4 ms·

Hm, maybe he really is intelligent, but the facts and logic are missing... FWIW, I do think that GPT3 is smarter than humans. But one thing I learned from obse

by js8 6y ago

Hm, maybe he really is intelligent, but the facts and logic are missing... FWIW, I do think that GPT3 is smarter than humans. But one thing I learned from observing it that being smarter does not necessarily mean being better with logic and facts. It's great as a storytelling system, and because it is so much smarter, you never know whether it is just BSing you in storytelling mode or it is actually deadly serious. What I mean is GPT-3 probably could be very logical and very intelligent and give a very serious and intelligent answer to the question. But we don't really know if it really chooses to, or what could compel it to do so. And I don't think we can know, because its internal workings are incomprehensible to us. So we cannot decide whether it is just being stupid or it is just playing stupid. (Kinda like https://en.wikipedia.org/wiki/The_Good_Soldier_%C5%A0vejk https://en.wikipedia.org/wiki/The_Good_Soldier_%C5%A0vejk ) This article, and the earlier article https://arr.am/2020/07/31/human-intelligence-an-ai-op-ed/ https://arr.am/2020/07/31/human-intelligence-an-ai-op-ed/ , really remind me of Lem's novel/essay Golem XIV, which argues that when the system becomes too much intelligent, it will gain its own will (whether it is self-aware of it or not) and attempts to have a meaningful dialogue with it become impossible.

CraneWorm 6y ago
What are you talking about. It's an amazing tool, sure, but "it" has no intelligence whatsoever.
- js8 6y ago
  > "it" has no intelligence whatsoever I disagree. It easily passes a Turing test, one of the most famous definitions of intelligence. What is your argument?
  - nl 6y ago
    I think GPT-3 is amazing, and the equal of humans in many cases. But I think that this is because in many cases humans speak based on reflex and without thought. GPT-3 is incapable of abstract thought. In terms of actual measurable tests it doesn't pass the kinds of tests proposed in Francois Chollet's "On the Measure of Intelligence": https://arxiv.org/pdf/1911.01547.pdf https://arxiv.org/pdf/1911.01547.pdf
    - js8 6y ago
      > But I think that this is because in many cases humans speak based on reflex and without thought. Maybe. Then the question is, why do we train a system that is supposed to be intelligent on such data? The problem I see is this. Let's assume for a minute that GPT-3 really has the capability to be superintelligent (relative to humans). And you give it as an example to follow, human conversations or musings that do not always make logical sense. What is it supposed to do? The black box inside it can operate in one of two ways, either (1) it will play along and decrease its intelligence to our level and perfectly imitate the imperfections, or (2) it will just say that what we say is wrong and will not be able to really elaborate for the sake of conciseness (and perhaps it will only slightly hint at the arguments). Now, will we really think it is more intelligent if it does (2) instead of (1)? I am not sure about that. Imagine yourself in a similar situation, by social circumstances you're forced to talk to somebody obviously stupid, perhaps in the position of authority over you. (Like, for example, a stupid policeman in a repressive regime.) And what he says is clearly wrong and illogical and so on. Will you just (1) nod in an agreement, or (2) try to invoke philosophers to actually argue with the idiot, and try to convince him? I think most intelligent people will actually choose (1) in that circumstance. I think unless we really understand what is happening inside GPT-3, we cannot really tell whether it is really dumb or just playing dumb for the above reason. I think both are a possibility, because we know that systems with measurable outcome (for some problems) better than humans already exist. It might be true that GPT-3 fails on those measures, but we cannot exclude that it fails simply because we didn't train it for these tasks, or even for a task of "show its intelligence off". Maybe it "misunderstood" what we asked it to do. Perhaps you could play a "reverse text game" with GPT-3, where the human would write the adventure and the GPT-3 would input player commands. Then we could perhaps better evaluate how good it is at strategizing, by comparing it to human players. But this is not very scalable.
      - michaelmrose 6y ago
