The IQ around you
https://twitter.com/interpretantion had a tweet recently that got me thinking:
My guess is IQ is actually a test of how many layers of recursion a person can handle in their raw working memory. Pretty key component of abstract thinking, engineering, and even management, but definitely not going to tell you everything.
Nassim Taleb has an article called “IQ is largely a psuedoscientific swindle” which is extremely chaotic and hard to cite as a whole, but has one point I find myself reaching for often. Suppose you have an X value of “score on some intelligence test” and a Y value of “general success in life”. Being deficient at this test indicates major troubles with life and perfectly correlates with your general success - if you can improve your score “one point” you’ll also improve your general success “one point”. But in the positive values, there’s absolutely no correlation whatsoever: improvement your score from zero to positive infinity doesn’t do anything to your general success in life.
This pattern of perfect correlation on the negative half and no correlation on the positive half leads to an 85% overall correlation between X and Y. (There’s a certain intuition that it might be 50%, which isn’t true.) But what’s really interesting is what happens when you add random noise to the test scores. The correlation drops, as you’d expect — but it also visually seems to migrate to “both sides”. By the end, you end up with a .46 correlation (still obviously too high to be pure chance) but visually just a loose cloud that doesn’t show the one-sided nature of the correlation:
IQ is an extremely noisy measure where peoples scores on multiple tests can often differ substantially. Asking whether it “correlates to life success” isn’t a very sensible question, since it might correlate very strongly on the low levels while being basically meaningless on the high ones. The idea of IQ-as-levels-of-recursion seems to fit nicely in to this. If you mostly need to see things literally spelled out, without the flexibility of parsing multiple levels of rules, there’s a lot of concepts that you flat out won’t be able to understand. But as the levels of recursion increase, it becomes more important to error-check all of the rules you think you’re following against the world than to be able to perfectly articulate the result of following the rules.
But the thing that really interests me is comparing this idea of depth of mental recursion in relation to environmental cognition. Very skilled practitioners of an intellectual field are often very particular about how they arrange their notes, or their desk, or whatever. They configuration the space around them in such a way to make it easier to store information and retrieve it later.This seems to be more evidence towards IQ being highly meaningful only on the low end. It matters a lot how much you can “load into” your working memory from the space around you, but you don’t gain a lot by trying to keep increasingly large edifices of thought straight, since you can just make your environment keep the score and focus on doing things correctly one local step at a time, like in a math problem or a grocery list.
I can’t say I’m very interested in measuring the efficacy of IQ in general — I think that people who fixated on the results of standardized tests tend to be dull and unimaginative. But I’ve got a suspicion that this mental model of humans as computers reading from the hard drive of the world around them has a lot of potential for explaining what we’re actually doing when we think.