I disagree with your interpretation of the SLODR; I think that happens because IQ tests test relatively high-level tasks for stupid people, and relatively low-level tasks for smart people.
When you solve a complicated mental problem, you have to do many different steps of thinking building on top of each other. The higher these build up, …
I disagree with your interpretation of the SLODR; I think that happens because IQ tests test relatively high-level tasks for stupid people, and relatively low-level tasks for smart people.
When you solve a complicated mental problem, you have to do many different steps of thinking building on top of each other. The higher these build up, the more they will use the low-level steps in similar proportion. This is why, for example, there arent CPUs which trade off speed at AND for speed at NOT: Our tasks for the computer are so high-level that they all use AND and NOT in similar ratios, and so theres just an optimal way to make a CPU which everyone does. And this stacking to higher levels is also how you get to more advanced cognitive functions.
So if you test someone on a task thats high-level relative to their ability, theyll have to stack up their "simple steps", and so their performance will average out their aptitude at those simple steps. When you then give them another high-level task, itll average in a slightly different but similar way, giving highly correlated performance.
But a more intelligent person may be able to learn more complex tasks natively, so that theyre "simple steps" for them. If you then test them on this simple-for-them task, theres not necessarily a lot of correlation with other simple-for-them tasks.
So when you have a test at a fixed level of difficulty, there will be more correlation among the less intelligent. This explains the observations without any real diminishing returns. It also explains why correlation among different abilities is higher within humans than across animal species, even though the humans are more intelligent (so opposite to what generalising Spearman would get): animals often dont stack but have specialised hardcoded skills in specific areas.
Thanks for the comment. If I understand correctly - and please correct me if I don't - you basically subscribe to the sampling model of g, namely that even alleged tests of one ability actually involve multiple stacked abilities and that's why they correlate. SLODR is thus an artefact of more intelligent people relying on fewer of these abilities in their stacks. My first question is, if this model is correct, how do you even know there are more generally intelligent people at all? The second question is does this make any difference to the argument that spatial ability should be ignored when calculating Asheknazi IQ?
As regards comparison with animals, it's obvious that many animals have incredibly high-functioning abilities without being intelligent at all in a sense we would recognise so maybe they are just too different to draw inferences from. I would find it more persuasive if it could be demonstrated correlation is lower among chimpanzees.
>My first question is, if this model is correct, how do you even know there are more generally intelligent people at all?
Imagine intelligence was simply processing speed of the brain, and theres two ways in which brains get faster: Nerve cells can get smaller or react to signals more quickly. It makes no difference to cognition how you got your speed, and cell size and reaction time dont correlate. In that scenario, does general intelligence exist or is there only cell size and reaction time? I think clearly the former.
In the same way, I dont think the stacking model means general intelligence doesnt exist. Now, if the tests involved multiple abilities in a more linear-combination-y way, then you could make the case against g. In that case, you could have isolated ability tests that are a similar kind of thing as the tests you actually have - the constructors of the test just failed to do it that way. But in the stacking model, the "simple steps" are a very different sort of thing from the compound abilities youre testing. The stacking is inherent to advanced cognition, and theres no way to separate out the basic abilities without reducing the difficulty.
Even in the linear combination scenario though, if cognitive tasks youd realistically do in the real world are combined in a similar way to those in the test, the practical utility of IQ testing is mostly the same, even if IQ is less "real".
>does this make any difference to the argument that spatial ability should be ignored when calculating Asheknazi IQ?
Your argument is that general intelligence comes apart at the higher end, and so you want to focus in on the part that closest tracks the intuitive concept of intelligence. This model implies that general intelligence doesnt come apart.
>I would find it more persuasive if it could be demonstrated correlation is lower among chimpanzees.
That depends on what you mean exactly. Id put my prediction as: Those abilities where humans are most ahead of chimps, are also the ones that most correlate with each other across people.
Is there a body of research for the stacking model, or is it your idea?
"This model implies that general intelligence doesnt come apart."
