Unfortunately this is all too believable. Taleb’s covid hysteria was off the charts. Apparently he’d dismissed all the epidemiology that showed its seriousness was limited to those already vulnerable, such as the elderly. Instead he bought into the fear. It’s always disappointing when smart people don’t use their heads.
I have not read Ferguson’s vita, but I do have an acquaintance with his previous involvement in epidemiology in the past. Here is the first paragraph from a simple ChatGPT query: “A famous British epidemiologist known for modeling COVID-19 fatalities and other pandemics is Professor Neil Ferguson.” You can best take up your view with ChatGPT.
I went in the opposite direction when covid hit. I believed it was a relatively benign virus and hoped I would get it to develop natural immunity. I was so wrong, because I was operating from a false fact, that covid was a natural virus rather than tweaked in a lab. The spike protein can trigger so many pathologies. A friend avoided the virus and the shots, and I wish I had taken that approach.
Covid artificial or not, the electron microscope picture of the coronavirus looks the same as when I encountered it almost 50 years previously at university. The spikes may have been altered, but the virus was always classified as a cold back then and looked identical to today’s Covid virus.
The ultimate irony is that Taleb’s whole body of work points at tails being underrated as a driver of basically, well, everything. Then he dismisses the tails when talking about IQ, despite the fact that it’s the people at the high end IQ tail who clearly have an absolutely outsized impact on literally everything. IQ literally proves his point and fits in with his worldview, but he singles it out.
I have a suspicion that he might not really believe IQ is useless, but that he uses his objection to it as a club to beat back liberals and blank slatists and keep his western consensus bona fides intact.
"Then he dismisses the tails when talking about IQ, despite the fact that it’s the people at the high end IQ tail who clearly have an absolutely outsized impact on literally everything."
Yep, and is one of the reasons (smart fraction) why our society has been shielded these last few generations from effects of our intellectual decline through increases in third world low IQ migration and spiteful mutants.
Greg Cochran showed that NNT's "barbell" theory of investing based on his Black Swan outlook did not actually make more money. Cochran has in general not been impressed with Taleb. Up until I started reading Greg, I was a fan of Nassim and found his theories persuasive. He certainly is a master of finding persuasive anecdotes. But once the seal was broken I could see things wrong with a few of NNT's theories, and stopped following him.
I actually think Taleb has done a lot of good work, and very much enjoyed 'The Black Swan' and 'Antifragile'. But his arguments against IQ are just bad.
Ironic, I was a fan of Cochran until he went crazy wrt the Covid scamdemic. As I remember, he became enamored with fatality models by the likes of epidemiologists like Neil Ferguson whose track record in these matters had been extremely poor in the past. According to Cochran, Covid was going to kill 2M in the USA alone. His stridency toward those commenting in the negative on his blog turned me off and I never went back to reading him. Until then he seemed curmudgeonly, but fairly smart, quick, and insightful. Now I’m not so sure about the later. In short, he panicked which is a bad look for a scientist in the best of times—much less a pandemic.
He is rude to people who disagree with him (including me), and he has an excellent memory about it. As for Covid, at least 1.1 M did die in the US, and that is probably an undercount. The final total is very unlikely to have been 2M however.
We will never know the true number of U.S. citizens who died of Covid. Never. The statistics are totally confounded. Death certificates vary from State to State, hospitals were incentivized to record COVID for admissions, etc. No need to argue the point further, better people than me have describe such previously.
I very much dislike resorting to "no need to argue the point further" as an attempt to evade disagreement. So I will argue the point, via Scott Alexander, whose rebuttal and making the case I find persuasive.
Yes, Taleb is just a guru. He got lucky but much of what he says isn’t true at all, it’s just appealing. It’s not surprising he is revered by the finance bros and it’s not surprising he comes up with all kind of bad opinions when it comes to real matters.
Finance is mostly a scam for the rich to get even richer so I’m rather suspicious of anyone coming from that field anyway, even though they are definitely very smart people working there field, they are most definitely corrupt, so not to be trusted.
