[sitkack]

> IQ also correlates with research success and research is not test.

We should define research success and test. Research is definitely a fitness function that is intertwined with lots of dependent and independent variables no?

Most of the Taleb argument is that you can't compress a high dimensional space down to one dimension is it not?

[sanxiyn]

Yes, but dimensionality reduction is a thing. IQ is just PCA. Intelligence is multidimensional, but it happens to have one large principal component. It could have been otherwise, but it is just not so, here on Earth with Homo sapiens.

[sitkack]

Dimensionality reduction isn't reversible function. Video and one channel of audio is at least 10 dimensions, arguably IQ is more, you are going to project that to a single scalar?

The sample size, the experimental design and control are all out of whack in these papers, I was trying to PCA Taleb's argument.

I don't think you are measuring what you think you are measuring.

[sitkack]

There was a now deleted insightful comment, by another author which I will paste below and then reply to.

----

No. That would be a pat objection. He is saying the value of IQ as a measure of intelligence-explained performance for anything that isn’t isomorphic to an IQ test is low. The reason IQ looks like a good measure or produces any correlation with anything outside of itself is because it is predictive on the low-side of the distribution. When you include the low-side data you get correlations on the high-side because of the correlation on the low-side. However, its predictive power on the high-side is mostly noise (a stick in the mud gets its support from the mud, not the wind). He demonstrates this by clipping the low-side data and comparing to salary (which he argues is a ‘real-world’ set not isomorphic to an IQ test).

In other words, a low IQ (very low) suggests lack of intelligence but, above some fuzzy point, a higher IQ does not suggest more intelligence (and could even be anti-correlated).

He is not saying some people aren’t smarter than other people or that dimension reduction is always bogus. He is saying that IQ is a bogus way to measure intelligence because intelligence is supposed to explain performance and when the clip the low tail it simply does not.

----

reply inline since I am out of post karma.

Thanks for that explanation.

Let me paraphrase from down here, to put it bluntly, the supposedly predictive value of this continuous variable (IQ) should really be replaced with a high, medium and low or a binary value. That the predictive power of IQ is much more coarse grained, and it doesn't do what it is commonly used for.

Slight defense, I wasn't trying to say dimensionality reduction was bogus either, nor was I advocating for pat objection. I was trying to reduce the argument down to something manageable so it could be pulled back into a discussion.

I do think we should always focus on the predictive power of any theory, value or mathematics. It is easy for one to intentionally or unintentionally go off into the weeds. How quickly does the measure become the goal? :)