I believe these run into legal grey areas because they have to be obviously relevant to the job in question, which likely leads to the tests getting progressively more domain-specific as time goes on.

Programming does seem to be a field in which general intelligence is a big predictor of success. Whereas a field like management probably personality and general intelligence are more equally weighted.

Well yes. And that is what the current crop of "programming puzzle" interviews essentially are; the next iteration of attempting to measure people on a cardinal scale instead of an ordinal scale.

The key to remember is that evaluating on a cardinal scale is much more efficient than evaluating on an ordinal scale, so companies will do everything they can to transform the "secretary problem" into one where they measure absolute metrics instead of relative metrics.

Now, whether those metrics align with business value is a separate issue. All evidence suggests that they are largely independent of business value.

Basically a scientific experiment, have someone else ask and grade the questions, but those results were not considered for hire/nohire. Then 6 months after a bunch of people were hired we pull the results of the test and compare to their performance reviews. If the questions were a good predictor or future success we could use them on future interviews.

The overhead of this was more than we were willing to pay so we went back to HR's allowed questions.

Legality...not sure.

That's ludicrous nonsense. The primary breakthrough of the theory of relatively was taking what the math had been telling physicists about the universe for years seriously. In hindsight, everything necessary was there before Einstein, it just was very deeply assumed not to be reasonable. Relativity builds on numerous pre-existing foundations and is very much in the "standing on the shoulder of giants" tradition.

https://en.wikipedia.org/wiki/Theory_of_relativity#Developme... , first paragraph

Either you've misinterpreted Adam Frank's work, or it isn't excellent.

I'll counter-cite Reflections on Relativity, which contains a lot of deep analysis of the mathematical history of relativity both before and after Einstein: https://mathpages.com/rr/rrtoc.htm which makes it quite clear both what he did, and did not do.

That's the sign of an expert, not a generalist.

But he wasn't ignorant of them. He knew the existing work. He was extremely well read in physics and maths. He wasn't a generalist!

If you get a group of random generalists to build your compiler, or your graphics engine, or your cryptography stack, they aren't going to know what they don't know, and it'd take them decades to come up to speed.

The amount of unique stuff in each subfield is not that great, and if you have a wide and solid base in different subfields, picking up the unique ideas and techniques isn't that hard or time consuming.

But after a generalist has it, he is a lot more productive and creative because he often applies stuff from other subfields to come up with better/novel solution to the problem at hand - something a specialist cannot do.

Most opportunities look for some high degree of specificity. For contract work or a pre-defined task, that makes sense but for companies that need a rich talent pool, I don't know why there appears to be little opportunity for generalists.

I don’t think anyone considers it a difficult barrier compared to getting into FNG etc.

https://en.wikipedia.org/wiki/Phoenix_pay_system

In short, it's costing Canada billions of dollars and has affected countless people's livelihoods.

That doesn't mesh so well with the modern era of employers hating spending a dime on training, even if it would save them a dollar, and employees wanting to jump ship every other year for that sweet sweet 10% compensation bump

Thought in the US you will run into problems due to Griggs v. Duke Power Co.

One is purely statistical -- that if you offer any metric of success, and try to correlate it with any measure of intelligence, even if the two are actually uncorrelated, then having dead people in the sample (that is, people who score 0 on both) will increase the quality of the correlation. Applicable because when you eliminate people who score low on IQ tests and low on performance tests from the sample, the correlations tend to disappear in these studies.

The second is more intuitive, that intelligence is multifaceted and attempts to project it into a low-dimensional space will mask the inherent complexity, but will be capable of identifying individuals that score low across many dimensions. That leads to the condition in the first issue.

The third is probably the most controversial; that IQ tests in particular do not measure generalized intelligence, but measure how well you can do "drone work", so for places where performance can be objectively measured, it will tend to be in tasks that do not require creativity or initiative in unpredictable directions because those are not easily measured, so thus reflective of the same sorts of tasks that IQ tests measure.