James Lee discusses recent progress in the genomic prediction of complex traits such as cognitive ability and educational attainment.
- Imagine a Future Without Sex: Reproductive technology may lead us to realize too late that being human is better than playing God
- James Lee academic web page
- Social Science Genetic Association Consortium (SSGAC)
- Nature Genetics: Gene discovery and polygenic prediction from a genome-wide association study of educational attainment in 1.1 million individuals
Steve Hsu is Professor of Theoretical Physics and of Computational Mathematics, Science, and Engineering at Michigan State University. Previously, he was Senior Vice President for Research and Innovation at MSU and Director of the Institute of Theoretical Science at the University of Oregon.
What is Manifold?
Steve Hsu is Professor of Theoretical Physics and Computational Mathematics, Science, and Engineering at Michigan State University. Join him for wide-ranging conversations with leading writers, scientists, technologists, academics, entrepreneurs, investors, and more.
Steve Hsu: Welcome to Manifold One, which is a continuation of the old podcasts that Corey Washington and I used to do called Manifold. We have been on hiatus for some time now, and we are kicking off the podcast again, except that, , this time it's mostly going to be me and less Corey.
Steve Hsu: Our first guest today is an old friend of mine and a very prominent researcher in psychology, behavior genetics, and computational genomics. His name is James Lee. I'm going to introduce James a little bit and go over his biography a little bit, because it's quite interesting. And then we'll get into topics like polygenic prediction of educational attainment.
Steve Hsu: He's part of one of the leading, perhaps the leading, collaboration in the world studying that. We'll also talk about his recent Wall Street Journal editorial about embryo selection and perhaps even cover some additional topics beyond that.
Steve Hsu: Does that sound okay for you, James?
James Lee: Sure.
Steve Hsu: Okay. I've known you for over a decade and I know a little bit about your background.
Steve Hsu: So I'm going to try to introduce you to our audience and just ask you to embellish or correct the facts that I reveal about your background. Nothing too embarrassing I hope.
Steve Hsu: Okay.
James Lee: Only the audience for this podcast will be interested in this, but yeah, go ahead.
Steve Hsu: Oh, no, they love it. I mean, I think people have a genuine interest in the life history of prominent scientists and intellectuals like yourself.
Steve Hsu: So, , you were born or you grew up in Torrance, California, is that correct?
James Lee: No, actually I, a bit tonier than that even I grew up in Palos Verdes, California.
Steve Hsu: Oh, Palos Verdes. Okay. Well, tell us a little bit about growing up as an Asian American Korean American in Southern California in that era.
James Lee: Well at that time, there were, there were some Chinese Americans, but fewer Korean Americans, Palos Verdes was not as Asian dominated as it was today. In fact, I guess we were sort of the pioneers there. My uncle lived there although later he moved to Orange county. Visiting my uncle once, my mom saw the place and said, wow, this place is beautiful. This is where I'd like to live. And that's, that's shortly after that, that's where we ended up.
Steve Hsu: Were you a serious kid in high school? A serious student.
James Lee: Oh by high school was totally different, by high school. Our high school was maybe 30% east Asian, some classes like calculus, BC, were virtually a hundred percent. And by then those numbers forced you to maybe think a little differently than I did in elementary school.
James Lee: I mean, the schools there had always been strong, although I don't think the school has much to do with it, but, , I mean, they're always been good students there. Let's put it that way. And, , and that's actually what explains the Asian influx that they, they go where, you know, the test scores are high and all that.
James Lee: And that's that, that that's how the place was transformed.
James Lee: Well, let's put it like this. In fourth grade, after fourth grade ended - I always remember that year, because that was the year that Dodgers won the World Series with Kirk Gibson, cinematic home run. My family moved briefly to Korea and so we were there about two and a half years.
James Lee: And then we went back to Palos Verdes and, and it was as if overnight the place had been transformed as large numbers of east Asians, although some other groups to Persian, south Asians, dominating all of the honors classes and so on. And I would say that there was a noticeable change in the character of the place from that point on.
James Lee: So you, you were a serious student and I gather you, you took a lot of STEM. We didn't probably use that word back then, but you took a lot of science and math courses in high school. But I understand when you entered Berkeley for college, you were an English major. Is that true?
James Lee: Yes. That's true.
Steve Hsu: So tell me what you were, what you were thinking when you entered college, what did you want to do with your life? What were your aspirations?
James Lee: I probably wasn't thinking much at all. I mean, it was, it was surely not, not a great decision for me to major in English. I'd always been an avid reader, so I thought that well, yeah, well, that's, that's a natural thing to do if you want to read is to major in English, study literature.
James Lee: , to some extent I did get that at Berkeley, early on when taking classes going over, like, you know, Chaucer, Shakespeare, Milton. But then later on, you know the classes became laden with all this, , , I would say pure nonsense. , you know, highly ideological. And also just plain silly. I mean, there were classes you could take about Tupac Shakur or the Godfather. And in fact, I became so disillusioned that I chose to graduate early, so I could just stop with this and, and actually took a year off or I just read on my own.
