gwern comments on Suppose HBD is True - Less Wrong

-12 Post author: OrphanWilde 21 April 2016 01:34PM

You are viewing a comment permalink. View the original post to see all comments and the full post content.

Article Navigation

Comments (178)

CC Licenced

Comments (178)

Sort By: Best

You are viewing a single comment's thread. Show more comments above.

Comment author: gwern 21 April 2016 04:50:42PM * 8 points [-]

To which I must respond that, while it should surprise us if those genetic causation factors don't vary in frequency across population groups, it should also surprise us if the frequency distribution of given genetic factors consistently advantages or disadvantages a single population group. (That is, there should be genetic causal factors which improve intelligence that are more common in black people, as well.)

No. Think about a random walk on a 1D line, or about generating a normal based on the sum of a lot of random variates. If you do 2 random walks A and B, do you think that A and B will wind up in the exact same position because 'there should be steps which increase the position in B as well'? Or, 'A and B should sum to the same, because B will have some variables which were higher than in A'? In expectation, A and B may have the same average value, and may asymptotically converge given enough steps or variates summed but in any particular realization, with a fixed number of steps of summed variates, they can and probably will be quite far apart. Genetic drift is a considerable force, and more still if you invoke any kind of different selection pressures (such as, say, for other expensive phenotyptic traits like infection resistance at the expense of highly expensive brain development).

But that isn't the position HBD advocates generally take, to put it mildly.

I disagree. This may not be the position that racists take, but HBDers often suggest outbreeding & hybrid vigor as good things and from Jensen onward have emphasized the problematic effects of inbreeding depression & homozygosity. Razib & Jayman practice it, and HBD Chick, Cochran also come to mind as bringing up these issues fairly regularly, I think Chuck probably has too. More historically, I believe this was one of the disputed topics among early eugenicists; for example, Razib gives a quote from R.A. Fisher from page 238 of The Genetical Theory of Natural Selection:

The general consequences of race mixture can be predicted with confidence…Their general character will therefore be intermediate, but their variability will be greater than that of the original races. Morever, new combinations of virtue and ability, and of their opposites, will appear in the mixed race, combinations which are not necessarily heterozygous, but may be fixed as permanent racial characters. There are thus in the mixed race great possibilities for the action of selection. If selection is beneficient, and the better types leave the greater number of descendants, the ultimate effect of mixture will be the production of a race, not inferior to either those from which it sprang, but rather superior to both, in so far as the advantages of both can be combined. Unfavorable selection, on the other hand, will be more rapidly disastrous to a mixed race than to its progenitors. It should of course be remembered that all existing races show very great variability in respect of hereditary factors, so that selections of the intensity to which mankind is exposed would be capable of producing rapid changes, even in the purest existing race.

So I'm surprised you have this impression.

Comment author: OrphanWilde 21 April 2016 06:20:33PM -1 points [-]

No. Think about a random walk on a 1D line, or about generating a normal based on the sum of a lot of random variates. If you do 2 random walks A and B, do you think that A and B will wind up in the exact same position because 'there should be steps which increase the position in B as well'? Or, 'A and B should sum to the same, because B will have some variables which were higher than in A'? In expectation, A and B may converge, but in any particular realization, they can and probably will be quite far apart; there's not that many genetic variants.

There are, last I checked, at least 100 genetic markers associated with intelligence. And convergence, short of genetic drift/selection pressures, should be expected; while it's possible for there to be a difference, given the law of large numbers, we should expect the distribution to be pretty similar.

Genetic drift is a considerable force, and more still if you invoke any kind of different selection pressures (such as, say, for other expensive phenotyptic traits like infection resistance at the expense of highly expensive brain development).

Selection pressures like plagues that wiped out the majority of the population on more than one occasion?

I disagree. This may not be the position that racists take, but HBDers often suggest hybrid vigor and from Jensen onward have emphasized the problematic effects of inbreeding depression. Razib, Jayman, HBD Chick, Cochran all come to mind as bringing up these issues regularly and I'm surprised you have this impression.

Why are you surprised?