In addition to the evidence from predictive validity, it's worth mentioning the implausibility of Taleb's claim from the point of view of other forms of psychometric validity. Taleb asserts that IQ tests only measure unintelligence; therefore, differences in above average scores say nothing about the differences in intelligence between those score earners. But a little reflection shows what a strange notion this is. It would mean, for example that the primary reason an average person can solve certain items on Raven's Progressive Matrices and a below average person cannot is due to intelligence, but the reason a smaller percentage of people can solve items of even higher difficulty has nothing to do with intelligence, despite the more difficult items being nearly identical in content! Common sense says there's no reason difficult items drawn from the same pool should loose all validity past the median. Taleb has even asserted that past a certain point IQ tests measure nothing more than (paraphrasing) facility for mindlessly desk work. But there's no prima facie reason to believe that.
Taleb is also fond of mocking IQ for being circular, but this is also a mistake. Fundamentally at least, we don't judge the difficulty of professions by nothing their correlation with IQ. Rather, we take it for granted that certain jobs like professor of physics, engineer, or hedge fund manager require high intelligence, and then we note the positive correlation IQ scores have with ending up in such positions despite the fact the IQ scores were derived from tests consisting of items which, on the face of it, have little to nothing to do with these professions.
Thanks for the comment. I think the low GWAS heritability is due to a combination of factors: insufficient sample sizes, rare variants, incomplete tagging of common variants, measurement error etc.
"Insufficient sample sizes"? Even GWAS analysis of millions have utterly failed and are not compaginate with the results of the famouse (or rather infamous) twin studies with their very high heritibility estimates.
As for 'incomplete tagging of common variants" and "measurement error", I'd invite you to submit a preprint of your findings to a journal because if you think you've really found a common methodological error to GWAS studies than you'd be lauded and published in a high ranking journal before the end of the month.
But I suspect you're just throwing stuff at the wall and seeing what sticks hence all these vague plausible-sounding but fundamentally specious explanations.
As of right now, we simple know nothing about the nature of intelligence or its heritability because the most sophisticated tools we have so far have come up with nothing and contradicted previously held beliefs.
Most sophisticated tool ≠ most accurate tool or most useful tool for making predictions. This is fallacious reasoning.
It's well known tagging is incomplete in molecular genetics, and by measurement error I think he's referring to measurement error with regards to the measures of IQ used vs how twin study estimates use much better measures of IQ. This is also something well known by psychologists and behavioral geneticists for decades so your demand for him to publish a new study is silly.
If he thinks he's found legitimate methodological issues with GWAS then yes, he absolutely should send a pre-print to a journal - it's a very large claim and it requires substantiation. Otherwise, this is just throwing things at the wall to see what sticks. It's simply your opinion that twin studies 'use much better measures of IQ' - and it's based on nothing and does nothing to resolve the present heritability conundrum.
Needless to say, the twin studies have bad controls given ethical limitations and abysmal sample sizes (compared to GWAS). Currently we're in a foggy area and all these ex cathedra statements are ridiculous in the extreme. We simple don't know what's going on.
And using genetic variations to establish correlations between phenotypic variations removes a lot of intermediate noise - it is more 'accurate' in that sense. I never stated anything about 'making predictions' - that's a different matter entirely. My statement was very very circumscribed and is only related to explanations of observed variations in IQ (or more accurately 'g').
First, GWAS and causal are not really connected. GWAS as an approach is strictly statistical. Causal links are generally only feasible for strong monogenic causes, because those can be experimentally verified. For GWAS, NC has given some good points, so I'll write a bit about causal links:
Genetic variants are extremely disproportionally connected to diseases and disabilities - it's easier to break something that works than to improve something, especially by random chance.
Highly intelligent brains are arguably among the finest machinery there is, so you wouldn't expect a high-impact monogenic cause to begin with, and the most plausible model is rather a larger number of weakly associated variants.
I never said that GWAS is linked to causality. It obviously isn't - that's what the A in GWAS means, it seeks to establish correlations, which despite vast sample sizes it has failed to do to any meaningful level viz. IQ or "g". The rest of what you wrote is a non-sequitur.