OK, so what do you do practically with a group like Ashkenazim where, on an empirical level, it does seem to be coming apart? Why are they natively learning certain tasks, but still doing stacking for other tasks that are of equivalent difficulty?
>OK, so what do you do practically with a group like Ashkenazim where, on an empirical level, it does seem to be coming apart?
Your model suggests that theres a general trend for things to come further and further apart the higher intelligence you look at, and that I definitely disagree with. Obviously there can be some differences in how strongly things correlate within different groups, because some skills may correlate for reasons other then the stacking-averaging.
What I would do with the Ashkenazim is look at how they do on problems that are on their upper limit. This is difficult because we dont have a way to do tests that are independent of prior knowledge up there, but if you care about knowing yourself rather than proving things very legibly, you can still form an opinion based on observation.
My impression is that spatial abilities continue to be helpful for high-level science, and that successfull jewish scientists werent bad at them. There are some fields of mathematics where the Grothendiecks of this world get by without, but thats the exception. So you shouldnt ignore spatial scores when judging the intelligence of smart people, though maybe you should weigh it a bit less (or more?); measuring by eye cant rule that out.
On second thought, I guess I should explain a little bit why I believe this. I was thinking about why there is a general intelligence, and found that straightforward theories generally implied that theres a simple way by which anyone can do any task. For example, if intelligence is processing speed, then anyone would be able to solve any problem if only he has long enough to think. Thats obviously not true. If it was working memory, youd only need pen and paper, if it was reliability/precision, you could attempt the problem many times and give the average answer etc.
This is because if intelligence was some simple *thing*, then you could improve someones effective intelligence arbitrarily without doing any cognitive work for him (i.e. a "one weird trick"). The way to avoid this is that cognitive abilities depend on other cognitive abilities - that way, theres no guaranteed way to improve them without doing cognitive work (e.g. giving him a calculator).
That already restricts you to something in the general area of the stacking model, and the details are just based on what I think complex tasks are like in general.
I disagree with your interpretation of the SLODR; I think that happens because IQ tests test relatively high-level tasks for stupid people, and relatively low-level tasks for smart people.
When you solve a complicated mental problem, you have to do many different steps of thinking building on top of each other. The higher these build up, the more they will use the low-level steps in similar proportion. This is why, for example, there arent CPUs which trade off speed at AND for speed at NOT: Our tasks for the computer are so high-level that they all use AND and NOT in similar ratios, and so theres just an optimal way to make a CPU which everyone does. And this stacking to higher levels is also how you get to more advanced cognitive functions.
So if you test someone on a task thats high-level relative to their ability, theyll have to stack up their "simple steps", and so their performance will average out their aptitude at those simple steps. When you then give them another high-level task, itll average in a slightly different but similar way, giving highly correlated performance.
But a more intelligent person may be able to learn more complex tasks natively, so that theyre "simple steps" for them. If you then test them on this simple-for-them task, theres not necessarily a lot of correlation with other simple-for-them tasks.
So when you have a test at a fixed level of difficulty, there will be more correlation among the less intelligent. This explains the observations without any real diminishing returns. It also explains why correlation among different abilities is higher within humans than across animal species, even though the humans are more intelligent (so opposite to what generalising Spearman would get): animals often dont stack but have specialised hardcoded skills in specific areas.
Thanks for the comment. If I understand correctly - and please correct me if I don't - you basically subscribe to the sampling model of g, namely that even alleged tests of one ability actually involve multiple stacked abilities and that's why they correlate. SLODR is thus an artefact of more intelligent people relying on fewer of these abilities in their stacks. My first question is, if this model is correct, how do you even know there are more generally intelligent people at all? The second question is does this make any difference to the argument that spatial ability should be ignored when calculating Asheknazi IQ?
As regards comparison with animals, it's obvious that many animals have incredibly high-functioning abilities without being intelligent at all in a sense we would recognise so maybe they are just too different to draw inferences from. I would find it more persuasive if it could be demonstrated correlation is lower among chimpanzees.