Comment author: gwern 21 April 2016 06:33:25PM * 5 points [-]

There are, last I checked, at least 100 genetic markers associated with intelligence.

Many more actually, the upcoming SSGAC paper alone reportedly identifies 80+ hits but even that paper's full polygenic score only explains about a third to a fifth of SNP contributions, so there's going to be many more hits to come and probably thousands of non-zero variants beyond that. But it doesn't make a difference because both group and individual differences are based on the same set. Individual differences arise from very small net differences (due to the CLT), and so you only need small changes in allele frequencies to also produce group differences on the order of individual differences. Take a look at my calculations and simulations in http://www.gwern.net/Embryo%20selection#limits-to-iterated-selection making concrete the issue of how much absolute genetic difference translates to observed relative differences; it's not much, and it would take a very small average difference to produce group differences like we see. Or look at the scale of Piffer's polygenic scores. Also consider 'soft sweeps'.

(I was looking into this because it has some important implications: the small variance means that embryo selection is going to be weak since you don't get embryos with large differences in their polygenic scores, but it also means that there is an enormous amount of potential improvement you could make with direct embryo editing. If the genetics of intelligence were just 50 genes or something small like that, selection would be more profitable since the sum of 50 random binomials is much more spread out than 10,000, but it also means that once you've edited all 50, you've 'run out' of genes to tweak and have topped out at a relatively few SDs of relative intelligence improvement. But with 10,000, you have so many knobs to tweak that you can go straight to whatever the neurophysiological limit of a human brain is.)

Selection pressures like plagues that wiped out the majority of the population on more than one occasion?

'Majority'? I don't think even the Black Plague killed a majority of the European population, if that's what you mean. And no. Only a few occasions doesn't create much of a selection pressure, compared to constant disease and parasite load over deca-millennia. Try the breeder's equation on the impact of a few dozen selection events killing 1/3 of the population versus say 10,000 selection events killing 10% of the population.

while it's possible for there to be a difference, given the law of large numbers, we should expect the distribution to be pretty similar.

'Pretty similar' is not nearly enough, and again, you can't neglect genetic drift and selection pressures. It's a fact that human populations do not experience significant gene flow and so genetic drift will be operating.

Why are you surprised?

Because if you're going to claim that HBD is irrelevant and futile and has no policy or real-world implications, I assume you must have been reading extensively about HBD to understand all the threads that go into it, which will lead you to those writers frequently.

Comment author: OrphanWilde comment score below threshold [+] (13 children)

Comment author: OrphanWilde 21 April 2016 07:24:27PM -3 points [-]

Individual differences arise from very small net differences (due to the CLT), and so you only need small changes in allele frequencies to also produce group differences on the order of individual differences. Take a look at my calculations and simulations in http://www.gwern.net/Embryo%20selection#limits-to-iterated-selection making concrete the issue of how much absolute genetic difference translates to observed relative differences; it's not much, and it would take a very small average difference to produce group differences like we see

Again, absent selection pressures, and given that genetic drift is responsible, we should see the same kinds of variability on a more localized scale; we should see African subpopulations (given the large number of distinct genetic populations) with higher IQs than white people.

'Majority'? I don't think even the Black Plague killed a majority of the European population, if that's what you mean. And no. Only a few occasions doesn't create much of a selection pressure, compared to constant disease and parasite load over deca-millennia. Try the breeder's equation on the impact of a few dozen selection events killing 1/3 of the population versus say 10,000 selection events killing 10% of the population.

The worst occurrence did, yes. But assuming this is the case, we should expect black populations to have higher disease resistance relative to white populations; is this the case?

'Pretty similar' is not nearly enough, and again, you can't neglect genetic drift and selection pressures. It's a fact that human populations do not experience significant gene flow and so genetic drift will be operating.

Except we're not talking about one white population and one black population, but dozens of each.

Because if you're going to claim that HBD is irrelevant and futile and has no policy or real-world implications, I assume you must have been reading extensively about HBD to understand all the threads that go into it, which will lead you to those writers frequently.

Ah. No. After the first few just-so stories based on armchair philosophy while deliberately ignoring half the studies out there in a deeply conflicted field of study, I got bored with the topic.