Correct, your original post was a non-sequitur: First you write about GWAS, which looks at statistical correlational relationships. Then your second sentence moves on to causal links, which have nothing to do with GWAS, and to which my post replies.
Oh dear, this is tiresome. My contention was that we can currently say nothing about the heritability of intelligence. IQ/g leads to poor heritability results CORELLATIONALLY due to GWAS (obviously) - we have only a handful of genes here that are statistically significant (In any case <6% of total var explained by them). We have not a single gene that can be causally linked (this is a separate statement - and is equally true).
Taleb is a very odd duck. I think he mostly likes to stir the pot by taking extreme positions... just to be controversial.... There is a lot to be said, however, for his promotion of anti-fragility.
What original contributions has Taleb made to the field of rare events? The ideas in the field of finance contained in books like Fooled by Randomness and Black Swan, that thrust Taleb into prominence, are mostly a rehash of the ideas of French mathematician Benoit Mandelbrot (who, to be fair, Taleb credits)
His concept of the "fourth quadrant" and his concept of "anti-fragility" are quite original. I also think his popularisation of work on rare events counts as an important contribution.
A minor gripe of mine: people report IQ scores like, say, 125 or 145, which to the lay public conveys very little intormation, except maybe that 145 must be "a genius." It would be much more informative to phrase it in terms of how rare (or common) a given score is. I don't have a conversion table in front of me, but an IQ of 145 is not rare at all, certainly not in a society of many millions of people (is it one out of a thousand, I forget?). I've been told that one out of a hundred thousand is the threshold for doing top notch theoretical physics. That's much more informative than saying an IQ of 170 or whatever it is.
My bigger issue is that many tests have substantial ceiling issues, sometimes due to the test design (if 29/30 is 130 IQ, there is not much room left), but more importantly almost always due to validation problems: 160+ IQ is so rare that getting enough together to validate these scores is hard and has not been done for all tests. So in reality, the 70-130 range is highly informative and can be done with various tests, 130-160 requires special high-IQ tests but it is often not bothered so many of the claims are sketchy, 160+ is basically bunk. Needless to say, that goes especially for those alleged ultra-high IQ individuals. That's not to say those people aren't smart, it's just their specific claimed value is much less trustable.
Hi, I think most of the points have already been addressed by Taleb in his general corpus. I would try to address them here:
Sifting through various datasets until you find something that IQ has *some* explanatory power (the best one you mentioned was 25-30%) is not a rigorous scientific approach. Because one doesn't see the data that is NOT presented, I have to assume that what is presented is the best the researcher could come up with. And if the best a researcher can come up with is a model with less than 10% variance explained (with many cases of the data you cited), that is a huge signal that the relationship is actually just nothing.
Those minimal improvements on the tiny R^2 values by Brown when changing models tell us that we go from no explanatory power to no explanatory power! Even if small correlations exist, they lose predictive usefulness is fat tailed domains.
RE Norway/Finland/quantiles study: Indeed Taleb's implications follow from the nature of extreme variations in fat tailed domains. When you compress the extreme variations using quantiles, you get rid of that problem, but now you are quite uninformative about actual outcomes (if im in the top 25% of income earners, this doesnt tell me much about how much i might earn). You have not explained the variance, you have hidden it away/ignored it!
"For example, if you simulate a correlation of 0.5 with, say, 500 datapoints and then cover the bottom half of the chart, it will be hard to discern much of a relationship in the top half."
Unfortunately this is all too believable. Taleb’s covid hysteria was off the charts. Apparently he’d dismissed all the epidemiology that showed its seriousness was limited to those already vulnerable, such as the elderly. Instead he bought into the fear. It’s always disappointing when smart people don’t use their heads.
Worse, Taleb bought into discredited epidemiologist, Neil Ferguson’s modeling.
WTF? Ferguson is a mediocre historian with a knack for self-promotion, not an epidemiologist.
Hirsi Ali definitely married down there.
Neil, not Niall. https://en.wikipedia.org/wiki/Neil_Ferguson_(epidemiologist)
I have not read Ferguson’s vita, but I do have an acquaintance with his previous involvement in epidemiology in the past. Here is the first paragraph from a simple ChatGPT query: “A famous British epidemiologist known for modeling COVID-19 fatalities and other pandemics is Professor Neil Ferguson.” You can best take up your view with ChatGPT.
