3 ms·

You can also get it to agree that it is not conscious. I can write a one line program that claims to be conscious right now. Lamda and GTP-3 are amazing achieve

by simonh 4y ago

You can also get it to agree that it is not conscious. I can write a one line program that claims to be conscious right now. Lamda and GTP-3 are amazing achievements, but they're still just chat bots that look good when they're being lobbed softball questions. If you ask something that maps to the input texts they're trained on you get credible looking responses, but critical or adversarial questioning very easily exposes how shallow these things are.

eru 4y ago
Turing's original paper is worth a read here. He already pre-empted a lot of the objections back then. You are right that Turing's original test is much harder than just passing a few shallow softball questions. See https://academic.oup.com/mind/article/LIX/236/433/986238 https://academic.oup.com/mind/article/LIX/236/433/986238 for the original paper. However, I do count responding well to softball questions as important markers of our progress. Ten years ago, no one was able to pass even this mark. (And from a commercial point of view, good enough 'understanding' and responses might be good enough for many tasks. For example, I am a native German speaker, but I found that Google Translate often already does a better job of translating English to German than I could do, unless I spent unreasonable amounts of time on the task. At the moment the most efficient way for me to produce a translation is to let Google Translate do the work, and then proof-read to spot cases were the machine didn't have enough context to resolve an ambiguity the right way.)
- janekm 4y ago
  You might like to try out GPT-3 (seems access is widely available now) for translation some time, it's remarkably good, better than myself (well my German is pretty rusty) and I think much better than Google Translate... (Basically just write a prompt like "translate the following to German/Chinese/Whatever:")
vbezhenar 4y ago
The same reasoning could be applied to many people. You wouldn't deny their consciousness on that ground. The issue with general AI is like issue with chess. People are not happy when robot can win some games. People can only accept that robot really plays chess, when robot defeated best of the best. General AI is the same. Looking like some guy is not enough. You need to reason better than smartest human on the planet, only then people will admit that general AI is there. That's a high bar. You can definitely talk someone to agree that he's not conscious using some logic tricks. Not everyone, for sure, but someone who may be doesn't really care that much about that concept. You also can tune AI so it'll not agree that he's unconscious no matter what. What I'm trying to deliver is that there's no easy border which can divide conscious from unconscious. The only border was Turing test and robots passed it. Now you can test whether that particular robot talks like a very smart person, etc. But you can't distinguish bot from some random not-so-smart person.
- simonh 4y ago
  This line of reasoning reduces to the philosophical zombie argument. It's true I can't prove to you that I have a first person experience of the world, but it doesn't therefore follow that anything that makes that claim does have such an experience. For myself I'm satisfied that, when questioned thoroughly, these models fail so completely and utterly and their output is so ludicrously nonsensical that I don't see how their behaviour can be consistent with consciousness, as I understand it. It's not just that the output is incoherent, but it's the ways that they fail too. Do I completely understand consciousness? No, but that doesn't mean these things are conscious either.
  - akimball 4y ago
    People frequently output ludicrous nonsense. Does this disqualify their consciousness?
    - simonh 4y ago
      No it doesn't. So what？We have this exact problem to deal with when people suffer brain damage. To what extent are they conscious, or still people? Yet we do come to medical conclusions on that question. There's always a degree of uncertainty, but you still have to make the call based on the information you have. The information I have indicates to me with a very high probability, in my estimation, that these things are not conscious.
akimball 4y ago
I can also get Daniel Dennett to agree that he is not conscious.
hsn915 4y ago
> You can also get it to agree that it is not conscious. Careful about how you word this. Only a conscious being can agree that is not conscious. I think you mean you can get the program to output text that proclaims its lack of consciousness.

by simonh 4y ago

eru 4y ago

Turing's original paper is worth a read here. He already pre-empted a lot of the objections back then. You are right that Turing's original test is much harder than just passing a few shallow softball questions. See https://academic.oup.com/mind/article/LIX/236/433/986238 https://academic.oup.com/mind/article/LIX/236/433/986238 for the original paper. However, I do count responding well to softball questions as important markers of our progress. Ten years ago, no one was able to pass even this mark. (And from a commercial point of view, good enough 'understanding' and responses might be good enough for many tasks. For example, I am a native German speaker, but I found that Google Translate often already does a better job of translating English to German than I could do, unless I spent unreasonable amounts of time on the task. At the moment the most efficient way for me to produce a translation is to let Google Translate do the work, and then proof-read to spot cases were the machine didn't have enough context to resolve an ambiguity the right way.)

janekm 4y ago
You might like to try out GPT-3 (seems access is widely available now) for translation some time, it's remarkably good, better than myself (well my German is pretty rusty) and I think much better than Google Translate... (Basically just write a prompt like "translate the following to German/Chinese/Whatever:")

vbezhenar 4y ago

The same reasoning could be applied to many people. You wouldn't deny their consciousness on that ground. The issue with general AI is like issue with chess. People are not happy when robot can win some games. People can only accept that robot really plays chess, when robot defeated best of the best. General AI is the same. Looking like some guy is not enough. You need to reason better than smartest human on the planet, only then people will admit that general AI is there. That's a high bar. You can definitely talk someone to agree that he's not conscious using some logic tricks. Not everyone, for sure, but someone who may be doesn't really care that much about that concept. You also can tune AI so it'll not agree that he's unconscious no matter what. What I'm trying to deliver is that there's no easy border which can divide conscious from unconscious. The only border was Turing test and robots passed it. Now you can test whether that particular robot talks like a very smart person, etc. But you can't distinguish bot from some random not-so-smart person.

simonh 4y ago
This line of reasoning reduces to the philosophical zombie argument. It's true I can't prove to you that I have a first person experience of the world, but it doesn't therefore follow that anything that makes that claim does have such an experience. For myself I'm satisfied that, when questioned thoroughly, these models fail so completely and utterly and their output is so ludicrously nonsensical that I don't see how their behaviour can be consistent with consciousness, as I understand it. It's not just that the output is incoherent, but it's the ways that they fail too. Do I completely understand consciousness? No, but that doesn't mean these things are conscious either.
- akimball 4y ago
  People frequently output ludicrous nonsense. Does this disqualify their consciousness?
  - simonh 4y ago
    No it doesn't. So what？We have this exact problem to deal with when people suffer brain damage. To what extent are they conscious, or still people? Yet we do come to medical conclusions on that question. There's always a degree of uncertainty, but you still have to make the call based on the information you have. The information I have indicates to me with a very high probability, in my estimation, that these things are not conscious.

akimball 4y ago

I can also get Daniel Dennett to agree that he is not conscious.

hsn915 4y ago

> You can also get it to agree that it is not conscious. Careful about how you word this. Only a conscious being can agree that is not conscious. I think you mean you can get the program to output text that proclaims its lack of consciousness.