This podcast is brought to you by the Albany Public Library main branch in the generosity of listeners like you. Like,
God getting these people talking,
cons unsupervised learning.
Hey, everybody, this is Razib Khan with the other supervised learning podcast. And I am here with Dr. Ruben our slot. Ruben, can you introduce yourself?
Sure. I'm a postdoctoral researcher at the University of Leipzig, currently, and my research focuses on biological influences on individual differences. So I'm very interested in the genetics of intelligence. But I've also done some work on hormonal effects on sexuality, especially across the menstrual cycle. And yeah, I did my PhD training in a sort of evolutionary psychology slash personality psychology department, and moved around a little after that.
So I want to talk about some papers and topics that you've been covering in, you know, if anyone goes to your Google Scholar, they will see that, you know, you obviously cover those topics, but I mean, I, I feel like, you know, there's some scholars who, I don't know, they, they kind of focus on one specific topic. And you know, you're using similar methods, I guess. But I mean, you go from topic to topic a little bit. And so I want to first talk about this a paper you have that just came out that intelligence can't be, or can be detected was not found attractive in videos, a lot of interactions? And can you talk about the context with Jeffrey Miller's work in sexual selection and evolutionary psychology and all that stuff?
Sure. So I read Jeffrey's book, the mating mind, when I was a student in Berlin. And yeah, I actually found it super interesting. It's very inspiring. I went on to do a very brief internship with los pangkor. My when my later PhD supervisor, then and then some of my first research topics were motivated in part by finding this theory of is interesting, and trying to submit it to some empirical tests coming at it from different angles, because it's not, it's not a trivial prediction to test in the end. So I mean, I'm not going to summarize the whole book here. So it's been a while since I really read it. But basically, the idea that human intelligence has been driven to this, to the extreme levels we observe today, by not only natural selection, which I think is the common understanding of the topic, but also by sexual selection. Basically, human beings, vetting each other's intelligence and going for partners with higher intelligence, and that this can more easily explain why we have such an abundance of intelligence and also the ability in areas that aren't immediately relevant for survival, like music, dance, so and I found that very intriguing. The book was super readable as well, I think I recovered. I gave it away as a gift during my first seminar I taught when I started my PhD. So I was definitely a fan. But I've kind of fallen out of love with it.
Yeah. Yeah, let me reiterate, it's a really good book, it has a lot of food for thought, that's the way I would put it. You know, I'm friendly with Jeffrey like, I like him as a person, you know, and in his defense, or the defense that he would put up. And I actually, I've talked to him recently about the book. You know, he wrote that book basically started writing in the late 90s. And so he he did, he did say that he would revise a lot of it at this point. There are some things he's just totally disavowed. I think, just because, you know, you put hypotheses out there to test and they're tested. Right. And this is one issue where, you know, you tested one of his hypotheses. And so can you. Can you talk about that? Yeah.
I would also be curious to see which parts, Jeffrey dispels I think that actually, would be a really nice cultural change in science or psychological sciences people. Yeah, did that kind of thing, not only in conversations and hotel boss, but we're actually more explicit about it. My friend Julie Aurora actually recently added this project, together with tell your Kony, the loss of confidence project where scientists basically disavowed certain papers that they had published because they'd lost confidence in them over the years after this. Very replication, about reflecting reputation, all sorts of problems that, yeah, things that people may be considered best practices that are just weren't overly critical of had. And so some researchers actually came together and said, like, Well, actually, in this paper IP hack died. I didn't do this super well. And in hindsight, I don't really trust it anymore. And I mean, that's super helpful, because I mean, it takes a lot of time to replicate studies and to bad things. And so I think, yeah, people who have changed their mind can do a great service to fuel by actually going public about that. I mean, maybe Jeffrey has dominated I
mean, yeah, this is a private conversation. So I don't want to get into too much detail. But he did basically say that, you know, he's thought about basically doing a second edition of the book where he just talks about all the things that he no longer accepts, and what he's revised his views on. I mean, it's been, I mean, it's probably been almost 25 years since he started thinking about a lot of things. Because you I don't know if you know, I mean, I don't I, I've never written a book myself. But my understanding when I talk to authors is, you know, even before they start writing, they're assembling ideas. And so the latency is just incredibly huge from like, when the book is published to when they started thinking. So if you think about all of that, that how fast science in certain areas, moves, you know, so for example, a topic that I'm super interested in human evolutionary genomics, it's extremely difficult right now to write a book. Because the time cycle of writing a book makes the book out of date, by the time the book is published, and you can even talk to people there, it's exhausting to keep up with this particular field. So after 20 years, I'm assuming that, you know, there's going to be a lot of turnover in any given book. So your your paper,
I mean, I, I can totally admit to the fact that I was a little scared when I saw you put a paper I published in 2015, on the list of things you want to talk about today? Because that's such a long time and the paper is genetic. So yeah, I mean, I totally empathize with that. I'm really just saying, Yeah, I, I would love to hear what he's changed his mind on. And I mean, like, like I said, in the beginning, I wanted to kind of set the frame, I came to this topic as somebody who thought this film was interesting and worth testing. And, yeah, it's it's unfortunate if, if somebody tried to test the theory becomes seen as something adversary. Right, bigger and bigger. Obviously, I didn't control the results. Yeah. And talking about timelines, this is also actually a massively time lag project. We first conducted data for this years ago, I think, might have been 2014 or so. Oh, wow. So massive time lag, something I worked on, during the beginning of my PhD, it was actually somebody else's project to supervise in the beginning. And yeah, there were some things we weren't so happy about. Main thing was actually, when women read men's attractiveness, and the first version of our study, women just very negatively evaluated every month and attractiveness. So we didn't have a lot of variants to work with. And we were, we were afraid that this was would undermine our conclusions. So let me let me ask you
a question. Let me ask you a question. This, this sounds kind of familiar from a lot of the online dating data I've seen where women just do not rate the opposite sex very highly.
