If you dump a prehistoric person at the age of let's say 50 into modern society they would be considered stupid and incapable of most things and would dump any IQ test we could give them because they wouldn't understand the concepts. We would think of them as less intelligent.
If you take the same persons new born child and could fast forward into today and let them grow up here they would be having no problem growing up in society and being plotted onto that distribution curve somewhere.
The distribution of intelligence based on one set off principle like IQ, in a society have nothing to do with the ability understand or explain. Most IQ tests have to do with language and the lack of vocabulary. That some people are better at it than others doesn't mean anything. There are things very smart people don't understand either.
In short, would you say that Deutsch’s theory is just too abstract? Ironically he might be too smart for his own good if intelligence is defined in humans as the ability to abstract and manipulate the abstractions. In other words, yes, if given unlimited time and resources, a stupid person could discover relativity or invent AI or whatever, but on a practical, real-world level, they can’t.
Since stupid people are also universal explainers, the only meaningful difference between them and smart people is the speed and memory of the computer they are running on.
I think this is wrong. The difference is the ideas the people understand, not hardware.
I think Deutsch’s concept of intelligence is very different from the concept of our society’s which measures intelligence in terms of IQ. For Deutsch intelligence is the ability of creating new explanations not just mere pattern recognition, which is what IQ test is, your 10th IQ test will be much higher than your 1st IQ test, why?? because your brain is now somewhat familiar with the pattern.
I think this view on intelligence is what makes Deutsch believe that our current AI technology is far away from being general intelligence because all it does is recognize patterns and not create any new explanations.
I loved this article and the provocative theory but certain aspects quoted have been already proved wrong by many studies in socio economics psychology and ethology.
I mention a few that anyone can search:
IQ is well known to be a terrible measure of intelligence
Income and ethnicity s a cofounder of "intelligence" if you are poor you can be street smart but unable to read or explain a theorem
Prodigies: people like Einstein are prodigies with exceptional abilities but if they were born in the wrong conditions we wouldn't be heard of them (so many examples of that).
Animales can beat people yes like the chimpanze number tapping simply because their brain has LESS cognitive baseline so they can be good at one task but cannot be average good to any other tasks.
Language and Intelligence: those are not causally related, both genetics and experiments have proven that.
Dwarkesh asks why do we care if AI meets your definition of intelligence if it can do a huge suite of economically valuable tasks? Wouldn’t it be a great outcome if this were the result?
This is a bet that we can separate capability from the capacity to learn end goals or purposes. This is why there is so much focus on g as an underlying valid “intelligence” trait that could be maxed - it’s a capability that is agnostic; it generalises regardless of the purpose.
I think it would be a great outcome and the bet is probably correct. But i don’t quite agree that it’s a distinction that doesn’t matter - an AI that can generate a trillion dollars but can’t ask why it should probably will have some limitations. From a civilisational perspective if you totally handover authority to massively capable AIs that cannot modify their reward function I expect that to lead to a society that cannot evolve new purposes once the ones we inherited from our ancestral environment are substantially solved
I agree. I too was quite puzzled by this passage in the book: “Indeed, it is tantamount to an appeal to the supernatural, with all the arbitrariness that is inherent in such appeals, for if we wanted to incorporate into our world view an imaginary realm explicable only to superhumans, we need never have bothered to abandon the myths of Persephone and her fellow deities.”
It should be obvious that very few humans are capable of understanding quantum mechanics and the like, and while that inability may to a great extent have been in the past (and can indeed be at present) the result of inauspicious cultures or environments, it is still the case that, without exceptional raw cognitive ability, no environment would by itself be sufficient to further our understanding of the universe. And it doesn’t at all follow, as Deutsche claims, that if we adopted the view (correctly) that the intricacies of the laws of physics are explicable only to “superhumans”, we should have just stuck to the myths and deities of old.
AI can have infinite memory and instantiate best-fit logical associations across a seemingly infinite number of narrow areas. They don’t technically need to be more “intelligent” than humans (smart will do) and I actually wonder if this is uneven acceptance between what is smart and what is intelligence is where the confusion lays, particularly with the unintelligent.
I think it comes down to memory and speed. Most of the standardized tests are timed. Probably, I can solve more questions correctly in the GRE exam if given more time. But, practically speaking speed is important in the world we live in. We have to solve problems fast at the workplace and are evaluated BOTH on - speed and accuracy. I think David Deutsch is saying that theoretically everyone could get a GRE score of 340/340 given enough practice, effort & memory, speed augmentation. So, until we have memory and speed augmentation, I think there is going to be a visible practical difference between humans.
