5 ms·

I mean, they'd be right that my intelligence is not as well distributed as theirs. But that's a specific feature of their intelligence, that's not general intel

I mean, they'd be right that my intelligence is not as well distributed as theirs. But that's a specific feature of their intelligence, that's not general intelligence. If an octopus can beat me at Sokoban, I don't care whether it thinks with its brain or its arms, I would like to hire it as a programmer for my startup. If an octopus tries its best to play Sokoban, beats the tutorial but can't figure out the levels with actual puzzles, I won't consider it to possess general intelligence, even if it kicks my ass at say visual perception cognitive tests. (Am I missing your point somehow? I think my answer is relevant but I'm not 100% sure)

dboreham 6y ago
John Searle Octopus Test
- SonOfLilit 6y ago
  I assume you're referring to this? https://www.aclweb.org/anthology/2020.acl-main.463.pdf https://www.aclweb.org/anthology/2020.acl-main.463.pdf Gotta say in my opinion it didn't age well for a paper published in 2020 (but pre GPT-3). In practice, we have an O that is pretty good at describing how to build weapons against bears despite having absolutely no idea what "weapon" or "bear" refers to. But also, I don't seem to understand how this relates to the discussion, so can you help me there?
  - quonn 6y ago
    It may not have an idea of a picture of a bear. But it certainly knows all kinds of words that would describe that bear. And it relates those to the bear and also to other things. So I would argue that there is some „idea“. It‘s just not a visual idea and there is not conscious reflection. As far as the first part is concerned, what about a blind person? I would argue they can have a very good idea, too.
    - SonOfLilit 6y ago
      In my understanding of the "Octopus Test" argument from the linked PDF, GPT-3 is exactly an octopus test, and has exactly the same information that the octopus would have about bears, so GPT-3 being able to perform this kind of task disproves the argument conclusively.
  - hackinthebochs 6y ago
    >Gotta say in my opinion it didn't age well for a paper published in 2020 (but pre GPT-3). Agreed. The glaring issue with their argument falls out of their own introduction. What is the "linguistic intent" of the sentence "When was Malala Yousafzai born"? Well, it falls right out of a natural language training corpus which will, with high likelihood, be followed up with her actual birthdate! So there is some non-trivial signal within the training corpus of linguistic intent. Capturing this intent is then just a computational problem. But the objective function of predicting the subsequent word has within the space of solutions the linguistic intent of the writers of the training corpus.
grawprog 6y ago
I might be wrong, but I think the point of the parent comment was more along the lines of, the things we find to be 'general intelligence' may not be as general as we believe. Something like your sokoban or 2048 example, which you view as a thing that any being with general intelligence should be able to solve, may not be such general intelligence thing and more of a 'human specific' thing and we're too wrapped up in our anthropomorphic view points to recognize it.
- SonOfLilit 6y ago
  "General Intelligence" is a well defined term and there's a lot of literature on why it makes sense. There are many human cognitive skills that seem to be very helpful for some specific tasks (e.g. noticing that someone is watching you, identifying the pitch of a strummed guitar string) but are not very useful in general, but others (e.g. analogous thinking) that we found to be relevant to a vast range of cognitive tasks. There are also tasks that require mostly the second kind of skill. An animal that can perform them well should be considered intelligent even if it doesn't have perfect pitch and can't notice that it's being watched. An animal that we can't get to perform them well may or may not be intelligent - perhaps we're not good at convincing it to try hard. Some relevant links: https://en.wikipedia.org/wiki/G_factor_(psychometrics) https://en.wikipedia.org/wiki/G_factor_(psychometrics) and also, I guess https://en.wikipedia.org/wiki/Artificial_general_intelligence https://en.wikipedia.org/wiki/Artificial_general_intelligenc...
  - mike00632 6y ago
    What if the animal can't hear or is disabled? I know I sure can't tell what different pitches are.
    - SonOfLilit 6y ago
      I think you're agreeing with me?
  - grawprog 6y ago
    That was kind of my point, all of that is through a human centric point of view. What if higher intelligence is something else? Look at say plants, they've been shown to be intelligent in various ways. https://academic.oup.com/aob/article/92/1/1/177536 https://academic.oup.com/aob/article/92/1/1/177536 Now, when you take an ecosystem like a forest, something interesting happens. The plants begin to communicate with eachother using electrical signals passed along the mycorrhizal networks of the fungi that live symbiotically with them. https://academic.oup.com/aobpla/article-abstract/doi/10.1093/aobpla/plv050/201398 https://academic.oup.com/aobpla/article-abstract/doi/10.1093... https://link.springer.com/chapter/10.1007/978-3-319-75596-0_10 https://link.springer.com/chapter/10.1007/978-3-319-75596-0_... So, these networks facilitate learning by individual plants within the network, they facilitate long term memory, complex coordinated long term behaviour. Now, a lot of these behaviours and things happen over spans of decades to hundreds if not thousands of years. That sounds very much like some level of higher intelligence, yet those plants will never complete a 2048 or sokoban puzzle.
mnky9800n 6y ago
I feel like you are hyper focused on the idea of general intelligence which is why you are not finding what you want. Most people studying intelligence don't buy into a general intelligence idea. It certainly isn't present in most people who perform transfer of knowledge tasks.
- SonOfLilit 6y ago
  I've never met a healthy adult human being for whom I would bet against their ability to solve level 2 of this [1] given some time and motivation. Have you? (I'm referring to the last sentence. I can accept that many animal intelligence researchers don't buy into the idea of a general intelligence) [1] https://www.youtube.com/watch?v=yix2AVVUwe0 https://www.youtube.com/watch?v=yix2AVVUwe0
jancsika 6y ago
> If an octopus can beat me at Sokoban, I don't care whether it thinks with its brain or its arms, I would like to hire it as a programmer for my startup. You'd need to know something-- indeed, a lot-- about the entity in question in order to design tests to measure its general intelligence in the first place. Otherwise you're going to default to anthropomorphize the entity in question and misunderstand what it is you are measuring. E.g., vastly overestimating the language skills of a gorilla, or perhaps believing that ML algorithms somehow provide a higher level of objectivity than the data they were trained on: https://www.wired.com/story/algorithms-shouldve-made-courts-more-fair-what-went-wrong/ https://www.wired.com/story/algorithms-shouldve-made-courts-...
- SonOfLilit 6y ago
  Sokoban puzzles are very close to math-complete (in the same sense as in Turing-complete or NP-complete). Ockham's Razor leads me to believe it's much more probable to find an alien (or animal or ML) entity good at math than an alien entity overfitted for complex Sokoban puzzles. Now, I may undermeasure (maybe they're very intelligent but suck at spacial reasoning) but it would be hard to overmeasure (if they can do this kind of math, I'd like to hire them, period).