Comment author: gwern 21 April 2016 07:58:54PM * 4 points [-]

we should see African subpopulations (given the large number of distinct genetic populations) with higher IQs than white people.

A few points here:

we already know European and East Asian populations are probably the smartest genetically because they are the smartest now phenotypically; this has already been conditioned on, so it's illogical to say 'well, we should expect an African subpopulation to be higher'. The race has already been run and the first and second place prizes handed out, it makes no sense to say 'there were a lot of other runners so maybe one of them is in first place'.
the jury is still out on whether some African subpopulations might approach European/East Asian genetic levels. There is already a wide spread of per capita incomes there. African immigrants to the USA perform famously better than the native Africa-American population, and we don't know if this is solely due to a migrant selection effect. Nigerians in particular seem to do well, IIRC. (Piffer's results don't indicate any African subpopulations like this, but I have a lot of doubts about the method and whether he has enough SNPs or reference genomes to rule out sampling error and systematic biases.)
African subpopulations could be expected to be more closely related than to Europeans or Eastern Europeans, reducing the variability
and of course, selection cannot be ruled out. Selection is pretty important because we still need to explain why intelligence, which is so useful, has not been driven to fixation, and some of the older theories like mutation load are dead in the water, leaving only a few viable theories like balancing selection, which would directly imply that different genetic levels are due to tradeoffs for metabolic resources or faster lifecycles.

The worst occurrence did, yes.

WP says 30-70% is the usual range, but all of the majority percentages are isolated to small area or are urban areas which make up only a small fraction of national populations in that time period & would be expected to have much higher mortality due to density. In any case, even if the Black Death did kill 70% of the population, one event cannot compare to many millennia of constant disease burden in selection power.

Except we're not talking about one white population and one black population, but dozens of each.

That makes your point worse, not better. The more subpopulations, the smaller each on average and the more powerful drift is, and the larger the spread of extremes. The maximal point drawn out of 2 samples with small variance is smaller than drawn out of dozens of samples with a larger variance. (Also relevant to embryo selection: the more embryos you generate, the better your chance of getting an unusually high scoring one to choose. Diminishing returns, of course, but still large increases initially; the equations & simulations are included in my embryo selection essay.)

Ah. No. After the first few just-so stories based on armchair philosophy while deliberately ignoring half the studies out there in a deeply conflicted field of study, I got bored with the topic.

I see. In any case, I'm glad to see that based on your replies so far, you've abandoned your position that 'the truth of HBD cannot matter' and are settling for arguing 'HBD is false'.

Comment author: Desrtopa 22 April 2016 03:58:18AM 3 points [-]

we already know European and East Asian populations are probably the smartest genetically because they are the smartest now phenotypically; this has already been conditioned on, so it's illogical to say 'well, we should expect an African subpopulation to be higher'. The race has already been run and the first and second place prizes handed out, it makes no sense to say 'there were a lot of other runners so maybe one of them is in first place'.

It's likely, but I think it's important to remember that there are a lot of environmental factors which can depress IQ, and some populations may have high genetic potential which is being depressed by circumstance.

The Dutch are the tallest nationality in the world today, but 150 years ago, they were among the shorter ones; as average height has risen in Western nations, the Dutch significantly overtook various populations that used to be much taller than they were.

Comment author: OrphanWilde 21 April 2016 08:27:59PM -1 points [-]

we already know European and East Asian populations are probably the smartest genetically because they are the smartest now phenotypically; this has already been conditioned on, so it's illogical to say 'well, we should expect an African subpopulation to be higher'. The race has already been run and the first and second place prizes handed out, it makes no sense to say 'there were a lot of other runners so maybe one of them is in first place'.

Given that European and East Asian populations have benefitted from substantial increases in intelligence on the order of at least 20 IQ points since we started measuring (the Flynn effect), it's not illogical at all.

African subpopulations could be expected to be more closely related than to Europeans or Eastern Europeans, reducing the variability

Africa is -huge-, with eight distinct mountain-separated geographic areas, and multiple island populations. IIRC, most genetic diversity in the human race exists in Africa.

and of course, selection cannot be ruled out.

