Comment by exoverito | Hacker News Reader

Comment by exoverito | Hacker News Reader

Parent Full thread

Human ethnic groups are measurably different in genetic terms, as based on single nucleotide polymorphisms and allelic frequency. There are multiple PCA plots of the 1000 Genomes dataset which show clear cluster separation based on ancestry:

https://www.researchgate.net/figure/Example-Ancestry-PCA-plo...

We know ethnic groups vary in terms of height, hair color, eye color, melanin, bone density, sprinting ability, lactose tolerance, propensity to diseases like sickle cell anemia, Tay-Sachs, stomach cancer, alcoholism risk, etc. Certain medications need to be dosed differently for different ethnic groups due to the frequency of certain gene variants, e.g. Carbamazepine, Warfarin, Allopurinol.

The fixation index (Fst) quantifies the level of genetic variation between groups, a value of 0 means no differentiation, and 1 is maximal. A 2012 study based on SNPs found that Finns and Swedes have a Fst value of 0.0050-0.0110, Chinese and Europeans at 0.110, and Japanese and Yoruba at 0.190.

https://pmc.ncbi.nlm.nih.gov/articles/PMC2675054/

A 1994 study based on 120 alleles found the two most distant groups were Mbuti pygmies and Papua New Guineans at a Fst of 0.4573.

https://en.wikipedia.org/wiki/File:Full_Fst_Average.png

In genome wide association studies, polygenic score have been developed to find thousands of gene variants linked to phenotypes like spatial and verbal intelligence, memory, and processing speed. The distribution of these gene variants is not uniform across ethnic groups.

Given that we know there are genetic differences between groups, and observable variation, it stands to reason that there could be a genetic component for variation in intelligence between groups. It would be dogmatic to a priori claim there is absolutely no genetic component, and pretty obviously motivated out of the fear that inequality is much more intractable than commonly believed.

…but that just sounds like scientific racism.

Rather than judging an individual on their actual intelligence, these kinds of statistical trends allow you to justify judging an individual based on their race, because you feel you can credibly claim that race is an acceptable proxy for their genome, is an acceptable proxy for their intelligence.

Or for their trustworthiness, or creativity, or sexuality, or dutifulness, or compassion, or aggressiveness, or alacrity, or humility, etc etc.

When you treat a person like a people, that’s still prejudice.

Well, the rational thing is obviously to be scared of what ideas sound like.

> Rather than judging an individual on their actual intelligence

Actual intelligence is hard to know! However, lots of factors allow you to make a rapid initial estimate of their actual intelligence, which you can then refine as required.

(When the factors include apparent genetic heritage, this is called "racism" and society doesn't like it. But that doesn't mean it doesn't work, just that you can get fired and banned for doing it.)

((This is of course why we must allow IQ tests for hiring; then there's no need to pay attention to skin color, so liberals should be all for it.))

> Well, the rational thing is obviously to be scared of what ideas sound like.

Yes, actually. If an idea sounds like it can be used to commit crimes against humanity, you should pause. You should reassess said idea multiple times. You should be skeptical. You shouldn't ignore that feeling.

What a lot of people are missing is intent - the human element. Why were these studies conducted? Who conducted them?

If someone insane conducts a study then yes - that is absolutely grounds to be skeptical of said study. It's perfectly rationale. If extremely racist people produce studies which just so happen to be racist, we should take a step back and go "hmm".

Being right or being correct is one thing, but it's not absolutely valuable. The end-result and how "bad" it is also matters, and often times it matters more. And, elephant in the room, nobody actually knows if they're right. Making logical conclusions isn't so, because you are forced to make thousands of assumptions.

You might be right, you might not be. Let's all have some humility.

The assertion that "actual intelligence is hard to know" followed almost immediately by "apparent genetic heritage" is what's wrong with your opinion. And no, it doesn't work -- at least it doesn't work for identifying intelligence. It just works for selecting people who appeal to your bias.

IQ tests are not actual measurements of anything; this is both because nobody has a rigorous working definition of intelligence and because nobody's figured out a universal method of measuring achievement of what insufficient definitions we have. Their proponents are more interested in pigeonholing people than actually measuring anything anyway.

And as a hiring manager, I'd hire an idiot who is good at the job over a genius who isn't.

Actual intelligence is hard to know precisely/accurately/confidently, is what I meant, my bad. It's not hard to place a guess, and there's nothing wrong with it either so long as you're open for refinement. Your brain does it automatically in the first few seconds of seeing somebody, you can't even prevent it, and skin color is by no means the largest factor in it.

IQ as a metric is correlated with almost every life outcome. It's one of the most reliable metrics in psychology.

As a hiring manager, if you think an idiot can be good at the job, you either hire for an undemanding job or I'm not sure if you're good at yours.

