> I know at least octopi can use computer screens, and I know they enjoy hard puzzles that reward them with food, so it should be easy to teach them some computer puzzle game?
I'm not sure it's that easy.
For example, suppose a race of hyperintelligent octopuses beam down and decide to study your intelligence. They notice in one test that if you begin scrambling eggs for breakfast and they cut off both your arms at the shoulder, your arms aren't able to finish scrambling the eggs.
"My, the intelligence in this species isn't very well-distributed," they and their tentacles agree.
> "My, the intelligence in this species isn't very well-distributed," they and their tentacles agree.
And then maybe they conspire to orchestrate your firing from that cushy corporate office job. They note that you then go home and start working on your career skills, in hopes of joining another company.
"My, the intelligence of this creature is quite distributed," the tentacles and central brain agree, of the corporation.
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)
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.
"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.
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.
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.
Those seem like great reasons to focus our efforts on things that are very close to us in their mode of being first, and expand from there as we learn more examples. Plants and fungi being broadly sessile means that their cognition, if they have it, will be expressed in radically different ways from ours. Whereas birds or dolphins are a lot like us: brains and bodies.
If plants were highly generally intelligent (as I suspect corvids are) and if I were able to convince them that they would get a prize they highly desire if they solve a difficult Sokoban level with controls they can use (which has been done many times with corvids) then I would definitely expect them to solve the level given a few decades or centuries, yes.
(I personally don't believe plants to possess general intelligence, but I would be delighted to discover I'm wrong)
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.
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)
> 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:
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).
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?
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.
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.
>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.
I'm not sure it's that easy.
For example, suppose a race of hyperintelligent octopuses beam down and decide to study your intelligence. They notice in one test that if you begin scrambling eggs for breakfast and they cut off both your arms at the shoulder, your arms aren't able to finish scrambling the eggs.
"My, the intelligence in this species isn't very well-distributed," they and their tentacles agree.