The biggest lesson or this project? Okay. Yeah, I mean, they have ABF data from OkCupid, showing kind of the same thing. So whereas for men, their judgments of attractiveness of women, it's, it's basically a bell curve, a normal distribution. But then on OkCupid, you can see that men excessively messaged the women in the Upper 10% or something like that, so they face also some lopsidedness there. But yeah, when you look at women's attractiveness ratings on OkCupid, at least you will also kind of see a sort of kind of threshold effect. So some men just don't work for them at all. And others are found sufficiently attractive to maybe getting in touch with Yeah, and so we sound saw something similar in our study. So basically, we had women raped men that I saw on video and, and a lot of them came away with a conclusion. Yeah, well, that person is just not attractive to me. And, I mean, that's fine. I mean, if that's the result, that's fine, but from a statistical perspective, of course, you you lose something, you lose information when when the when the range is compressed like that, when when many men get a value of zero. So basically, we will rephrase that item to Go from actually very repulsive to very attractive. And then we had a little bit more variation to work with. But kind of a side side note to all this, but maybe worth knowing?
Well, I mean, just so you know, I do want to for the listener, I do want to emphasize what you say, as a side note, I mean, this is a lot of science, in terms of just getting the data format, right, getting usable data doing, you know, just like making it so that you have something that you can draw inferences on. So there's a lot of a lot of work that goes into producing a paper that might not be even totally evident, or explicit. Even if you read the method. Sometimes the methods are I feel, having been through the process, sometimes I feel like the methods make a cleaner impression than was the reality. So I think it's good to actually talk about this.
Yeah, I agree. I mean, this diaper just was in the pipeline, and also kind of not been worked on for a few years, actually. So it's not all that kind of stuff. But but a lot of stuff is is really the nitty gritty work of making sure your assumptions are met, that the models run nicely, and just doing a lot of due diligence work, to see that you can actually trust your conclusions. But yeah, I mean, we were starting to talk about the details of this paper ready? I haven't really described it yet. So we should just do that quickly. Yeah, I don't know. Yeah. So what we contributed to that study, that was a team from getting around Lars pangkor. And the first offer from that team is, Julie dreeben. was a study where we had, we recruited a sample of men. The people who ended up in the study were 88 men and total in the end. And we try to sample a broad swath of men, but we didn't have a representative sample or anything. It's it's Yeah, we, but we tried not to just recruit in university students, because obviously, when the topic is the attractiveness of intelligence, you can do better when looking only at the upper end of the distribution. So we pick people up from the street as well. And, yeah, we had these men complete a bunch of questionnaires and and intelligence tests, first in an online study, and then additional tests in a lab setting. And from this data, we could extract the G factor and rank these men in terms of their tested intelligence. So the G factor stands for the general factor of intelligence. It's just a standard finding that I guess listeners of your podcasts know about that. Basically, our cognitive ability tests tend to correlate positively. So you can, you can summarize a lot of information in one number, maybe that's the number we used here. There's a general factor. And these men when it came to the lab and did a bunch of things on, on video. So when they spoke, they read out headlines. That's kind of the standard task from a personality literature that has been shown to be useful for assessing intelligence. So you have complex headlines, containing words like archdiocese and things like that. So basically, the listener can maybe tell a little bit about comprehension and, and vocabulary of a person speaking, and their vocabulary is actually a highly g loaded ability. And so these men did that. And they also will, were taught to make the female experimenter laugh. But it was kind of a freeform exercise. And what we had in mind was basically to sort of bring initial stages of romantic attraction into the lab, so to be not super far away from reality, but of course, to control the process much more to chunk it in, into into bits, basically. So in the end, what women saw when they were writing these men on intelligence and attractiveness, were slices of his behaviors and, and central analysis for the paper. This is a sample sample of women who first saw a picture of the guy. Then they heard his voice, just speaking of vowels. And then they saw him read the headlines, and when they saw an old joke, and after each step, they rated very attraction to the guy. So the idea was, yeah, we want to kind of separate any correlations that intelligence might have with things like style of dress or height or health, things that you might be able to see in the picture, and photo and theory, whatever behavioral information, but somebody is behaving intelligently. So it's not making mispronouncing obvious words is not having trouble with reading comprehension. This may be able to be funny also in the last task, and that, to isolate the effect of that, from things that might be correlations with other traits that could be engendered fruit fly atrophy, or other or effects of income and wealth on social status, things like that.
Well supplied by Adobe for the listeners, just like when one gene has multiple output effects. So genetic correlation, right.
So yeah, so that's what we did. In additional sub samples. Women also rated the same men in on intelligence and funniness and, and physical attractiveness. So, yeah, and what we could do with this data in the end was we could show Yes, women can, can tell whether somebody is intelligent or not, to a degree, it's not, not a huge accurate accuracy. But it's, it's a range of point free correlation, if you correct for the measurement error and the criterion, and habits that's consistent with what other people find as well for for these kind of thin slices paradigms, where you don't see a person for a very long time. And they do find the people that are being judged as intelligent, more attractive as well. So there's a correlation there. But there's actually not any predictive validity to the measure of intelligence. So when we take the measure of intelligence and have it predict attractiveness in this paradigm that we use with a staggered display of information, we don't see an associate association with attraction. So and minimal, optimal, significant ended and also very unlikely to be significant in a in a real way in a for an actual decision in the end, because it's a very small effect.