SES, personality, and education are also correlated with life outcomes. Craniometry used to be one of the most reliable metrics in psychology.

Sure, I'm not saying IQ is everything. Do you have articles on craniometry actually being validated by studies? That sounds interesting.

SES, education, and personality are intimately interrelated with "IQ". Once you control for those, what's the leftover signal, and how do you know?

I mean, if IQ is real then SES and education are plausibly strongly caused by it. In that case, controlling for SES and education and then saying "there's not much signal left, thus IQ is bunk" just means that you've basically renamed IQ into "the implicit thing that determines SES and education, which I will not calculate directly." With such related properties it becomes hard to determine causation in general. The obvious test would be to improve education and see if IQ shifts. I don't have a study for this on hand, but I expect (just to register my prediction, not to make an argument) that if I look I'll find that the signal is weak that direction.

Or if IQ is less real than IQ-ists think it is, then it is plausibly strongly caused by SES and education. See?

I'm not even saying you're wrong (I think you are, but I don't have to defend that argument). I'm just saying the level of epistemic certainty you kicked this subthread off with was unwarranted. You know, "most reliable metrics in psychology" and all that.

I don't see how your argument puts my initial argument into doubt, tbh. If IQ isn't real but SES-and-education are, well then SES-and-education is the thing that you pick up at a glance. I'm not sure that the specific construction of causation here matters.

But also sure, I tend to assert my opinions pretty strongly in part to invite pushback.

All I'm saying is that "most reliable metrics in psychology" is less a mic drop than that sentence would make it sound. The arrows of causality here are extremely controversial --- not politically, but scientifically.

Sure, that's fair. I kinda wish the topic wasn't politicized so we could just get scientists to hash it out without having to ask "is that a scientific conclusion or do you think it would be politically disadvantageous to come to another answer".

My own view is "IQ is real and massively impactful", because of the people I've read on the topic, my understanding of biology, sociology and history, and my experience in general, but I haven't kept a list of citations to document my trajectory there.

That multiple correlation is just more evidence the test is flawed. No qualified psychologist would call it reliable, and that's been the case for some years now.

Not all work is knowledge work. You might want to broaden your horizons.

Your treatment of IQ is ridiculous. Give me access to a child for seven months, and I can increase their IQ score by 20 (and probably make myself an enemy in the process: IQ test drills are one of the dullest activities, since you can't even switch your brain off while doing them).

Intelligence is not a single axis thing. IQ test results are significantly influenced by socioeconomic factors. "Actual intelligence is hard to know" because it doesn't exist.

I have never yet known scientific racism to produce true results. I have known a lot of people to say the sorts of things you're saying: evidence-free claims that racism is fine so long as you're doing the Good Racism that Actually Works™, I Promise, This Time It's Not Prejudice Because It's Justified®.

No candidate genetic correlate of the g factor has ever replicated. That should be a massive flashing warning sign that – rather than having identified an elusive fact about reality that just so happens to not appear in any rigorous study – maybe you're falling afoul of the same in-group/out-group bias as nearly every group of humans since records begin.

Since I have no reason to believe your heuristic is accurate, we can stop there. However, to further underline that you're not thinking rationally: even if blue people were (on average) 2× as capable at spacial rotation-based office jobs than green people, it still wouldn't be a good idea to start with the skin colour prior and update from there, because that would lead to the creation of caste systems, which hinder social mobility. Even if scientific racism worked (which it hasn't to date!), the rational approach would still be to judge people on their own merits.

If you find it hard to assess the competence of your subordinates, to the point where you're resorting to population-level stereotypes to make hiring decisions, you're an incompetent manager and should find another job.

> Your treatment of IQ is ridiculous. Give me access to a child for seven months, and I can increase their IQ score by 20

That would be remarkable! Do you have a write-up/preprint on your protocol?

There's nothing remarkable about it:

• Drill.

• Goto step 1.

Does this make the child "more intelligent"? Not in any meaningful way! But they get better at IQ tests.

It's a fairly common protocol. I can hardly be said to have invented it: I was put through it. (Sure, I came up with a few tricks for solving IQ-type problems that weren't in the instruction books, but those tricks too can be taught.)

I really don't understand why people think IQ test results are meaningful. They're among the most obvious cases of Goodhart's law that I know. Make up a sport that most kids won't have practised before, measure performance, and probably that's about as correlated with the (fictitious) "g factor" as IQ tests are.

I mean, have you actually done this and have you documented your results? Preferably with a placebo and double blind?

The problem with "I've gone through this" is it's hard to analyze the counterfactual.

I'm not sure how you'd run a double-blind experiment on this. You can single-blind the experimenters, but the participants are always going to know whether they've been drilling IQ tests.