If you dump a prehistoric person at the age of let's say 50 into modern society they would be considered stupid and incapable of most things and would dump any IQ test we could give them because they wouldn't understand the concepts. We would think of them as less intelligent.
If you take the same persons new born child and could fast forward into today and let them grow up here they would be having no problem growing up in society and being plotted onto that distribution curve somewhere.
The distribution of intelligence based on one set off principle like IQ, in a society have nothing to do with the ability understand or explain. Most IQ tests have to do with language and the lack of vocabulary. That some people are better at it than others doesn't mean anything. There are things very smart people don't understand either.
In short, would you say that Deutsch’s theory is just too abstract? Ironically he might be too smart for his own good if intelligence is defined in humans as the ability to abstract and manipulate the abstractions. In other words, yes, if given unlimited time and resources, a stupid person could discover relativity or invent AI or whatever, but on a practical, real-world level, they can’t.
Since stupid people are also universal explainers, the only meaningful difference between them and smart people is the speed and memory of the computer they are running on.
I think this is wrong. The difference is the ideas the people understand, not hardware.
Misses the entire point Deutsch makes about culture…
I think Deutsch’s concept of intelligence is very different from the concept of our society’s which measures intelligence in terms of IQ. For Deutsch intelligence is the ability of creating new explanations not just mere pattern recognition, which is what IQ test is, your 10th IQ test will be much higher than your 1st IQ test, why?? because your brain is now somewhat familiar with the pattern.
I think this view on intelligence is what makes Deutsch believe that our current AI technology is far away from being general intelligence because all it does is recognize patterns and not create any new explanations.
I loved this article and the provocative theory but certain aspects quoted have been already proved wrong by many studies in socio economics psychology and ethology.
I mention a few that anyone can search:
IQ is well known to be a terrible measure of intelligence
Income and ethnicity s a cofounder of "intelligence" if you are poor you can be street smart but unable to read or explain a theorem
Prodigies: people like Einstein are prodigies with exceptional abilities but if they were born in the wrong conditions we wouldn't be heard of them (so many examples of that).
Animales can beat people yes like the chimpanze number tapping simply because their brain has LESS cognitive baseline so they can be good at one task but cannot be average good to any other tasks.
Language and Intelligence: those are not causally related, both genetics and experiments have proven that.
Dwarkesh asks why do we care if AI meets your definition of intelligence if it can do a huge suite of economically valuable tasks? Wouldn’t it be a great outcome if this were the result?
This is a bet that we can separate capability from the capacity to learn end goals or purposes. This is why there is so much focus on g as an underlying valid “intelligence” trait that could be maxed - it’s a capability that is agnostic; it generalises regardless of the purpose.
I think it would be a great outcome and the bet is probably correct. But i don’t quite agree that it’s a distinction that doesn’t matter - an AI that can generate a trillion dollars but can’t ask why it should probably will have some limitations. From a civilisational perspective if you totally handover authority to massively capable AIs that cannot modify their reward function I expect that to lead to a society that cannot evolve new purposes once the ones we inherited from our ancestral environment are substantially solved
I agree. I too was quite puzzled by this passage in the book: “Indeed, it is tantamount to an appeal to the supernatural, with all the arbitrariness that is inherent in such appeals, for if we wanted to incorporate into our world view an imaginary realm explicable only to superhumans, we need never have bothered to abandon the myths of Persephone and her fellow deities.”
It should be obvious that very few humans are capable of understanding quantum mechanics and the like, and while that inability may to a great extent have been in the past (and can indeed be at present) the result of inauspicious cultures or environments, it is still the case that, without exceptional raw cognitive ability, no environment would by itself be sufficient to further our understanding of the universe. And it doesn’t at all follow, as Deutsche claims, that if we adopted the view (correctly) that the intricacies of the laws of physics are explicable only to “superhumans”, we should have just stuck to the myths and deities of old.
AI can have infinite memory and instantiate best-fit logical associations across a seemingly infinite number of narrow areas. They don’t technically need to be more “intelligent” than humans (smart will do) and I actually wonder if this is uneven acceptance between what is smart and what is intelligence is where the confusion lays, particularly with the unintelligent.
I think it comes down to memory and speed. Most of the standardized tests are timed. Probably, I can solve more questions correctly in the GRE exam if given more time. But, practically speaking speed is important in the world we live in. We have to solve problems fast at the workplace and are evaluated BOTH on - speed and accuracy. I think David Deutsch is saying that theoretically everyone could get a GRE score of 340/340 given enough practice, effort & memory, speed augmentation. So, until we have memory and speed augmentation, I think there is going to be a visible practical difference between humans.