Selection can be ruled out, because it begs a serious question: What selection process is universal across a continent with dozens of distinct biomes which are in no way unique to the continent, but somehow fails to appear anywhere else in the world?

WP says 30-70% is the usual range, but all of the majority percentages are isolated to small area or are urban areas which make up only a small fraction of national populations in that time period & would be expected to have much higher mortality due to density. In any case, even if the Black Death did kill 70% of the population, one event cannot compare to many millennia of constant disease burden in selection power.

Except that Europe wasn't exempt from the disease burden, and had it worse because of population density.

That makes your point worse, not better. The more subpopulations, the smaller each on average and the more powerful drift is, and the larger the spread of extremes. The maximal point drawn out of 2 samples is smaller than drawn out of dozens of samples. (Also relevant to embryo selection: the more embryos you generate, the better your chance of getting an unusually high scoring one to choose. Diminishing returns, of course, but still large increases initially; the equations & simulations are included in my embryo selection essay.)

Because the genetic drift of each geographically distinct subpopulation is going to be independent, it should really surprise us if they all happen to drift in the same direction. We should expect a normal distribution among subpopulations - meaning that we should expect some of the subpopulations to be smarter than average for global populations, once the Flynn effect is finished and all nutrition and health and society related gained can be made.

Comment author: gwern 21 April 2016 09:03:10PM * 4 points [-]

Given that European and East Asian populations have benefitted from substantial increases in intelligence on the order of at least 20 IQ points since we started measuring (the Flynn effect), it's not illogical at all.

The Flynn effect is dubious and always has been, so this is not much of a counter-argument. No one has demonstrated a meaningful increase in overall intelligence, just on one or two types of subtests, and if you've been following the latest research - since you've surely been reading up on these topics and following Thompson's blog if you are going to authoritatively sweep HBD to the ash heap of history, doubtless - the gains seem to reflect an attitude or conceptual shift rather than any sort of real intelligence gain. These hollow gains are why we see nothing on backwards digit span, nothing on reaction time, nothing on all the other aspects of intelligence, no 10x+ increase in tails/genius as expected from a +1.3SD in means...

Africa is -huge-, with eight distinct mountain-separated geographic areas, and multiple island populations.

And yet, those geographic areas are still generally closer to each other and the island populations than they are to East Asia, say, and so my point stands, there will be more gene flow between them than more distant populations. They will be correlated and not as variable as one would expect. Which of the African groups will be highest genetically? I dunno, and it'll be interesting to see what the refined polygenic scores indicate; disease and parasite load might be a red herring and the relevant evolutionary environmental condition completely different.

IIRC, most genetic diversity in the human race exists in Africa.

This is true but it does not mean that African populations are more closely related to random non-African populations than to each other. (See also Lewontin's fallacy.) This has as much to do with the rapid expansion of non-African populations from small colonizing populations; the number of variants in a population reflect its long-term size, and humans have been in Africa for hundreds of thousands of years while in some places like South America those populations have been there for maybe 10,000 years at most, so they reflect a lack of initial diversity and not enough mutations have accumulated to make them as genetically diverse as the African population as a whole.

What selection process is universal across a continent with dozens of distinct biomes which are in no way unique to the continent, but somehow fails to appear anywhere else in the world?

Who said it fails to appear anywhere else in the world? There are a lot of undeveloped countries which have serious problems. Look at Brazil for a highly dysfunctional country on the equator with similar issues and which has continually failed to develop well despite considerable apparent potential.

Except that Europe wasn't exempt from the disease burden, and had it worse because of population density.

Yes, it was, because it was further north and colder and had different ecologies.

Because the genetic drift of each geographically distinct subpopulation is going to be independent, it should really surprise us if they all happen to drift in the same direction. We should expect a normal distribution among subpopulations - meaning that we should expect some of the subpopulations to be smarter than average for global population

Again, since populations are phylogenetic trees, there are going to be clusters of more and less related populations which will have less variability than expected, so what we see with Europe and East Asia could well be the two clusters which drifted highest and won the race to the Great Divergence. In an African cluster, there are enough that we can expect some to go near Europe/Asia levels, but if they exceeded Europe/Asia levels, they would have broken out before or at some point shortly afterwards. But there is no South Korea of Africa.

