3 ms·

Defining the term is hard. So hard in fact, that no one has managed to do it in millennia (it's not as if this question is new). We can, however, use proxies t

by qayxc 4y ago

Defining the term is hard. So hard in fact, that no one has managed to do it in millennia (it's not as if this question is new). We can, however, use proxies to get around the problem. Just like "life", "species", or "time" - all of which are surprisingly poorly defined and might not even exist in strictest sense - there are ways to work with terms without having a waterproof definition for them. No matter the set of criteria you choose, you will always find some that match the subject in question (here: a chatbot). The question is, at which point you draw the line and whether that necessarily arbitrary line has any relevance (see the discussion amongst biologists whether viruses are alive).

rendall 4y ago
> Defining the term is hard. So hard in fact, that no one has managed to do it in millennia (it's not as if this question is new). Indeed! It's almost like any statement about sentience pro or con is unfalsifiable, to be accepted or rejected according to faith. Each of us two, ourselves, could insist that the other is not (or is) sentient, but neither of us could objectively prove it. We humans all just have to decide that we must act as if everyone else must feel internally like we do, even though we all could just be automatons, reacting deterministically to stimuli. In the absence of a formal definition, any statement about "has sentience" is not only unfalsifiable, it's absurd. OP maybe should have titled his article "The question of LaMDA's sentience is meaningless" or maybe "No, LaMDA does not feel internally like we humans do".
- qayxc 4y ago
  > In the absence of a formal definition, any statement about "has sentience" is not only unfalsifiable, it's absurd. I have to disagree with you there. While it's indeed a question of faith (i.e. unscientific) to give a definite positive answer - or rather that a positive answer depends on a an arbitrarily chosen set of criteria - the opposite is simply not true. More formally put: making a positive claim would require a sufficient condition that has to be met. I agree that such condition/criterion doesn't exists (yet). Required conditions, however, do exist. Sentience in any scientific sense requires the ability to actively perceive- and interact with the environment, as well as a physical mechanism for processing and storing information. So we can definitely rule out rocks being sentient, for example. On the other hand, depending on the set of criteria you chose, you may or may not classify a sophisticated chatbot as sentient. It'd be sufficient to present such set of testable criteria and apply them to the system under test. Whether you or I accept these criteria as being sufficient, is another question entirely and indeed not a scientific one.
  - rendall 4y ago
    > Sentience in any scientific sense... I'm still not seeing a definition there though. Let's arbitrarily try: "sentience is here defined as having sensations, self awareness and sense of place in the universe". Good? Feel free to change it. I like your necessary versus sufficient phrasing, though. Let's take a look. Can we agree to remove "in any scientific sense" from there? It lends it an air of authority I'm not sure is justified. Feel free to push back, though. So: a required condition for sentience is "the ability to actively perceive- and interact with the environment, as well as a physical mechanism for processing and storing information" Hmm. "Actively perceive" "And interact with the environment" and "stores information". Couldn't this be a just-so story to include that which we feel to be most like us and exclude that which is least like us? Could we then say "the degree to which something is sentient is the degree to which it matches a human"? I think I can imagine a patient who contacts a horrific virus that cuts off all external and bodily sensations. She is still fully cognizant but now feels herself to be bodiless, floating in nothingness. From the outside, she seems to be asleep or paralyzed. She is kept alive by caretakers and her own autonomous processes but she can perceive none of it. Is she sentient? Id say yes, but she nevertheless lacks some of the required conditions.
    - int_19h 4y ago
      I believe that you're spot-on here. We don't really know what sentience is and cannot define it, so we resort to one of the two approaches: inclusionary, where anything that behaves as we'd expect a sentient entity to behave is considered sentient; and exclusionary, which is basically coming up with excuses as to why things that behave as sentient really aren't, because their behavior is implemented in ways that are "fundamentally" different from ours.
    - qayxc 4y ago
      > Can we agree to remove "in any scientific sense" from there? No we can't, that's the entire point. As soon as you remove that, you remove any testable, repeatable observation, which is the whole point. > I think I can imagine a patient who contacts a horrific virus that cuts off all external and bodily sensations. Think harder. There's a reason some patients are declared "brain dead". There are observable processes associated with human sentience, e.g. brain waves and body functions. If one fails (e.g. the heart stops and the vital signs collapse), they can sometimes be restored externally. If both fail and cannot be restored, the patient is declared dead and thus no longer sentient in the scientific (i.e. observable and testable) sense. Everything else is religion. It's really that simple. edit: as far machines go, different criteria have to be selected, because obviously language models don't have a pulse or emit delta waves. edit2: also forgot to mention that the perception and interaction proposal applies to the system (or organism) in principle - accidents or illnesses don't count; a system that's by design incapable of interacting with and perceiving its environment cannot be said to be sentient (again, because I cannot stress this enough: in an observable/testable way).