I went in the opposite direction when covid hit. I believed it was a relatively benign virus and hoped I would get it to develop natural immunity. I was so wrong, because I was operating from a false fact, that covid was a natural virus rather than tweaked in a lab. The spike protein can trigger so many pathologies. A friend avoided the virus and the shots, and I wish I had taken that approach.
Covid artificial or not, the electron microscope picture of the coronavirus looks the same as when I encountered it almost 50 years previously at university. The spikes may have been altered, but the virus was always classified as a cold back then and looked identical to today’s Covid virus.
I covered the entirety of Taleb’s article as a five-year anniversary: https://hereticalinsights.substack.com/p/iq-and-talebs-wild-ride
Excellent piece
—NC
The ultimate irony is that Taleb’s whole body of work points at tails being underrated as a driver of basically, well, everything. Then he dismisses the tails when talking about IQ, despite the fact that it’s the people at the high end IQ tail who clearly have an absolutely outsized impact on literally everything. IQ literally proves his point and fits in with his worldview, but he singles it out.
I have a suspicion that he might not really believe IQ is useless, but that he uses his objection to it as a club to beat back liberals and blank slatists and keep his western consensus bona fides intact.
"Then he dismisses the tails when talking about IQ, despite the fact that it’s the people at the high end IQ tail who clearly have an absolutely outsized impact on literally everything."
Indeed, that is true.
Yep, and is one of the reasons (smart fraction) why our society has been shielded these last few generations from effects of our intellectual decline through increases in third world low IQ migration and spiteful mutants.
Greg Cochran showed that NNT's "barbell" theory of investing based on his Black Swan outlook did not actually make more money. Cochran has in general not been impressed with Taleb. Up until I started reading Greg, I was a fan of Nassim and found his theories persuasive. He certainly is a master of finding persuasive anecdotes. But once the seal was broken I could see things wrong with a few of NNT's theories, and stopped following him.
I actually think Taleb has done a lot of good work, and very much enjoyed 'The Black Swan' and 'Antifragile'. But his arguments against IQ are just bad.
—NC
Ironic, I was a fan of Cochran until he went crazy wrt the Covid scamdemic. As I remember, he became enamored with fatality models by the likes of epidemiologists like Neil Ferguson whose track record in these matters had been extremely poor in the past. According to Cochran, Covid was going to kill 2M in the USA alone. His stridency toward those commenting in the negative on his blog turned me off and I never went back to reading him. Until then he seemed curmudgeonly, but fairly smart, quick, and insightful. Now I’m not so sure about the later. In short, he panicked which is a bad look for a scientist in the best of times—much less a pandemic.
He is rude to people who disagree with him (including me), and he has an excellent memory about it. As for Covid, at least 1.1 M did die in the US, and that is probably an undercount. The final total is very unlikely to have been 2M however.
“As for Covid, at least 1.1 M did die in the US…”
We will never know the true number of U.S. citizens who died of Covid. Never. The statistics are totally confounded. Death certificates vary from State to State, hospitals were incentivized to record COVID for admissions, etc. No need to argue the point further, better people than me have describe such previously.
I very much dislike resorting to "no need to argue the point further" as an attempt to evade disagreement. So I will argue the point, via Scott Alexander, whose rebuttal and making the case I find persuasive.
https://www.astralcodexten.com/p/the-evidence-that-a-million-americans
Yes, Taleb is just a guru. He got lucky but much of what he says isn’t true at all, it’s just appealing. It’s not surprising he is revered by the finance bros and it’s not surprising he comes up with all kind of bad opinions when it comes to real matters.
Finance is mostly a scam for the rich to get even richer so I’m rather suspicious of anyone coming from that field anyway, even though they are definitely very smart people working there field, they are most definitely corrupt, so not to be trusted.