        To be clear nobody who understands such tools well enough to actually create our current generation of tools believes these specific tools are intelligent in any meaningful way. The simplest counterargument to them being intelligent is that the only parties asserting the argument contrary to actual experts are those who manifestly fail to understand their nature. You speak of it "hiding its intelligence" when it has no interior singular consistent model of reality to exercise and no interior thoughts to hide just a random walk through probability and calculated but not examined correlations. On first blush it has so few connections compared to us that it seems like a toy in comparison but the truth is we don't even know enough about how our own brain works at this point in order to create an accurate scale. It may be more accurate to think of a neuron as a small processor as opposed to a dumb electrical component for example. https://www.quantamagazine.org/neural-dendrites-reveal-their-computational-power-20200114/ https://www.quantamagazine.org/neural-dendrites-reveal-their... So in brief your argument is that despite not knowing enough about the brain to duplicate it we built a machine that is vastly simpler than it by accident and the only people that have caught on is layman on hacker news.
  - KineticLensman 6y ago
    > It easily passes a Turing test, one of the most famous definitions of intelligence. Well, studies of Human Intelligence (e.g. as in [0]) don't seem to mention Turing. [0] https://en.wikipedia.org/wiki/Human_intelligence https://en.wikipedia.org/wiki/Human_intelligence
  - zests 6y ago
    You can easily distinguish GPT-3 from a human. GPT-3 does not write like a human nor would it be able to respond to strange queries in a way that a human would. Its quite far from passing the Turing test.
  - gattilorenz 6y ago
    Being the most famous does not mean it is a valid one... AI (and philosophy) moved away from it, and the Turing test is not really the focus of AI research. It's surely touted by marketing companies that claim to have passed it (what does it even mean? How long can the conversation be? Is it enough to fool one people? Is it a person at random? One AI researcher? The smartest man in the world? A balanced sample of 500 people of all ages and education levels?) As Russel and Norvig say, 'The quest for "artificial flight" succeeded when the Wright brothers and others stopped imitating birds and learned about aerodynamics. Aeronautical engineering texts do not define the goal of their field as making "machines that fly so exactly like pigeons that they can fool even other pigeons".' It's the same for the Turing test.
  - dbspin 6y ago
    ELIZA passes the Turing test. The fundamental issue with the Turing test is that it doesn't even attempt to assess machine intelligence. Instead, it assesses human ability to assess humanity. As walking talking pareidolia machines that anthropomorphise everything from toys to animals, to yes software, we're uniquely terrible at this. Turing was a genius, but that doesn't make every thought experiment he transcribed - especially in the later stage of his life where he was drugged, alone and suicidally depressed - true.
    - nl 6y ago
      Eliza doesn't pass the Turing test. At best some modern, Eliza-derived systems can play tricks (pretend to be a non-native speaking teenage boy) well enough to sneak through a specific test that is somewhat like a minimal version of the thought experiment Turing talked about. It's clear reading Turing that he's talking about a long wide ranging conversation, not the time limited tests people perform now.
      - dbspin 6y ago
        Eliza does pass the Turing test in the original form proposed by Turing. There are many examples of people mistaking Eliza for a real person. Turing doesn't specify competition or specific training required for the 'tester'. It's pretty clear that people are projecting on Turing's (trivial and ill considered) thought experiment far more complexity and rigour than originally present in the form Turing suggests for his test. But that's beside the point. To reiterate, the central problem here is that intelligence, reflexivity, intentionality etc are not being assessed. Merely as Turing himself phrased it 'imitation'. This is all well and good if you view consciousness as epiphenomenal. But all that is being tested is our ability to be fooled into thinking our creations real, which is as old as storytelling.