Your point about counterfactuals is good, but… subjectively, I ended up with a better understanding of IQ test genre conventions (which is also why I bang on so much about "culturally-specific": they really are). My speed at solving the problems doubled or tripled, and my accuracy went from 80%-ish to near 100%. This did not translate to any improvements to my real-life skill at anything (although, I suppose it might've generalised a bit to other multiple-choice exams). I've got a lot more evidence to analyse than just an n=1 scatter plot.

> the participants are always going to know whether they've been drilling IQ tests.

I'm not asking you to actually do this, but the participants (and experimenters!) don't necessarily have to know what you're testing. Maybe get one to drill IQ tests, one to drill Latin, one to drill chess and one to drill the piano.

Does your ability extend to IQ tests with other patterns? Also, does it extend to logic puzzles?

No, it actually slows me down on IQ tests that don't follow the genre conventions. (I used to approach them with a fresh mind – and I can still do that, if I have time to get into that mindset, but it's not always my default.) But I almost never see those, so…

It doesn't extend to logic puzzles, which I've always been quite bad at. (I find the Professor Layton games hard enough to be actively unfun, despite their beauty.) I can solve problems if they're contextualised, but my approach for solving logic puzzles is "identify a general algorithm, then execute it", which is quite slow.

As I've been telling you: IQ is extremely artificial; and doesn't measure general intelligence, because there's no such thing as "general intelligence". The "g factor" is a statistical regularity, but any statistician can tell you that while all sustained statistical regularities have explanations, they don't necessarily correspond to real things.

From time to time I see a press release to the effect that some movie star (Alyssa Milano was one) got an IQ score around the max of Raven's Progressive Matrices. I bet somewhere there is a psychologist who will coach you on it, across the street one will test you, and on the next block a PR agency that will make the press release.

I mean, there aren't that many questions on Raven, you could memorize them all, particularly if you've got the kind of intelligence that actors have -- being able to memorize your lines. (And that's something, I have a 1950-ish book about the television industry that makes a point that people expect performers to be a "quick study", you'd better know your lines really well and not have to be told twice that you are expected to do this or that. That's different from, say, being able to solve really complex math problems.)

Isn't the point of Raven that you can synthesize variants at will? Or am I misunderstanding?

I'd consider it well plausible that top movie stars are also very smart.

We do allow IQ tests for hiring. That we don't is a super-pernicious online myth. Some of the largest companies in America, with extremely deep pockets for employment law plaintiffs attorneys to come after, do in fact use IQ tests. Most companies don't, though, because they don't work.

Huh. Interesting, thank you, I didn't know that!

(The "they don't work" thing is my editorial gloss; the rest of it though is I think objectively verifiable.)

Um, I asked Grok and aren't IQ tests soft-outlawed under disparate impact? As opposed to skill qualifications, it's hard to prove that, say, you need to be "intelligent" to work at a certain job. You would end up having to argue that intelligence even exists in court, which I can see why companies would try to avoid- ie. disparate impact creates a reversed weight of evidence. If your test has disparate impact, you now have to prove that it's necessary, creating a chicken-and-egg problem.

Nope. It's trivially easy to pull up (a small minority of) Fortune 500 companies that use IQ tests in their hiring processes. The companies that offer these tests brag about the companies that use them. Whatever Grok thinks about this doesn't really much matter in the face of that evidence.

I think they use "aptitude tests" or "personality tests" that are at least packaged up to look relevant to the job, not direct naked IQ tests? I can't offhand find companies using actual IQ tests in hiring.

And if you're saying "well those are just repackaged IQ tests, so doesn't it count", then 1. it sure seems like IQ tests are illegal then, but 2. it also seems like they're so useful that companies are trying to smuggle them in anyway?

No, they use general cognitive tests, advertised as such. I don't think there's a way to wiggle out of this: IQ testing in hiring is fully lawful and accepted in US employment law.

Well, I couldn't find any that were general cognitive tests rather than some sort of skill or culture-fit test, at least ostensibly. Link?

>> That we don't is a super-pernicious online myth.

> Um, I asked Grok and

Where does Grok's training set come from?

>Some of the largest companies in America . . . do in fact use IQ tests.

Can you name one, please?

I saw your claim in this thread that the companies that make money by supplying these test to employers brag about how many large employers use the tests, but plenty of people brag falsely when bragging will tend to increase their revenue.

You can literally just go to Wonderlic and look at the logo crawl on their front page. It's not a secret.

I don't know what you're trying to say. That there is no race IQ difference? That we should not determine if there is one? How does one act on the feelings you've expressed?

GWAS results are so inhospitable to race/IQ talking points that race/IQ people have for the last several years been slagging GWAS, and molecular genomics more generally, and saying the twin studies work was the right approach all along. It's a whole thing.

As we all know, genetics and evolution only apply from the neck down.

True, nature is egalitarian although only intracranialy.

Reply on news.ycombinator.com