Yeah. So I guess the issue here is so exact, extrapolating from this, if you're not seeing that correlation, you can't have selection for intelligence, I guess a, you know, a question that I would have, though, is, you know, the human encephalization, or lineage kind of P kind of leveled off 200,000 years ago. So we could have a situation, we're just kind of at the limits of, you know, what is necessary? Right. So I mean, you could imagine a situation where in the past, there was selection for intelligence, but now we're smart enough. I mean, is that is that a reasonable conjecture? Your perspective,
I think it depends on what you're talking about. But I think it might be reasonable conjecture for our study in the West. In fact, one interpretation I draw from potential interpretation and refinement, make, we obviously do still have massive variation and intelligence of the human population today. And there are people who never learned to speak and they are intellectually disabled to a degree that is incompatible, incompatible with day to day functioning, and also, I think, fairly incompatible with being successful on the dating market. We did not have such people in our study here, right. So we were working with people who were in what you would call the normal range of intelligence. We, we didn't use a standardized normed IQ test for a study because we were running some of the tests online to reach more people. And your copyright makes it difficult. But we did have a few tests in the lab that have been normed. And we're, yeah, our best guess is that we have people in a range from 80, IQ points to 130 IQ points, maybe some were higher, it's always a bit difficult to tell with these simple tests. So they're not the entire range of intelligence distribution, and definitely not going into the range where what some people would call mild intellectual disability starts at 70 IQ points. And so yeah, this is something that I'm interested in. Is this basically a linear effect? Have we been are we boosting basically intelligence all the way and more is always better? Or is there this is good enough point, right? And in what I've read, I mean, a lot of people don't really focus on on this, but in what I've read on this, I've always thought it was kind of implicit assumption, but actually, it's linear. More than And we don't expect to see a nonlinear effect. And yeah, more more recent work in many areas has kind of led me to believe that. So yeah, I can see that maybe over the course of human evolution. That's what happened, we got good enough. And even though the differences in the normal range of intelligence seem pretty significant to us today, because of course, in our very education focused society, there's a big difference in what somebody with an IQ of 80 can achieve versus what somebody with an IQ of 120 can achieve. I mean, in terms of income, and things like that the differences are quite substantial. But from an evolutionary perspective, maybe that stuff doesn't matter. But yeah, I've the way I've seen it described in the literature was always well, maybe it doesn't matter now, right? Because we assumption going, going back, he is basically being well, we don't live in the ancestral environment anymore. We have social transfers, we have all these kind of things that mess with evolution, in the sense. And so what we observe now isn't what has been happening. And I think this is kind of a perspective that influenced early eugenics movement, right. So they were seeing negative correlations between number of siblings and intelligence. And they were worried that like, as a modern society, there was going to be a selective effect against intelligence, but they were, but their assumption was that this wasn't how it used to be. And yeah, I think that's basically the assumption that most people make nowadays, as well. So if our article provides something of a counterpoint to that, that would be good.
Well, so, you know, I want to move to the next paper, because I think this one, this next one got a lot of attention for years ago. But I do want to say, when it comes to intelligence, you know, from an evolutionary perspective, when you see a normally distributed trade that's affected by a lot of genes, you know, complex like architecture, you can often make the inference that strong uni directional selection isn't continuing on that tray right now, because or hasn't for a long time, because the genetic variance should be gone. Like if you're really strong selection, right? That's the stylized assertion. You know, obviously, it's a heuristic, you know, you can't prove it. And then there's balancing selection. So for example, when it comes to human body size and height, you know, I think it's quite quite obvious that there's balancing factors on size where like, for example, taller males probably have higher reproductive success, I feel attraction, but the human body, like your mortality rate starts increasing above a, you know, height six, three, and there's all these sorts of correlations. And so I think when it comes to a complex trait, like intelligence, I think we do run into this thicket of, you know, alternative, or like different forces that are pressuring it. On the other hand, you know, we just, we do have to admit that our brains as a lineage kept getting bigger for on and off for 2 million years until relatively recently. So you know, some people because I, oh, that's just like random, but I mean, this is ridiculous. It's metabolically expensive, 20 to 25%, of our caloric intake, you know, we have these big heads, we look super weird compared to other primates. So I think, you know, there's a smoking gun there. Now, how it affects the modern world is a little different, because as you said, there might be nonlinear effects here, where there's a lot of gains to some intelligence, and there's a lot but there's not that much gains to a lot more intelligence. And I think the reproductive distributions that I've seen today indicate that, you know, if you have extremely low intelligence, your reproductive value is not high, but then it rises up really quickly, even if you're well below average. And then there's not that much gains, perhaps there's even negative results if you go if you scale up the bell curve. So um, it's a complex issue that I think is, you know, that's why there's still people are still doing research, like you, and other people will do research and I do want to loop back to where we are with intelligence research. That's you did give a lot of thought to this about six years ago, you put out the paper, but I want to ask you about another paper. I'm kinda
because, I mean, this association with fitness that you mentioned, right, that with number of children, I mean, that's that's a thing that people have taken issue with, because obviously, family sizes are influenced not only by who's found attractive and and who succeeds in life, but also nowadays, by contraception. And by when people start having families, and I think that's the main reason people just never really treated this as very important evidence for or against ongoing selection on intelligence. Because they just assumed Well, whatever we see here is not is now very system dependent. And I mean, they are obvious reasons to, to believe that, right? I mean, like, if you stay in education for a long time, it takes a long time to basically get to that level where you have a stable income and a stable situation and I want to have kids and and since we attract people by intelligence in our education system, this kind of thing could nullify or even reverse correlation that you would see in a different society. But we, we don't generally have a lot of societies that we can look at, that don't have a tracking based education system. So that was the motivation, why we were going to look at attraction here, which is, of course, way more indirect recommend this, this lab choices that we looked at, or in the other study that we had the speed dating choices, is, of course, a very, very indirectly related to how many children these people will end up producing. But the idea was, well, maybe this hasn't changed that much, who is attractive and who isn't? Maybe that has changed less? When, yeah, how people nowadays who can control their fertility with contraception, and up with in terms of number of children. So that was the idea here. But in the end, I think it's it's, it's resulted actually consistent with best data I know for for fitness effects, which is from Sweden, I actually, I put this on Twitter right below the announcement for this paper. So we also see this kind of nonlinear effects. People who don't even take the men who don't even take the military intelligence tests, because they're, they have some sort of diagnosis that prevents them from being even conscripted, they tend to have fewer children, men who get very low scores on this test, they also tend to have fewer children. But after that, basically, there aren't that many differences even between, between below average people and above average people. But that's data from Sweden, and Sweden has very generous parental leave policies, they pay you to study all sorts of stuff. So in many ways, an unusual country. And in other countries, you do see a negative correlation between fertility and intelligence, although other countries again, usually don't have is really representative samples with the Scandinavian nations have because of a conscription. So yeah, it's it's difficult. But then you have these small positive associations with survival with longevity. But I mean, they basically don't matter in the big picture of the dwarf by the effect on fertility.