>My first question is, if this model is correct, how do you even know there are more generally intelligent people at all?
Imagine intelligence was simply processing speed of the brain, and theres two ways in which brains get faster: Nerve cells can get smaller or react to signals more quickly. It makes no difference to cognition how you got your speed, and cell size and reaction time dont correlate. In that scenario, does general intelligence exist or is there only cell size and reaction time? I think clearly the former.
In the same way, I dont think the stacking model means general intelligence doesnt exist. Now, if the tests involved multiple abilities in a more linear-combination-y way, then you could make the case against g. In that case, you could have isolated ability tests that are a similar kind of thing as the tests you actually have - the constructors of the test just failed to do it that way. But in the stacking model, the "simple steps" are a very different sort of thing from the compound abilities youre testing. The stacking is inherent to advanced cognition, and theres no way to separate out the basic abilities without reducing the difficulty.
Even in the linear combination scenario though, if cognitive tasks youd realistically do in the real world are combined in a similar way to those in the test, the practical utility of IQ testing is mostly the same, even if IQ is less "real".
>does this make any difference to the argument that spatial ability should be ignored when calculating Asheknazi IQ?
Your argument is that general intelligence comes apart at the higher end, and so you want to focus in on the part that closest tracks the intuitive concept of intelligence. This model implies that general intelligence doesnt come apart.
>I would find it more persuasive if it could be demonstrated correlation is lower among chimpanzees.
That depends on what you mean exactly. Id put my prediction as: Those abilities where humans are most ahead of chimps, are also the ones that most correlate with each other across people.
Is there a body of research for the stacking model, or is it your idea?
"This model implies that general intelligence doesnt come apart."
OK, so what do you do practically with a group like Ashkenazim where, on an empirical level, it does seem to be coming apart? Why are they natively learning certain tasks, but still doing stacking for other tasks that are of equivalent difficulty?
Im not aware of any research on this.
>OK, so what do you do practically with a group like Ashkenazim where, on an empirical level, it does seem to be coming apart?
Your model suggests that theres a general trend for things to come further and further apart the higher intelligence you look at, and that I definitely disagree with. Obviously there can be some differences in how strongly things correlate within different groups, because some skills may correlate for reasons other then the stacking-averaging.
What I would do with the Ashkenazim is look at how they do on problems that are on their upper limit. This is difficult because we dont have a way to do tests that are independent of prior knowledge up there, but if you care about knowing yourself rather than proving things very legibly, you can still form an opinion based on observation.
My impression is that spatial abilities continue to be helpful for high-level science, and that successfull jewish scientists werent bad at them. There are some fields of mathematics where the Grothendiecks of this world get by without, but thats the exception. So you shouldnt ignore spatial scores when judging the intelligence of smart people, though maybe you should weigh it a bit less (or more?); measuring by eye cant rule that out.
OK thanks. This is a perspective I haven't heard before and I appreciate you sharing it.
On second thought, I guess I should explain a little bit why I believe this. I was thinking about why there is a general intelligence, and found that straightforward theories generally implied that theres a simple way by which anyone can do any task. For example, if intelligence is processing speed, then anyone would be able to solve any problem if only he has long enough to think. Thats obviously not true. If it was working memory, youd only need pen and paper, if it was reliability/precision, you could attempt the problem many times and give the average answer etc.
This is because if intelligence was some simple *thing*, then you could improve someones effective intelligence arbitrarily without doing any cognitive work for him (i.e. a "one weird trick"). The way to avoid this is that cognitive abilities depend on other cognitive abilities - that way, theres no guaranteed way to improve them without doing cognitive work (e.g. giving him a calculator).
That already restricts you to something in the general area of the stacking model, and the details are just based on what I think complex tasks are like in general.
You should write an article. If you don't want your own substack, maybe DLA could guest post it.
Thanks. Ive been told to make one a few times, and I will when I have a bit more free time.