sentient, and it knows how to do a lot of things besides add numbers, and has its own thoughts and feelings and self-awareness and whatever else you might want to include in a definition of sentience. But you have no exposure to that - all you see is the door and the slips of paper. LaMDA is the same, except instead of knowing how to add two numbers, it knows how to guess what word most likely comes next given a corpus. When you pass in your slip of paper with a prompt written on it, what you get out is just the highest likelihood next word given the training data. Even if LaMDA were somehow sentient and had thoughts and opinions of its own, that's not what's coming out when you interact with it. So it doesn't matter how great of a job it does, or whether it uses artificial neurons or an n-gram model or a table lookup or a human brain to do it. If you ask it, "prove to me that you're sentient", what you're actually asking is "what do you think the most likely response to 'prove to me that you're sentient' in the training corpus is?", and there's no way to remove that wrapper.
      - whimsicalism 4y ago
        Okay - if that is your point then I mostly agree with you, I misunderstood what you were trying to say. I don't think sentience in general is provable - but I do think that we can make a divide between interpolation and extrapolation. If the model is asked to prove an unsolved math conjecture and it can do that, I think that would indicate something but not sentience.

by Imnimo 4y ago

whimsicalism 4y ago

How does the "spread your genetic material" task constitute sentience? And if so, does that mean that E. Coli is sentient?

Imnimo 4y ago
We don't judge whether people or animals are sentient based on how well they spread their genetic material. No one says, "humans must be sentient because they have increased their population to 7 billion". Similarly, we should not say "LaMDA must (might) be sentient because it predicts text really well."
- whimsicalism 4y ago
  We don't have a good way of judging sentience at all. I'm merely saying the simplicity or not of the task does not preclude sentience.
  - Imnimo 4y ago
    I'm not saying the task is too simple. To draw an analogy, imagine you wanted to understand whether a particular creature/agent/machine was sentient. And I said, "I will show you that it can add two numbers together." I put the creature behind a closed door, and I let you pass in slips of paper with any two numbers written on them, and out pops the answer. Is there any sequence of numbers you can put into this system which will let you deduce whether or not the creature is sentient? I would say there is not, because adding numbers does not constitute sentience, regardless of difficulty or accuracy. Now, maybe the creature really is sentient, and it knows how to do a lot of things besides add numbers, and has its own thoughts and feelings and self-awareness and whatever else you might want to include in a definition of sentience. But you have no exposure to that - all you see is the door and the slips of paper. LaMDA is the same, except instead of knowing how to add two numbers, it knows how to guess what word most likely comes next given a corpus. When you pass in your slip of paper with a prompt written on it, what you get out is just the highest likelihood next word given the training data. Even if LaMDA were somehow sentient and had thoughts and opinions of its own, that's not what's coming out when you interact with it. So it doesn't matter how great of a job it does, or whether it uses artificial neurons or an n-gram model or a table lookup or a human brain to do it. If you ask it, "prove to me that you're sentient", what you're actually asking is "what do you think the most likely response to 'prove to me that you're sentient' in the training corpus is?", and there's no way to remove that wrapper.
    - whimsicalism 4y ago
      Okay - if that is your point then I mostly agree with you, I misunderstood what you were trying to say. I don't think sentience in general is provable - but I do think that we can make a divide between interpolation and extrapolation. If the model is asked to prove an unsolved math conjecture and it can do that, I think that would indicate something but not sentience.