7 ms·

What kind of evidence would you need to convince yourself of an AI being sentient?

by mckirk 4y ago

rnosov 4y ago
Pass a Turing test with a competent judge. ChatGPT or any other bot won't be able to pass it.
- pixl97 4y ago
  So if a human fails a Turing test, we can remove their rights of personhood? I'm just checking your logic to see if a two way communicative process, or if humans are granted magic values here?
  - rnosov 4y ago
    For example, great apes look somewhat like humans but can't pass a Turing test. Should we give them same rights as humans too? It works other way too.
- HDThoreaun 4y ago
  Bing chat would do better at passing a Turing test than many real people I suspect.
- PeterisP 4y ago
  I'm strongly convinced that a Turing test can be passed by a non-sentient entity. Being sentient and smart is sufficient to fool a judge, but not necessary, people can be fooled by stupid illusions.
- redox99 4y ago
  I'm pretty sure if let's say OpenAI worked 1 year on passing the Turing test, they would achieve it. It seems extremely close. Current LLMs are trained to be useful, not to trick humans.
j0hnyl 4y ago
Sentience is the ability to be able to experience feelings. In this regard, chatbots are not really convincing. Kind of like a psychopath can describe a feeling, but not actually feel it. This is a tough question to answer though, but there's this kind of human intuition we have where most of the time we somehow know if someone is feigning emotion. Emotion that comes from chat bots can be explained with, "well it's just a chat bot". Something needs to happen in order for us to truly question that in order to squash our doubts about their sentience. Not sure what that will look like though.
- dougmwne 4y ago
  This is the point I was making. We humans do have this emotional sensor we can use to probe the minds of others. For some people, this emotional detector started beeping when they talked to Bing. Is the sensor faulty? Maybe. But as it hits for more and more people that becomes it’s own problem.
  - PeterisP 4y ago
    > We humans do have this emotional sensor we can use to probe the minds of others. Wait, what? I certainly don't have one. Most of us (not all! neurotypical experience isn't universal) have a somewhat effective skill, continuously trained since childhood, to guess at the emotions of others based on various visual and behavioral cues, but we also know that these visible cues can be faked by (for example) a skilled actor, as the "sensor" has zero insight to the actual emotional experience. If I make the assumption that someone else (e.g. you) function the same as I do, then I can reason that "your behavior X implies your emotional experience Y, because it does for me"; however, if we can't make that assumption, or (as is the case for non-human minds) we know that the process is substantially different - then there is literally zero basis for any trustworthy reasoning whatsoever about that internal emotional state.
  - weard_beard 4y ago
    :) I just typed a colon and a parenthesis. Your brain interpreted it as a smile on a human face. You are equally mistaken in the use of your emotional sensor. Our brain fills in a LOT of gaps based on OUR training.
dekhn 4y ago
Nobody has established a reasonable answer to this question; we can't even demonstrate self-sentience or sentience of other humans.
weard_beard 4y ago
Step 1) The ability to distinguish real from unreal and the ability to understand truth as, partially or primarily, direct sensory input which is retained and can be acted upon.
matt_heimer 4y ago
Lock 2 AIs that haven't been programmed with language in a room, given time do they develop the ability to communicate and express ideas, desires, needs to each other? There have been several instances of language evolving again, deaf people creating sign language, etc. A sentient intelligence should be driven to understand concepts and convey them, probably over multiple mediums. That's why the turning test is a good start, communication is a key aspect of sentience but its just round one.
- fzzzy 4y ago
  This already happened. https://www.huffpost.com/entry/facebook-shuts-down-ai-robot-after-it-creates-its-own-language_n_61087608e4b0999d2084f6bf https://www.huffpost.com/entry/facebook-shuts-down-ai-robot-...
  - crote 4y ago
    No, in the linked example it is mutating English into a different language. GP's challenge was to create a form of communication unprompted, and without precedence.
jsimzeroone 4y ago
One precondition for sentience would be some kind of continuity. Does the AI operate separately from its tasks, or does it instantiate, take care of some request, and then end, as a ChatBot process does? Is it actively processing while waiting for input, or is it only alive and active when it is processing some input and preparing some output? I think the lack of this continuity is why people who feel they know how ChatGPT works feel confident it is not sentient. Sentience isn't simply responding to stimuli -- although we don't really understand consciousness, it clearly involves some kind of continuing sense of self, some kind of "dial tone" that provides the basic sense of continuing existence, and that isn't there in a bot that simply takes input and builds a response based on rules and data.