Didn't Kareem Carr suggest 2+2 = 5?
https://beunbound.us/blog/harvard-two-plus-two-equals-five-kareem-carr/
In addition to the evidence from predictive validity, it's worth mentioning the implausibility of Taleb's claim from the point of view of other forms of psychometric validity. Taleb asserts that IQ tests only measure unintelligence; therefore, differences in above average scores say nothing about the differences in intelligence between those score earners. But a little reflection shows what a strange notion this is. It would mean, for example that the primary reason an average person can solve certain items on Raven's Progressive Matrices and a below average person cannot is due to intelligence, but the reason a smaller percentage of people can solve items of even higher difficulty has nothing to do with intelligence, despite the more difficult items being nearly identical in content! Common sense says there's no reason difficult items drawn from the same pool should loose all validity past the median. Taleb has even asserted that past a certain point IQ tests measure nothing more than (paraphrasing) facility for mindlessly desk work. But there's no prima facie reason to believe that.
Taleb is also fond of mocking IQ for being circular, but this is also a mistake. Fundamentally at least, we don't judge the difficulty of professions by nothing their correlation with IQ. Rather, we take it for granted that certain jobs like professor of physics, engineer, or hedge fund manager require high intelligence, and then we note the positive correlation IQ scores have with ending up in such positions despite the fact the IQ scores were derived from tests consisting of items which, on the face of it, have little to nothing to do with these professions.
Great revisit of this subject.
As I have said before, it appears Taleb has an agenda.
I guess there is a good reason I never heard of Nassim Taleb.
Why do you think GWAS picks up so little variation - and as of right now not a single gene can be casually linked to higher intelligence
Thanks for the comment. I think the low GWAS heritability is due to a combination of factors: insufficient sample sizes, rare variants, incomplete tagging of common variants, measurement error etc.
—NC
"Insufficient sample sizes"? Even GWAS analysis of millions have utterly failed and are not compaginate with the results of the famouse (or rather infamous) twin studies with their very high heritibility estimates.
As for 'incomplete tagging of common variants" and "measurement error", I'd invite you to submit a preprint of your findings to a journal because if you think you've really found a common methodological error to GWAS studies than you'd be lauded and published in a high ranking journal before the end of the month.
But I suspect you're just throwing stuff at the wall and seeing what sticks hence all these vague plausible-sounding but fundamentally specious explanations.
As of right now, we simple know nothing about the nature of intelligence or its heritability because the most sophisticated tools we have so far have come up with nothing and contradicted previously held beliefs.
Most sophisticated tool ≠ most accurate tool or most useful tool for making predictions. This is fallacious reasoning.
It's well known tagging is incomplete in molecular genetics, and by measurement error I think he's referring to measurement error with regards to the measures of IQ used vs how twin study estimates use much better measures of IQ. This is also something well known by psychologists and behavioral geneticists for decades so your demand for him to publish a new study is silly.
If he thinks he's found legitimate methodological issues with GWAS then yes, he absolutely should send a pre-print to a journal - it's a very large claim and it requires substantiation. Otherwise, this is just throwing things at the wall to see what sticks. It's simply your opinion that twin studies 'use much better measures of IQ' - and it's based on nothing and does nothing to resolve the present heritability conundrum.
Needless to say, the twin studies have bad controls given ethical limitations and abysmal sample sizes (compared to GWAS). Currently we're in a foggy area and all these ex cathedra statements are ridiculous in the extreme. We simple don't know what's going on.
And using genetic variations to establish correlations between phenotypic variations removes a lot of intermediate noise - it is more 'accurate' in that sense. I never stated anything about 'making predictions' - that's a different matter entirely. My statement was very very circumscribed and is only related to explanations of observed variations in IQ (or more accurately 'g').
First, GWAS and causal are not really connected. GWAS as an approach is strictly statistical. Causal links are generally only feasible for strong monogenic causes, because those can be experimentally verified. For GWAS, NC has given some good points, so I'll write a bit about causal links:
Genetic variants are extremely disproportionally connected to diseases and disabilities - it's easier to break something that works than to improve something, especially by random chance.
Highly intelligent brains are arguably among the finest machinery there is, so you wouldn't expect a high-impact monogenic cause to begin with, and the most plausible model is rather a larger number of weakly associated variants.