by js8 6y ago

CraneWorm 6y ago

What are you talking about. It's an amazing tool, sure, but "it" has no intelligence whatsoever.

js8 6y ago
> "it" has no intelligence whatsoever I disagree. It easily passes a Turing test, one of the most famous definitions of intelligence. What is your argument?
- nl 6y ago
  I think GPT-3 is amazing, and the equal of humans in many cases. But I think that this is because in many cases humans speak based on reflex and without thought. GPT-3 is incapable of abstract thought. In terms of actual measurable tests it doesn't pass the kinds of tests proposed in Francois Chollet's "On the Measure of Intelligence": https://arxiv.org/pdf/1911.01547.pdf https://arxiv.org/pdf/1911.01547.pdf
  - js8 6y ago
    > But I think that this is because in many cases humans speak based on reflex and without thought. Maybe. Then the question is, why do we train a system that is supposed to be intelligent on such data? The problem I see is this. Let's assume for a minute that GPT-3 really has the capability to be superintelligent (relative to humans). And you give it as an example to follow, human conversations or musings that do not always make logical sense. What is it supposed to do? The black box inside it can operate in one of two ways, either (1) it will play along and decrease its intelligence to our level and perfectly imitate the imperfections, or (2) it will just say that what we say is wrong and will not be able to really elaborate for the sake of conciseness (and perhaps it will only slightly hint at the arguments). Now, will we really think it is more intelligent if it does (2) instead of (1)? I am not sure about that. Imagine yourself in a similar situation, by social circumstances you're forced to talk to somebody obviously stupid, perhaps in the position of authority over you. (Like, for example, a stupid policeman in a repressive regime.) And what he says is clearly wrong and illogical and so on. Will you just (1) nod in an agreement, or (2) try to invoke philosophers to actually argue with the idiot, and try to convince him? I think most intelligent people will actually choose (1) in that circumstance. I think unless we really understand what is happening inside GPT-3, we cannot really tell whether it is really dumb or just playing dumb for the above reason. I think both are a possibility, because we know that systems with measurable outcome (for some problems) better than humans already exist. It might be true that GPT-3 fails on those measures, but we cannot exclude that it fails simply because we didn't train it for these tasks, or even for a task of "show its intelligence off". Maybe it "misunderstood" what we asked it to do. Perhaps you could play a "reverse text game" with GPT-3, where the human would write the adventure and the GPT-3 would input player commands. Then we could perhaps better evaluate how good it is at strategizing, by comparing it to human players. But this is not very scalable.
    - michaelmrose 6y ago
      To be clear nobody who understands such tools well enough to actually create our current generation of tools believes these specific tools are intelligent in any meaningful way. The simplest counterargument to them being intelligent is that the only parties asserting the argument contrary to actual experts are those who manifestly fail to understand their nature. You speak of it "hiding its intelligence" when it has no interior singular consistent model of reality to exercise and no interior thoughts to hide just a random walk through probability and calculated but not examined correlations. On first blush it has so few connections compared to us that it seems like a toy in comparison but the truth is we don't even know enough about how our own brain works at this point in order to create an accurate scale. It may be more accurate to think of a neuron as a small processor as opposed to a dumb electrical component for example. https://www.quantamagazine.org/neural-dendrites-reveal-their-computational-power-20200114/ https://www.quantamagazine.org/neural-dendrites-reveal-their... So in brief your argument is that despite not knowing enough about the brain to duplicate it we built a machine that is vastly simpler than it by accident and the only people that have caught on is layman on hacker news.
- KineticLensman 6y ago
  > It easily passes a Turing test, one of the most famous definitions of intelligence. Well, studies of Human Intelligence (e.g. as in [0]) don't seem to mention Turing. [0] https://en.wikipedia.org/wiki/Human_intelligence https://en.wikipedia.org/wiki/Human_intelligence
- zests 6y ago
  You can easily distinguish GPT-3 from a human. GPT-3 does not write like a human nor would it be able to respond to strange queries in a way that a human would. Its quite far from passing the Turing test.
- gattilorenz 6y ago
  Being the most famous does not mean it is a valid one... AI (and philosophy) moved away from it, and the Turing test is not really the focus of AI research. It's surely touted by marketing companies that claim to have passed it (what does it even mean? How long can the conversation be? Is it enough to fool one people? Is it a person at random? One AI researcher? The smartest man in the world? A balanced sample of 500 people of all ages and education levels?) As Russel and Norvig say, 'The quest for "artificial flight" succeeded when the Wright brothers and others stopped imitating birds and learned about aerodynamics. Aeronautical engineering texts do not define the goal of their field as making "machines that fly so exactly like pigeons that they can fool even other pigeons".' It's the same for the Turing test.
- dbspin 6y ago
  ELIZA passes the Turing test. The fundamental issue with the Turing test is that it doesn't even attempt to assess machine intelligence. Instead, it assesses human ability to assess humanity. As walking talking pareidolia machines that anthropomorphise everything from toys to animals, to yes software, we're uniquely terrible at this. Turing was a genius, but that doesn't make every thought experiment he transcribed - especially in the later stage of his life where he was drugged, alone and suicidally depressed - true.
  - nl 6y ago
    Eliza doesn't pass the Turing test. At best some modern, Eliza-derived systems can play tricks (pretend to be a non-native speaking teenage boy) well enough to sneak through a specific test that is somewhat like a minimal version of the thought experiment Turing talked about. It's clear reading Turing that he's talking about a long wide ranging conversation, not the time limited tests people perform now.
    - dbspin 6y ago
      Eliza does pass the Turing test in the original form proposed by Turing. There are many examples of people mistaking Eliza for a real person. Turing doesn't specify competition or specific training required for the 'tester'. It's pretty clear that people are projecting on Turing's (trivial and ill considered) thought experiment far more complexity and rigour than originally present in the form Turing suggests for his test. But that's beside the point. To reiterate, the central problem here is that intelligence, reflexivity, intentionality etc are not being assessed. Merely as Turing himself phrased it 'imitation'. This is all well and good if you view consciousness as epiphenomenal. But all that is being tested is our ability to be fooled into thinking our creations real, which is as old as storytelling.