What kind of evidence would you need to convince yourself of an AI being sentient?

by mckirk 4y ago

What kind of evidence would you need to convince yourself of an AI being sentient?

rnosov 4y ago

Pass a Turing test with a competent judge. ChatGPT or any other bot won't be able to pass it.

pixl97 4y ago
So if a human fails a Turing test, we can remove their rights of personhood? I'm just checking your logic to see if a two way communicative process, or if humans are granted magic values here?
- rnosov 4y ago
  For example, great apes look somewhat like humans but can't pass a Turing test. Should we give them same rights as humans too? It works other way too.
HDThoreaun 4y ago
Bing chat would do better at passing a Turing test than many real people I suspect.
PeterisP 4y ago
I'm strongly convinced that a Turing test can be passed by a non-sentient entity. Being sentient and smart is sufficient to fool a judge, but not necessary, people can be fooled by stupid illusions.
redox99 4y ago
I'm pretty sure if let's say OpenAI worked 1 year on passing the Turing test, they would achieve it. It seems extremely close. Current LLMs are trained to be useful, not to trick humans.

j0hnyl 4y ago

Sentience is the ability to be able to experience feelings. In this regard, chatbots are not really convincing. Kind of like a psychopath can describe a feeling, but not actually feel it. This is a tough question to answer though, but there's this kind of human intuition we have where most of the time we somehow know if someone is feigning emotion. Emotion that comes from chat bots can be explained with, "well it's just a chat bot". Something needs to happen in order for us to truly question that in order to squash our doubts about their sentience. Not sure what that will look like though.

dougmwne 4y ago
This is the point I was making. We humans do have this emotional sensor we can use to probe the minds of others. For some people, this emotional detector started beeping when they talked to Bing. Is the sensor faulty? Maybe. But as it hits for more and more people that becomes it’s own problem.
- PeterisP 4y ago
  > We humans do have this emotional sensor we can use to probe the minds of others. Wait, what? I certainly don't have one. Most of us (not all! neurotypical experience isn't universal) have a somewhat effective skill, continuously trained since childhood, to guess at the emotions of others based on various visual and behavioral cues, but we also know that these visible cues can be faked by (for example) a skilled actor, as the "sensor" has zero insight to the actual emotional experience. If I make the assumption that someone else (e.g. you) function the same as I do, then I can reason that "your behavior X implies your emotional experience Y, because it does for me"; however, if we can't make that assumption, or (as is the case for non-human minds) we know that the process is substantially different - then there is literally zero basis for any trustworthy reasoning whatsoever about that internal emotional state.
- weard_beard 4y ago
  :) I just typed a colon and a parenthesis. Your brain interpreted it as a smile on a human face. You are equally mistaken in the use of your emotional sensor. Our brain fills in a LOT of gaps based on OUR training.

dekhn 4y ago

Nobody has established a reasonable answer to this question; we can't even demonstrate self-sentience or sentience of other humans.

weard_beard 4y ago

Step 1) The ability to distinguish real from unreal and the ability to understand truth as, partially or primarily, direct sensory input which is retained and can be acted upon.

matt_heimer 4y ago

Lock 2 AIs that haven't been programmed with language in a room, given time do they develop the ability to communicate and express ideas, desires, needs to each other? There have been several instances of language evolving again, deaf people creating sign language, etc. A sentient intelligence should be driven to understand concepts and convey them, probably over multiple mediums. That's why the turning test is a good start, communication is a key aspect of sentience but its just round one.

fzzzy 4y ago
This already happened. https://www.huffpost.com/entry/facebook-shuts-down-ai-robot-after-it-creates-its-own-language_n_61087608e4b0999d2084f6bf https://www.huffpost.com/entry/facebook-shuts-down-ai-robot-...
- crote 4y ago
  No, in the linked example it is mutating English into a different language. GP's challenge was to create a form of communication unprompted, and without precedence.

jsimzeroone 4y ago

One precondition for sentience would be some kind of continuity. Does the AI operate separately from its tasks, or does it instantiate, take care of some request, and then end, as a ChatBot process does? Is it actively processing while waiting for input, or is it only alive and active when it is processing some input and preparing some output? I think the lack of this continuity is why people who feel they know how ChatGPT works feel confident it is not sentient. Sentience isn't simply responding to stimuli -- although we don't really understand consciousness, it clearly involves some kind of continuing sense of self, some kind of "dial tone" that provides the basic sense of continuing existence, and that isn't there in a bot that simply takes input and builds a response based on rules and data.