Prestigious black national merit finalist G-man (and others) wanted me to comment on Nassim’s Taleb’s recent tirade against IQ.

It’s amazing how much publicity this issue is getting in the HBD community. Taleb is famous for attacking mathematical models and for attacking “intellectuals”, so it was only a matter of time before he attacked IQ.

But because a lot of Taleb’s fans are high IQ people who love numbers, and thus love IQ, they may be shocked that their hero doesn’t share their enthusiasm.

Taleb writes:

“IQ” is a stale test meant to measure mental capacity but in fact mostly measures extreme unintelligence (learning difficulties), as well as, to a lesser extent, a form of intelligence, stripped of 2nd order effects. It is via negativa not via positiva. Designed for learning disabilities, it ends up selecting exam-takers, paper shufflers, obedient IYIs (intellectuals yet idiots), ill adapted for “real life”. The concept is poorly thought out mathematically (a severe flaw in correlation under fat tails, fails to properly deal with dimensionality, treats the mind as an instrument not a complex system)

Well he’s right that intelligence tests were originally created to identify the learning disabled, not the brilliant, and that the concept was not mathematically well thought out (IQ was originally measured in mental age units, ignoring the fact that the developmental trajectory is not linear, and that newborn babies do not have zero intelligence). However neither of these problems strike me as especially important, especially since IQ testers admit their tests are ordinal scales (or at best interval scales) and not absolute scales. Meanwhile the ceiling on IQ scales have since been extended to better measure giftedness.

But Taleb seems to think the flawed mathematical model of IQ tests is a severe problem under fat tails. I wish he would elaborate on this point. Modern IQ tests are forced to fit a bell curve and when the variable being correlated with IQ does not have a perfectly Gaussian distribution (i.e. income), one can either normalize the distribution as I did here, or one can take the logarithm of income.

But maybe I’m missing the point. Taleb is famous for writing about black swan events:

A black swan is an event or occurrence that deviates beyond what is normally expected of a situation and is extremely difficult to predict. Black swan events are typically random and unexpected.

So maybe Taleb is talking about all the times IQ predictions go wildly wrong, like when when Richard Feynman scores “only” 125 on an IQ test, despite being one of the greatest physicists of all time. Perhaps he’s saying these events are so unpredictable they can’t be foreseen from the correlation between IQ and achievement using the bivariate normal distribution and thus render IQ tests useless in some of the most important cases.

IQ enthusiasts will counter that Feynman’s IQ can easily be explained by the fact that the test may not have measured math ability, however this supports Taleb’s point that test’s fail to “to properly deal with dimensionality”.

Taleb also has a point when he says tests treat “the mind as an instrument not a complex system” if he means what I think he means. Even the best IQ tests like the WAIS measure cognitive abilities in isolation and your final score is just a composite of all your strengths and weaknesses. But in real life, intelligence is a dynamic system that maximizes strengths and minimizes weaknesses. On an IQ test, you’ll lose points for sucking at spatial ability no matter how brilliant you are verbally, but in real life, your spatial disability is irrelevant if you’re smart enough to avoid spatial situations.

Don’t get me wrong; I’m still a huge IQ enthusiast, but IQ testing has room for improvement.

Advertisements