by qayxc 4y ago

rendall 4y ago

> Defining the term is hard. So hard in fact, that no one has managed to do it in millennia (it's not as if this question is new). Indeed! It's almost like any statement about sentience pro or con is unfalsifiable, to be accepted or rejected according to faith. Each of us two, ourselves, could insist that the other is not (or is) sentient, but neither of us could objectively prove it. We humans all just have to decide that we must act as if everyone else must feel internally like we do, even though we all could just be automatons, reacting deterministically to stimuli. In the absence of a formal definition, any statement about "has sentience" is not only unfalsifiable, it's absurd. OP maybe should have titled his article "The question of LaMDA's sentience is meaningless" or maybe "No, LaMDA does not feel internally like we humans do".

qayxc 4y ago
> In the absence of a formal definition, any statement about "has sentience" is not only unfalsifiable, it's absurd. I have to disagree with you there. While it's indeed a question of faith (i.e. unscientific) to give a definite positive answer - or rather that a positive answer depends on a an arbitrarily chosen set of criteria - the opposite is simply not true. More formally put: making a positive claim would require a sufficient condition that has to be met. I agree that such condition/criterion doesn't exists (yet). Required conditions, however, do exist. Sentience in any scientific sense requires the ability to actively perceive- and interact with the environment, as well as a physical mechanism for processing and storing information. So we can definitely rule out rocks being sentient, for example. On the other hand, depending on the set of criteria you chose, you may or may not classify a sophisticated chatbot as sentient. It'd be sufficient to present such set of testable criteria and apply them to the system under test. Whether you or I accept these criteria as being sufficient, is another question entirely and indeed not a scientific one.
- rendall 4y ago
  > Sentience in any scientific sense... I'm still not seeing a definition there though. Let's arbitrarily try: "sentience is here defined as having sensations, self awareness and sense of place in the universe". Good? Feel free to change it. I like your necessary versus sufficient phrasing, though. Let's take a look. Can we agree to remove "in any scientific sense" from there? It lends it an air of authority I'm not sure is justified. Feel free to push back, though. So: a required condition for sentience is "the ability to actively perceive- and interact with the environment, as well as a physical mechanism for processing and storing information" Hmm. "Actively perceive" "And interact with the environment" and "stores information". Couldn't this be a just-so story to include that which we feel to be most like us and exclude that which is least like us? Could we then say "the degree to which something is sentient is the degree to which it matches a human"? I think I can imagine a patient who contacts a horrific virus that cuts off all external and bodily sensations. She is still fully cognizant but now feels herself to be bodiless, floating in nothingness. From the outside, she seems to be asleep or paralyzed. She is kept alive by caretakers and her own autonomous processes but she can perceive none of it. Is she sentient? Id say yes, but she nevertheless lacks some of the required conditions.
  - int_19h 4y ago
    I believe that you're spot-on here. We don't really know what sentience is and cannot define it, so we resort to one of the two approaches: inclusionary, where anything that behaves as we'd expect a sentient entity to behave is considered sentient; and exclusionary, which is basically coming up with excuses as to why things that behave as sentient really aren't, because their behavior is implemented in ways that are "fundamentally" different from ours.
  - qayxc 4y ago
    > Can we agree to remove "in any scientific sense" from there? No we can't, that's the entire point. As soon as you remove that, you remove any testable, repeatable observation, which is the whole point. > I think I can imagine a patient who contacts a horrific virus that cuts off all external and bodily sensations. Think harder. There's a reason some patients are declared "brain dead". There are observable processes associated with human sentience, e.g. brain waves and body functions. If one fails (e.g. the heart stops and the vital signs collapse), they can sometimes be restored externally. If both fail and cannot be restored, the patient is declared dead and thus no longer sentient in the scientific (i.e. observable and testable) sense. Everything else is religion. It's really that simple. edit: as far machines go, different criteria have to be selected, because obviously language models don't have a pulse or emit delta waves. edit2: also forgot to mention that the perception and interaction proposal applies to the system (or organism) in principle - accidents or illnesses don't count; a system that's by design incapable of interacting with and perceiving its environment cannot be said to be sentient (again, because I cannot stress this enough: in an observable/testable way).