dboreham 6y ago

John Searle Octopus Test

SonOfLilit 6y ago
I assume you're referring to this? https://www.aclweb.org/anthology/2020.acl-main.463.pdf https://www.aclweb.org/anthology/2020.acl-main.463.pdf Gotta say in my opinion it didn't age well for a paper published in 2020 (but pre GPT-3). In practice, we have an O that is pretty good at describing how to build weapons against bears despite having absolutely no idea what "weapon" or "bear" refers to. But also, I don't seem to understand how this relates to the discussion, so can you help me there?
- quonn 6y ago
  It may not have an idea of a picture of a bear. But it certainly knows all kinds of words that would describe that bear. And it relates those to the bear and also to other things. So I would argue that there is some „idea“. It‘s just not a visual idea and there is not conscious reflection. As far as the first part is concerned, what about a blind person? I would argue they can have a very good idea, too.
  - SonOfLilit 6y ago
    In my understanding of the "Octopus Test" argument from the linked PDF, GPT-3 is exactly an octopus test, and has exactly the same information that the octopus would have about bears, so GPT-3 being able to perform this kind of task disproves the argument conclusively.
- hackinthebochs 6y ago
  >Gotta say in my opinion it didn't age well for a paper published in 2020 (but pre GPT-3). Agreed. The glaring issue with their argument falls out of their own introduction. What is the "linguistic intent" of the sentence "When was Malala Yousafzai born"? Well, it falls right out of a natural language training corpus which will, with high likelihood, be followed up with her actual birthdate! So there is some non-trivial signal within the training corpus of linguistic intent. Capturing this intent is then just a computational problem. But the objective function of predicting the subsequent word has within the space of solutions the linguistic intent of the writers of the training corpus.

grawprog 6y ago

I might be wrong, but I think the point of the parent comment was more along the lines of, the things we find to be 'general intelligence' may not be as general as we believe. Something like your sokoban or 2048 example, which you view as a thing that any being with general intelligence should be able to solve, may not be such general intelligence thing and more of a 'human specific' thing and we're too wrapped up in our anthropomorphic view points to recognize it.