Yeah. And so speaking of fertility, you know, you have this paper that came out Proceedings of the Royal Society be older fathers have lower evolutionary fitness was for centuries and for populations. Can you talk about this, because this is like, I this paper, and this result actually did get a lot of attention, I think in the media, from what I can see, for obvious reasons, a lot of people are having kids at older ages in the developed world, and they really, really care about maternal effects. There's been increased, you know, correlation with things like autism spectrum disorder and whatnot, and possibly friendlier with older fathers due to the, you know, sperm, you know, mutational load hypothesis and sperm and whatnot, as you get older. So can you talk about this result and how you feel about it like?
Yeah, this was actually also a project motivated by my interest in weather. Intelligence is under a mutation selection balance. So I've done some earlier work work, actually my first paper, which actually the only media attention that got I think was on your blog, gene expression way back. So I didn't find an effect of paternal age on intelligence, in a twin study. Later work kind of went and sometimes confirmed that sometimes found a negative effect. So my notion back then was well, there had just been this con paper published with fontas, with very strong correlation with paternal age and number of de novo mutations that a child carries. And I mean, that had been I mean, it has been a known fact that because the male gametes keep on the progenitor cell, female genomic gametes keep on dividing, after puberty continuously, I think every 16 days, there was an understanding that this would lead to an accumulation of mutations with age in the germline in the male germline. But I think before it was called paper, nobody had actually compared the genomes of Mother, father and child and actually kind of imitations and put that in relation with the paternal age. And I think, at least that was my reaction. But I think that was a common reaction. We were impressed with how strong the association was. This is in part also maybe due because to to the maternal line, I mean, fathers and mothers ages are just heavily correlated. Obviously, people tend to have children with people of fairly similar age ages. But there was a strong correlation. And that seemed to make it possible to also use paternal age as a proxy variable for number Have mutations. And after I didn't see this effect of proton electron intelligence and some other much larger studies also hadn't seen it. I wondered, is this reason that mutations do not have such a big effect on intelligence differences in the normal range? Because I mean, we do know that a lot of different kinds of intellectual disability disability are caused by de novo mutations. But we weren't so sure about intelligence and differences in the normal range. And I didn't know Yeah. Is this just a result? Is this what we should believe? Or is this method just not feasible? paternal age just doesn't work as a good proxy of de novo mutations. And so I thought, What's the easiest prediction to make? Well, we do know that the average novo mutation should have a negative effect, right? I mean, most neutral, they just fall into somewhere into the junk DNA don't change anything and don't matter. And I mean, yeah, on occasion, mutation will also have a positive effect, but the vast majority of the ones that are neutral should be negative. With respect to fitness, right? So yeah, random change to a complex system should make it work less well, on average. And so I thought, Okay, well, that's very straightforward prediction. And also, there's lots of data available, where I have paternal ages and the the outcome and the number of children. And I started with a data set of British periods and the Royals, because that was, I could just order that on a CD online. But yeah, I ended up not using that data, it was really messy. And there were people who were supposed to be their own uncles or grandfather's stuff like that kind of time traveling people in the pedigree. And, also, I mean, it's the British period is not really a good population to use, if you want to make inferences about about a broader population. But yeah, but then basically, I've got access to three different kinds of datasets based on church books. So churches record, of course, births and deaths and marriages. And from that data, you can reconstruct pedigrees, and a number of researchers have done that, and also got access to the Swedish population database. The Scandinavian nations are famous for recording very diligently all the information about the citizens in a way that's easily linked, because everybody has one unique number. And so I had access to all the births and deaths and marriages in Sweden, and modern time, and also access to these church books from Quebec, from from the time of European settlement in Quebec. And also from the chrome horn region, that region in what's not, yeah, Bordeaux, Germany, Netherlands, and also, again, from Sweden, from various regions in Sweden, again, from church books, but from an earlier time, so basically, data on for populations and for centuries, and in each of his data sets, I could
count how many children somebody had had, how long they lived, and how old their parents had been at birth, at their birth. And because I did, and, well, I found a very consistent result. That children of all the fathers, when compared to their own siblings, had fewer children. This compared to their own siblings is very important. It's also an issue that's near and dear to my heart. Because if you compare between families or between sibling pairs, which is what genetic studies also, genomic studies also used to do in the beginning, when you don't get a clear causal inference, right? Because there's all kinds of confounds that can act at the between family level, so maybe richer people reproduce for longer, find it easier to remarry find it easier to find a new wife after the first one passes. And if you just look on a between person level, and there's so many compounds analysis, it's not even worth running, I think. And I think the effect is also in some of the populations positive on a between family level. But if you compare people to their siblings, well, lot of confounds kind of just drop out, right so the visit, obviously, on average, you inherit the same genes from your parents. And yeah, you can inherit a similar share of resources and share similar early environment. But yeah, that this paternal age differences remain. And yeah, what I found was that for every additional decade that father was older, depending on the population that they had a few percent fewer children. In terms of number of children, I think this varied from 0.2 to 0.4. Children. Yeah, so yeah,
well, I mean, so I'm looking at figure one right now. And, you know, you know, this is like a pretty consistent effect, as you were saying, in these three, or these four cultures, I guess. And it's, it's interesting. I mean, as you go really high up in age, the confidence interval increases and your sample size, obviously smaller, there's not that many 18 year old guys having children, right. So So what's really interesting to me is, um, you know, you're seeing, like, declined very much from the beginning. where it's like, you know, if you're, like, 20, you have more kids, or 30, you have more kids at 40. And that was a little surprising to me, because my intuition would have been that the decline really starts happening after your 30s. I mean, what what was your intuition coming ahead into those?