I never said that GWAS is linked to causality. It obviously isn't - that's what the A in GWAS means, it seeks to establish correlations, which despite vast sample sizes it has failed to do to any meaningful level viz. IQ or "g". The rest of what you wrote is a non-sequitur.
Correct, your original post was a non-sequitur: First you write about GWAS, which looks at statistical correlational relationships. Then your second sentence moves on to causal links, which have nothing to do with GWAS, and to which my post replies.
Oh dear, this is tiresome. My contention was that we can currently say nothing about the heritability of intelligence. IQ/g leads to poor heritability results CORELLATIONALLY due to GWAS (obviously) - we have only a handful of genes here that are statistically significant (In any case <6% of total var explained by them). We have not a single gene that can be causally linked (this is a separate statement - and is equally true).
You are duller than dishwater.
Taleb is a very odd duck. I think he mostly likes to stir the pot by taking extreme positions... just to be controversial.... There is a lot to be said, however, for his promotion of anti-fragility.
What original contributions has Taleb made to the field of rare events? The ideas in the field of finance contained in books like Fooled by Randomness and Black Swan, that thrust Taleb into prominence, are mostly a rehash of the ideas of French mathematician Benoit Mandelbrot (who, to be fair, Taleb credits)
His concept of the "fourth quadrant" and his concept of "anti-fragility" are quite original. I also think his popularisation of work on rare events counts as an important contribution.
—NC
I think the important word 'against' is missing in the paragraph about 'Scientific Accomplishment', between 'evidence' and 'his'.
Thanks — fixed.
—NC
A minor gripe of mine: people report IQ scores like, say, 125 or 145, which to the lay public conveys very little intormation, except maybe that 145 must be "a genius." It would be much more informative to phrase it in terms of how rare (or common) a given score is. I don't have a conversion table in front of me, but an IQ of 145 is not rare at all, certainly not in a society of many millions of people (is it one out of a thousand, I forget?). I've been told that one out of a hundred thousand is the threshold for doing top notch theoretical physics. That's much more informative than saying an IQ of 170 or whatever it is.
My bigger issue is that many tests have substantial ceiling issues, sometimes due to the test design (if 29/30 is 130 IQ, there is not much room left), but more importantly almost always due to validation problems: 160+ IQ is so rare that getting enough together to validate these scores is hard and has not been done for all tests. So in reality, the 70-130 range is highly informative and can be done with various tests, 130-160 requires special high-IQ tests but it is often not bothered so many of the claims are sketchy, 160+ is basically bunk. Needless to say, that goes especially for those alleged ultra-high IQ individuals. That's not to say those people aren't smart, it's just their specific claimed value is much less trustable.
Hi, I think most of the points have already been addressed by Taleb in his general corpus. I would try to address them here:
Sifting through various datasets until you find something that IQ has *some* explanatory power (the best one you mentioned was 25-30%) is not a rigorous scientific approach. Because one doesn't see the data that is NOT presented, I have to assume that what is presented is the best the researcher could come up with. And if the best a researcher can come up with is a model with less than 10% variance explained (with many cases of the data you cited), that is a huge signal that the relationship is actually just nothing.
Those minimal improvements on the tiny R^2 values by Brown when changing models tell us that we go from no explanatory power to no explanatory power! Even if small correlations exist, they lose predictive usefulness is fat tailed domains.
RE Norway/Finland/quantiles study: Indeed Taleb's implications follow from the nature of extreme variations in fat tailed domains. When you compress the extreme variations using quantiles, you get rid of that problem, but now you are quite uninformative about actual outcomes (if im in the top 25% of income earners, this doesnt tell me much about how much i might earn). You have not explained the variance, you have hidden it away/ignored it!
"For example, if you simulate a correlation of 0.5 with, say, 500 datapoints and then cover the bottom half of the chart, it will be hard to discern much of a relationship in the top half."
Can somebody visualize this?
You can do it with ChatGPT.
—NC
Who knew Talib was a science denier? I
suppose liberals fall hard, even on their own swords. Anyone who has had good as well as bad secretaries knows that IQ is real and meaningful.