SonOfLilit 6y ago
"General Intelligence" is a well defined term and there's a lot of literature on why it makes sense. There are many human cognitive skills that seem to be very helpful for some specific tasks (e.g. noticing that someone is watching you, identifying the pitch of a strummed guitar string) but are not very useful in general, but others (e.g. analogous thinking) that we found to be relevant to a vast range of cognitive tasks. There are also tasks that require mostly the second kind of skill. An animal that can perform them well should be considered intelligent even if it doesn't have perfect pitch and can't notice that it's being watched. An animal that we can't get to perform them well may or may not be intelligent - perhaps we're not good at convincing it to try hard. Some relevant links: https://en.wikipedia.org/wiki/G_factor_(psychometrics) https://en.wikipedia.org/wiki/G_factor_(psychometrics) and also, I guess https://en.wikipedia.org/wiki/Artificial_general_intelligence https://en.wikipedia.org/wiki/Artificial_general_intelligenc...
- mike00632 6y ago
  What if the animal can't hear or is disabled? I know I sure can't tell what different pitches are.
  - SonOfLilit 6y ago
    I think you're agreeing with me?
- grawprog 6y ago
  That was kind of my point, all of that is through a human centric point of view. What if higher intelligence is something else? Look at say plants, they've been shown to be intelligent in various ways. https://academic.oup.com/aob/article/92/1/1/177536 https://academic.oup.com/aob/article/92/1/1/177536 Now, when you take an ecosystem like a forest, something interesting happens. The plants begin to communicate with eachother using electrical signals passed along the mycorrhizal networks of the fungi that live symbiotically with them. https://academic.oup.com/aobpla/article-abstract/doi/10.1093/aobpla/plv050/201398 https://academic.oup.com/aobpla/article-abstract/doi/10.1093... https://link.springer.com/chapter/10.1007/978-3-319-75596-0_10 https://link.springer.com/chapter/10.1007/978-3-319-75596-0_... So, these networks facilitate learning by individual plants within the network, they facilitate long term memory, complex coordinated long term behaviour. Now, a lot of these behaviours and things happen over spans of decades to hundreds if not thousands of years. That sounds very much like some level of higher intelligence, yet those plants will never complete a 2048 or sokoban puzzle.

mnky9800n 6y ago

I feel like you are hyper focused on the idea of general intelligence which is why you are not finding what you want. Most people studying intelligence don't buy into a general intelligence idea. It certainly isn't present in most people who perform transfer of knowledge tasks.

SonOfLilit 6y ago
I've never met a healthy adult human being for whom I would bet against their ability to solve level 2 of this [1] given some time and motivation. Have you? (I'm referring to the last sentence. I can accept that many animal intelligence researchers don't buy into the idea of a general intelligence) [1] https://www.youtube.com/watch?v=yix2AVVUwe0 https://www.youtube.com/watch?v=yix2AVVUwe0

jancsika 6y ago

> If an octopus can beat me at Sokoban, I don't care whether it thinks with its brain or its arms, I would like to hire it as a programmer for my startup. You'd need to know something-- indeed, a lot-- about the entity in question in order to design tests to measure its general intelligence in the first place. Otherwise you're going to default to anthropomorphize the entity in question and misunderstand what it is you are measuring. E.g., vastly overestimating the language skills of a gorilla, or perhaps believing that ML algorithms somehow provide a higher level of objectivity than the data they were trained on: https://www.wired.com/story/algorithms-shouldve-made-courts-more-fair-what-went-wrong/ https://www.wired.com/story/algorithms-shouldve-made-courts-...

SonOfLilit 6y ago
Sokoban puzzles are very close to math-complete (in the same sense as in Turing-complete or NP-complete). Ockham's Razor leads me to believe it's much more probable to find an alien (or animal or ML) entity good at math than an alien entity overfitted for complex Sokoban puzzles. Now, I may undermeasure (maybe they're very intelligent but suck at spacial reasoning) but it would be hard to overmeasure (if they can do this kind of math, I'd like to hire them, period).