I mean, as psychologists, this was very unfamiliar territory to me, because I had to actually strong mechanism based theory. I mean, I knew the stuff about the germline I knew the progenitor cells, cells divide continuously they start in puberty. I mean, a more ideal comparison would actually be a paternal age, relative to when they enter puberty, but I didn't have data on that. And I mean, this doesn't vary so much. And yeah, I mean, it's a continuous process, you you start making sperm, and you keep on doing so and, and over time, the copy errors accumulate. And if I had seen a nonlinear effect, that would have been inconsistent with my mechanism that I anticipated. And I mean, what you're saying, I mean, maybe that intuition is more driven by what what we would expect to see on a between family level, right? I mean, I mean, some studies have looked at paternal age effects. Not in a sibling comparison, and in the way they do some often see that very young dads have children who have worse outcomes, which makes a lot of sense, right? I mean, if you have children at an age that's not close to the socially prescribed time, then maybe you are in a worse situation in many ways. And but maybe your children will then also be worse off in many ways. And so the sibling sibling comparison is really central to all this, right? Because this is something you haven't it's hard to formulate an intuition about that. Based on what you see around you, of course, right.
Now, what so you know, in, in the West today, especially among more educated, I mean, to be frank, the people that would be listening to this podcast, paternal age, you know, and I am, you know, I've had, I've had children and my second, let's just say that. So, I am not uncommon. Sorry, what's the takeaway for people looking at these data for the future? I mean, is there really a big takeaway, or is this just kind of academic? I think
it should not at all effect when you decide to have children? It's, I mean, on on a different level, on a family planning level, of course, all these contracts that I'm talking about, actually matter to you as a family, right? I mean, when you were in your late 20s, maybe you sperm had fewer mutations, but you also have less money, and maybe you weren't even with a partner you're with today. So waiting until you actually want to have kids makes a lot of sense. And these effects, I mean, they look large here, but yeah, it's, it's it's not a massive effect. From an individual point of view. I mean, it is, it is a real effect. I mean, if you look at specific diseases, like a pet syndrome, or achondroplasia, but we really understand quite well. Like diseases that occur occur almost. Yeah, congratulation. Maybe not anymore, but forget that syndrome almost exclusively occurs de novo. Yeah, there's an excess among all the all the deaths among the children of all the dads and you are risking that but of course, by having children at a much younger age, you're also risking maybe being with somebody, you end up finding quite low some later. And, and I think I think these things are problematic. More. It's keep hard, even real, really hard to real, better numbers. But I mean, I can I can I can put the effect on the on the fitness in numbers, but yeah,
yeah, well, I would say, you know, I would say that, you know, the individual choices can also different from social aggregates. So for example, let's think about, think about first cousin marriage, we're on the individual level doesn't have that huge of an effect like increases your recessive look to seize load of the offspring by like, you know, two to 3%. And maybe decreases the IQ, you know, less than a third of a standard deviation expected value, your kids are a little smaller, probably a little less attractive. Let's be frank. You know, going through Wilson's work using massive British data, there's a there's a modest effect of the offspring, right? But the key issue here is like what's happening at the social aggregate? So that starts adding up on the margin where it's like, okay, like, let's assume that you have 3% more recessive diseases in the society. I mean, that's, that's like non trivial amount of money in the aggregate. So things start adding up. And then you have fewer people at the tails of IQ distribution. Well, upper tail the IQ distribution, like
my response wasn't only academically relevant, right. And the thing on an individual level when you went to have children, I think you probably would bother to think about this too much. But I do, I do think there's other variables Saiful level what we are doing with the policies we set, which encouraged people to have children later, obviously. But at the same time, yeah, yeah. And also, quick note on the on the cousin marriage thing. Have you ever seen a study that has done that test, for example, runs of homozygosity? On a within family level? Yes. Because I mean, inbreeding, you know, what I would call it of course, with various variables that might also negatively affect all these things that we recognize us. I mean, obviously, heavy inbreeding is associated with clear health consequences that are undeniable. But again, these differences in the normal range that people have found only later research, do we know that this is not confounding? I mean, I would love to see that done again with the sibling data that is out now. Because they like these small effects on on height and and IQ, that could just as well be people being more rural or being more royal or whatever else, is correlated with inbreeding in the population.
Sure, sure. I'll, so I will actually get back to you on that, because we got Nelson, jF Wilson did a pretty extensive work on this. So I'll check that. So I want to I want to end our conversation, I want to go back to intelligence. Because this is a topic that you're interested in, obviously, and you wrote a review paper, I think it was in 2015 zeroing in on the genetics of intelligence in the Journal of intelligence. And, you know, a lot has happened since then listeners of this podcast will be familiar with educational attainment papers. You know, there's been a lot of work, I think there's going to be another educational attainment paper coming out of the social genomic Consortium. So this is just like marching along. I in in relation to what you said, in 2015. what you were thinking back then, how have things panned out?