Comment author: OrphanWilde 22 April 2016 01:50:56PM -2 points [-]

The Flynn effect is dubious and always has been, so this is not much of a counter-argument. No one has demonstrated a meaningful increase in overall intelligence, just on one or two types of subtests, and if you've been following the latest research - since you've surely been reading up on these topics and following Thompson's blog if you are going to authoritatively sweep HBD to the ash heap of history, doubtless - the gains seem to reflect an attitude or conceptual shift rather than any sort of real intelligence gain. These hollow gains are why we see nothing on backwards digit span, nothing on reaction time, nothing on all the other aspects of intelligence, no 10x+ increase in tails/genius as expected from a +1.3SD in means...

Yes. IQ is pretty obviously at least partially a trainable skill. This doesn't support HBD - and in fact argues against a substantial part of the evidence.

And yet, those geographic areas are still generally closer to each other and the island populations than they are to East Asia, say, and so my point stands, there will be more gene flow between them than more distant populations. They will be correlated and not as variable as one would expect. Which of the African groups will be highest genetically? I dunno, and it'll be interesting to see what the refined polygenic scores indicate; disease and parasite load might be a red herring and the relevant evolutionary environmental condition completely different.

Yet Africa has more genetic diversity than anywhere else.

Who said it fails to appear anywhere else in the world? There are a lot of undeveloped countries which have serious problems. Look at Brazil for a highly dysfunctional country on the equator with similar issues and which has continually failed to develop well despite considerable apparent potential.

You mean a country which LBJ deliberately helped destabilize? The fingerprints of the US are clear on the civil wars which tore apart South America, just as much as the fingerprints of Europe are all over the civil wars which tore apart Africa.

Yes, it was, because it was further north and colder and had different ecologies.

You realize portions of Africa are as far South as Europe is North from the equator?

Again, since populations are phylogenetic trees, there are going to be clusters of more and less related populations which will have less variability than expected, so what we see with Europe and East Asia could well be the two clusters which drifted highest and won the race to the Great Divergence. In an African cluster, there are enough that we can expect some to go near Europe/Asia levels, but if they exceeded Europe/Asia levels, they would have broken out before or at some point shortly afterwards. But there is no South Korea of Africa.

If we go back further in history, of course, we see a previous "Great Divergence", between North Africa and far Southern Europe and everywhere else. The northern Europeans were primitive savages during this era. Further back, and North Africa and the Middle East were the seat of civilization.

Comment author: drethelin 21 April 2016 07:33:04PM 1 point [-]

Why would you ever assume that selection pressures were absent?

Comment author: OrphanWilde comment score below threshold [+] (6 children)

Comment author: gwern 21 April 2016 08:20:20PM 4 points [-]

A universal selection pressure doesn't produce dozens of distinct population groups, it would homogenize them. What would help produce distinct population groups - in addition to the genetic drift and inherent random walk behavior of all partially or wholly separated population groups with non-panmictic mating behavior which would produce large differences anyway - is different selection pressures for different things in different places, potentially varying over time due to environmental and social changes. (Such as, to name one possibility, greater disease burden making intelligence less fit compared to diverting those metabolic resources into faster maturation or more robust immune systems.) Since the Arctic is not Asia is not Africa, there is no surprise that there might be different selection pressures, and the real question is why one would expect selection pressures to be universally identical such that evolution would stop at the neck.

Comment author: OrphanWilde comment score below threshold [+] (4 children)

gwern comments on Suppose HBD is True - Less Wrong

Article Navigation

Comments (178)

Nearest Meetups

Virtual Study Room

Recent Comments

Recent Posts

Recent Rationality Quotes

Recent Wiki Edits

Top Contributors, 30 Days

Recent Karma Awards

Comments (178)

Top Contributors, 30 Days

You'll need to login or register to do that

(Don't worry, it only takes a few seconds)

Create

Login