I know, I mean, what I expected them and I think it says on the paper was that there would be progress now that g was became a dominant paradigm. I mean, it was time when in my mind for candidate gene studies was still fresh in memory. But I think this paradigm should shift to much more robust methods that correct for multiple testing, really broad field for an ancestor method available of availability of massive amounts of data that really brought the field forward. And so we saw some slow upward movement of explained variance and educational attainment and, and cognitive ability. I think the last number that is published, but for the direct genetic effects is I think 6% explained in a sibling comparison. Yeah, you can probably go a bit higher, and I'm expecting the next paper will go high stone. And yeah, my conclusion or it wasn't really a certain conclusion. But I what I saw kind of as a, I mean, that paper was mainly concerned with evolutionary genetics of intelligence, not so much with and I mean, I think evolutionary genetics can can let us predict much better which which approaches will work but Yeah, I mean, much of recent work on that has been published is not so focused on evolutionary aspects at all right. So, um, but like the takeaway I had in that paper was that perhaps intelligence isn't under such heavy mutation selection balance would, which would make it quite hard to identify genes using g was because, yeah, very large portion of the variance would have to be explained by rare variants. But g was typically typically don't capture. Um, and yeah, I think that has a better.
What's that? Let me let me, let me Yeah, let me let me review for the for the listener, because it's been, you know, this is like, second nature to you, and I know this literature, okay. So you know, we have this, we have this model, we have this model of, kind of combinations, variants are like non trivial frequency in the population, they have small effect, then there's other idea that there could be rare variants of large effect in families. That could explain a lot of the variance of intelligence as a trade normally distributed trade. The candidate gene studies that you're talking about in the 2000s, basically are, okay, there's this gene is a candidate for various reasons, maybe there was some, you know, pedigree analysis, some some classical analysis, leakage, study, pinpointing it, and you're just kind of looking to see if there's some signal there, the sample size were small, there's a lot of false positives going on, not necessarily p hacking, but probably file drawer effect. This became very obvious by the end of the 2000s. And so people move to huge sample sizes with better genomic technology with snip arrays. Okay, so we got that. And now we have millions and millions of samples. And with these millions and millions of samples, we're starting to detect associations between lots of Jesus. So there's 19,000 genes of the human genome, probably like on the order of like 100 million snips per individual in the human genome. And then, you know, of the common ones is on the order of 10 million, I think, you know, and so these are the ones that you're looking through, and a lot of the variants have been accounted for by these common snips. So basically, where we're at today, if I, again, I don't follow this literature as closely as you is, it does look like the idea that it was common variation, you know, in small effect, snips. So basically, only accounting for a small proportion of the variance is the dominant driver of normal variation in intelligence. The idea that there were bigger effect variants that are segregating at really low frequency, perhaps within families and lineages. And so they're dragging down the intelligence on the low end, that does not seem to have panned out. Is that correct? From?
I'm not sure I'm really uncertain about all this. And I'm actually curious, I was curious to hear what you think about it. And it's, I mean, the, the twin studies usually find heritability of around 60%, for intelligence differences. And the, the gcda approaches, which are based on your genetic relatedness on molecular level, tend to find lower coefficients yet tend to be around 30%, the recent study of the sibling g was that I mentioned they found an even lower number, which was surprising to me. Because as far as I understand that method, the number should have been higher, because you should, yeah, because of linkage disequilibrium being higher in amongst siblings, you you should get a boost and expand parents. So there is still a lot of variance missing, if you believe that twin study estimates. And yeah, I mean, some people recently don't. I'm not sure. Like, with I agree with him, I think the twin paradigm has held up pretty well actually. I don't actually think the genomic era has overturned. Much of its really if I'm being if I'm being if I'm looking at this at a very high level mean there has to happen these papers on the nature of nurture, but I mean, what what they found was effects are pretty consistent with the shared environment effects, which were observed before. So I think this 60% number might well be right. And I did have a paper out with a colleague with David Hill and Charlie here and a bunch of other people where we did look at whever rare variants. In the intergeneration, Scotland data explained a larger fraction of the variations and variation in intelligence. And we found that they did. So we had the minor allele frequency on the x axis in that graph and the explained variance on the Y, and there was a clear deviation from there, just parity. However, the paper has also been criticized, people have said, Well, on the Reverend level, you're more likely to pick up cryptic confounding, but you didn't get rid of fusion using the standard methods. So basically, I don't think my paper convinced all the relevant people, and I'm, I'm uncertain myself. But I mean, there is still this gap. And, yeah, it's, I think people can agree that the common variant estimate is at best, half of what between two imprints that is fine, but people doesn't realize the interpretation of it. And some people think, Well, actually, probably the explanatory power of really just the additive effect of genes is lower. And we're looking at more gene environment reactions, or just really overestimates and twin studies due to methodological deficits, I'm sort of more focused on the fact that the genomic studies that tend to only look at common variants, and also they tend to have just much poorer measures of intelligence, the twin studies of net is very extensive batteries, because the hard part was finding the twins not keeping them in the lab for hours. But in the genomic studies, like the UK Biobank for the hard part is actually getting to half a million people. And so you don't want to waste their time, because we're completing lots and lots of measures. So their intelligence measure is in probably measures intelligence, but that you can say, but it's, it's it's not very reliable. So yeah, yes, yeah, both reasons are really good. Dad might still persist, for some time actually. And my focus in this in the zeroing in paper, or my main focus, but but the point I wanted to kind of make was, well, we don't have to, we don't have to assume directional selection. We don't have to assume that intelligence is under balance between de novo mutations occurring constantly and directional selection for higher intelligence, we could also believe that it's under stabilizing selection, right. So very high levels of intelligence are also costly to fitness for some reason. And of course, people have looked into this, to some extent, to see whether they can find a conflict, same same sort of pattern that they see for intellectual disability, an intellectual disability, we know a lot of mutations that cause cause it and we do know we have is interesting findings, like, if you have a severely disabled, intellectual, disabled child, their siblings will be unlikely to be of normal intelligence. But if you have a mildly intellectually disabled child, where siblings are also likely to have lower intelligence, not as low but lower.
And so the group around glomming try to look at that whether we see a similar effect of a higher end, right, whether we're what we're also is large effect mutations, maybe. Yeah, really increasing intelligence. But where we would find people should look at very, very smart people that their siblings are most strongly regress to the mean, when for just the standard, slightly above average smart person. And they didn't find evidence for this, but I didn't find that study. Super convincing. Yeah. And I, I don't know. I mean, it's, I don't I see it kind of as my job to be a little bit skeptical about all the people who say, well, best, probably directional selection for higher intelligence going on. Because I think many of the studies that have been used to support that have these sort of flaws to do with confounding, like I mentioned, there are these inbreeding effects on intelligence in a normal range that have been reported. But this could easily be confounded, because so far people haven't checked between siblings in a sibling comparison. They also find this for height, by the way, right? So earlier, I was skeptical that was just simple, linear directional selection on height. They all they also find evidence for that using that method of looking at runs of homozygosity without a sibling comparison. So I don't know I'm a little skeptical that we know that there is this directional selection for higher intelligence in the normal range. I do find it interesting to speculate about whether with maybe also you have this downward trend at the upper end. I don't think that's commonly the model that people have in mind. I also obviously don't have a great way to tested. But but but yeah, I mean, you could do the inbreeding analysis a little bit better. You could do this approach with regression to what we mean. with with with data that's, that's, that's a little bit better at actually getting people if a very high extreme. So, I mean, you could, you could, for example, look at the siblings of extraordinarily smart people to see whether they are more strongly regressed. And I mean, there are people who have made these genius databases, right, but I don't, I don't think people often use the sibling approach in words really, really powerful. If you want to understand what is what's happening. It is
powerful. Yeah, let me let me let me make it clear for the listener, because I, you know, I also have been thinking about the sibling approach for me probably about 10 years now just thinking about it. So we have whole genome sequencing ability now where, you know, if you're a big lab, it's, it's much closer to $100, for a good quality sequence than $1,000. Now, we have the technology where we could put people through a battery of intelligence tests, who are siblings for a couple of hours, you know, do like a cheat, do it, do it cheap, or do a cheek swab? And, oh, no problem, that'll be edited out. So do a cheek swab. And, you know, compared to the whole genome sequences, we have the technology wouldn't be that expensive. So if there's a billionaire listening out there, perhaps email Ruben, some of his colleagues, I don't know. I mean, this is a feasible scientific program, scientific plan that we could do, we could test right now we, this is totally feasible, the main issue is always you need huge sample sizes, if the effects are subtle to you know, tease everything apart, but we can, we can sequence the whole genome, relatively affordably, that would not be an incredible, like, pain point in terms of the price. I mean, you could do snip arrays for less than $25 an array now. But I mean, really, I don't see the point, not just going whole genome sequence. So you can discover rare variants, and you know, what mutations? And then just get a bunch of siblings? And you're going to answer a lot of these questions like we could do this within the next year, if we wanted to some some, you know, rich person decided to get involved in a perhaps it's happening right now, I don't know, I don't know. I don't know if this particular effort, but it's not technologically logistically, you know, prohibitive at this point, it was 10 years ago, when we first started thinking about so you can imagine a situation where like, you know, you're looking at like, let's say, inbreeding effect to be concrete, there's going to be a variance in the runs of homozygosity, in the realized genomic inbreeding across siblings. And so to just gauge the inbreeding effect, you just look at the correlation between these genomic inbreeding, and, you know, decrease in the expected values across the siblings. Because, hopefully, I mean, they're controlling for a lot of the other effects naturally, their siblings, right, raised in the same household, a lot of the genetic interaction effects that you might be wondering about, like, it should apply to both of them, you got a big enough sample size? And we can answer these questions. So there's some areas of science where we're still scratching our head and how we're going to figure this out. And then there's other areas where if we wanted to focus on it, we could do it. And so I think this is one area, where we now have all the technology there. And it's about getting the scientific program there. And you know, as you're applying here, there's been over a century of work in psychometrics and other fields, with a lot of methods, a lot of hypotheses, a lot of frameworks. So we have the technology, we have the intellectual tradition, do we have the curiosity? Do we have the will? That's really well, so
this is within family Consortium, they have collected sibling pairs, they have 150,000 100, almost 160,000 siblings from 70 cohorts. I'm not a part of that. But that's a really cool project. They recently had a preprint arch. First office, Laurens, how that's really great. But I mean, the one thing with siblings is you have to have less variation to examine, right? Because so much so much as control for so you even but you need even bigger samples to get good statistical power and, and so you just need to go this number needs to go into the Middle Ages as well. And, yeah, this would be super super inflated, let me let me make it would be one of the biggest possible contributions to actually bring in the causal inference forward and actually generating really reliable knowledge that you can work with.
Yeah, let me make it again, concrete for the listener. siblings are about 50 or 50%. Related the expected value this is the theory, but because of you know, just like the medalion process, there's variants across siblings do their realize relatedness. So standard deviation, last I checked was about 3%. Like I have, I have two siblings who are only 41% related genomically, not 50%. So most people are going to be off the 50%. And there's a distribution, there's gonna be some sibling pairs to find that are below 40%. And so if there's that, quote, unrelated, you know, what we're proposing here is, well, if they're that uncorrelated on genomic relatedness, they should be, you know, proportionally uncorrelated in their phenotypic characteristics in direct relation to how much genetics is contributing to that, right. And so there's going to be siblings that are 60% related siblings that are 40% related. And what you want to do is look at this relatedness look at the correlation of characteristics. And a lot of the questions that we have are going to be answered by the natural experiment. And this is something I've discussed with people over the last decade, I just feel like in the 2020s, we have the we have an opening to finally push really forward on this, because we have the genomic technologies to basically estimate relatedness between everybody that we want to I mean, you know, these personal directed consumer genomic companies like 23andme, they have relatedness on 10s of millions of individuals. Now, you know, maybe a lot of these people don't have sibling pairs, but I mean, probably if you add up 23andme, ancestry, all these companies, they probably have over 10 million snips.
Yeah, quick question, just because, you know, you know, that number that two of your siblings are only 41%. Related, did you run the numbers yourself? Or do the consumer genetics companies actually provide that now? Because I mean, they totally showed, right, I both,
they do provide it, they do provide it. So I that's I saw the number I was shocked, and I double checked with the raw genomic data, and it was correct. So I know. So I know, for example, with my children, I know what proportion of their genome is contributed by. Okay, so I know the variance on that. I also know because they are of mixed heritage. I also know for example, like they have a they have a Norwegian great grandmother. And by looking at the percentage of Norwegian ancestry, I can actually figure out how much ancestry they have from their great grandmother. And I can actually figure out that one of my, one of my sons is extremely enriched for this woman who died in the 1970s. You know, so there's like some really, really deep, quiet and like, it's not a coincidence that his baby pictures look a lot like, like his great, great uncle who died in the 1930s. Like, I mean, he has his facial features or not, his facial features have like a bit, he basically looks like a tamped Norwegian. And so it's like, you know, in hindsight, I'm like, Okay, this makes sense. He's heavily enriched for this one woman, of his of his great grandparents. So this is going to happen with like, with variants, and so we can figure out these things. Now, if we wanted to, it just has to be like getting the getting, like this paper that you had in 2015, and have all these questions, we can answer a lot of the questions now, we see the path forward, and the issue is like, are we gonna go forward, you know, on this path, and there's no other topics you're talking about, like the paternal mutation, all of these things, we're gonna have like huge, huge piles of data, of genomic data as basically, my projections that I've done. It's not with an outside the realm of possibility that half of Americans would get whole genome sequenced by 2030. You know, that's hundreds of millions of people whose data is going to be out there. And so we need to figure out, are we going to get any value out of it? What are we going to do with it? And you know, in places like in Europe, where you have more centralized socialized medicine systems like in, in the UK, you know, the UK, UK is punching above its weight in genomics. It has it's done the same thing for Coronavirus. Same thing with the UK Biobank, like they really push forward in this area. So I'm pretty optimistic about our ability to know as much as we could, there can be like limits in terms of our understanding, just because of the nature of the questions that we're asking. But I think we're definitely moved. We've definitely moved a lot further in the last decade. And, you know, I think you've been part of that in terms of just like exploring these, you know, edge questions that maybe some people find a little bit controversial, although I don't know, I mean, paternal age is is shockingly controversial to me, and so far as a lot of people don't want their choices to have any negative consequences. You know, especially around play games. You know, it's just people are just so scared. Yeah. Children forever. Yeah, really, really heavy. dramatic. Exactly. But
I just, I don't believe I can work or really improved my son's future by how many books I have lying around and, or,
yeah, correlation. It's a correlation as a content as well. Yeah. What's up? Um, you know, as we, as we close out Reuben, like, what do you what are you looking forward in the next decade? What do you what projects like what are you excited about?
Well, I do hope to get back into some genomic work. I've recently not done any work in that area. But I mean, I've also been on parental leave. I am really, really excited about marrying my interest in causal inference and, and genomics, because I just think that's, I mean, it's so much more powerful approach than what many of my other colleagues in psychology have to offer. But I don't, I don't know if I will ever contribute much to the whole question of genetics of intelligence. It's a very crowded field with a lot of very capable people, maybe by its very nature. I'm yeah, I'm working increasingly on hormonal effects on sexuality. And it's also a continuing interest of mine. And by
the way, what do you mean by sexuality, sexual orientation?
No, I mean, sexual interest and desire, but also who you're attracted to stuff like that. I see. And I've been most mostly working with menstrual cycle data and changes in this sort of stuff. And I'm hoping to actually connect this to my interest in genetics, because that's not something that has happened a lot. Especially in when it comes to female sex steroids. Probably because there's not, I mean, the individual differences in the steroids are dwarfed by the inter intra individual differences. So variation across the menstrual cycle is much larger than variation from woman to woman in in estrogen and progesterone progestogens. But what would be interesting, I think, would be to understand why some women respond so much more strongly to changes in the menstrual cycle, but also to the pill. Which is, I mean, after all, the most common medication that anybody takes, yeah, and now, I don't know, I, I kind of feel like this might be a bit more easy to break new ground in this area, and also might be more relevant because it's linked to an intervention that we carry out almost without thinking. Yeah, and just it's a it's a cultural natural experience. And so I think that's really interesting. I mean, I don't necessarily any longer believe some of the more involved evolutionary psychology theories around that. I don't know the pill would change who you're attracted to. Or that this basically is turned on its head in the fertile phase of the menstrual cycle. So I think the recent replication crisis hasn't been kind of that idea. But I do think there's this really interesting effect on just sexual desire and all sorts of other symptoms. And, yeah, I would love to understand what's going on there. So
great, great. All right. appreciate you taking your time out of your maternity leaves, great conversation. No long long paternity leave. Okay, wow. Well, I'm sure you're catching up and you're busy. So I'll definitely let you go now. But it was great conversation. I really appreciate your work. For the listeners, you know, Google Scholar, they can check that out. And they'll probably hear from you. The type of people that listen to me are interested in the sort of work that you're talking about. So you'll probably show up again in the 2020s. And hopefully, this will be an exciting time for psychological and genomic research at the intersection between the two,
but also to the time it replicates. Okay. This podcast for kids favorite