James Lee: So I read the things that I thought I should have been reading, but didn't get, did I get a chance to, or wasn't assigned? So I read, basically all the Shakespeare, Dickens, a lot of Tolstoy, Dostoyevsky, Chekhov, and sort of, sort of became an autodidact at that point.
Steve Hsu: So after Berkeley, you attended Harvard law school and was that a kind of classic situation where someone majors in the humanities and they're not quite sure what they want to do.
Steve Hsu: And so law school is always an option or were you passionate about some aspects of the law?
James Lee: The first? Yeah, it was not clear what to me, what I should do once I finished my undergrad. So, law school was where I went. I did become highly interested in the law once I arrived, because I did find it to be really interesting. So it was much more interested in what I was studying then, then earlier.
James Lee: So yeah, so yeah, and in law school, I was reasonably content, I suppose.
Steve Hsu: Now, at some point during law school, you, was it already in law school that you started getting interested in things like psychology and cognitive science, things like that.
James Lee: Pretty late in last school, , while I was reading things to prepare for a research paper that we're all supposed to write, that's when I started reading about cognitive science, behavioral, genetics, these kinds of things.
James Lee: In fact, one of the things I found most interesting was an account of the Minnesota study of twins reared apart. , this was a study that was actually conducted by my predecessor at the University of Minnesota Thomas Bouchard, who would find identical twins who had been separated at birth, or soon after, raised in different homes, later reunited sometimes reunited for the first time, actually at Tom's, lab, I guess, in Minnesota, in Minneapolis.
James Lee: And so he would bring these twins, to the university of Minnesota and then for a week or so he would just study the heck out of them, giving them all these tests, questionnaires. And basically discovered that even though they are not spent much time in contact, have been reared by different families, , that these twins did resemble each other often greatly as adults.
James Lee: In fact, the resemblance was not that much less than that between monozygotic twins. Monozygotic, meaning identical twins, having identical genomes. That these twins who have been reared apart were not that much less similar than twins who have been reared together by the same parents, usually their own parents. He also noticed these qualitative similarities that are not so easy to quantify. And perhaps that prevented some of them from being published. But he would notice that, for example, the first set of twins that he studied, the Jim twins. Recall the Jim twins cause by coincidence, they'd both been given the same name, Jim.
James Lee: , they both married women with the first, the same name. , they'd given their children the same names. Except one was James Allan with an a, the other was James Allen with E.
James Lee: They both work as volunteer sheriffs. They both had owned dogs named toy as children. They both installed circular benches around the trees in their front yards. , they both taken a vacation to the same strip of beach in Florida and missing each other by about a month. They drove there in the same make and model car, light blue Chevy. so it was actually reading about those twins that, and I guess it was a bit of a selection bias here, but, , that convinced Tom, Hey, this is pretty interesting. What I should do is crack down. As many of these twins I could find study them systematically. So that's what he did over the next 20 years, working with others, including Matt McGue, Nancy Segal.
James Lee: And this, this really struck me. This seemed to be a pretty important fact that genetics can exert this profound influence on, the way you think, your behavior. Pretty important fact, I thought for it, it never to have been, never been exposed to in high school and college and after college.
James Lee: And in fact, I became so fascinated by this, that I just, at some point I decided that I actually wanted to do science and not, and not law. I guess I was struck by the fact that these interesting questions about human nature could be answered by just systematic observation, measurement and quantification and, and theorizing.
James Lee: So, , that's what got me on a different path.
James Lee: Interestingly, I wasn't committed to doing behavioral genetics at first. I was actually sure what I wanted to specialize in. I even wrote down a list of things and sort of asked myself, well, what's the department or area that hits most of the things on this list.
James Lee: But as it turned out, I did go on to specialize in what drew me in, in the first place, which was behavioral genetics.
Steve Hsu: So I probably should have said when I introduced you that you are currently a professor of psychology at the University of Minnesota, and you just related to us, that is a storied department. So, I'm at a Big 10 university, the University of Minnesota is a Big 10 university. My father got his PhD there and my brother and many members of his family actually attended University of Minnesota by coincidence. So I just want to say, it isn't Harvard, but in this particular field of behavior genetics, these were the people that you mentioned were pioneers, who really drove the subject forward.
Steve Hsu: And it's, it's rather amazing now that you're a professor at Minnesota and you've actually in a sense taken the position that the spot in the faculty roster that was held before by Bouchard, who was really the leading pioneer in this area. So I should have said that when I introduced you, but I'm glad we got that out there.
Steve Hsu: Now when you decided then that Harvard law was not the place for you and perhaps something in related to behavior genetics or psychology was the place for you. How did you then end up doing a PhD at Harvard under Steve Pinker?
James Lee: Well, I decided that psychology was what I should get my PhD in because when I drew up my list of topics, it seemed like psychology was the one that hit most of them. So I applied to various psychology programs. I did not actually make it into very many of them, I think because precisely because my background was so odd, , what I'd majored in it.
James Lee: And so on. For example, I applied to the University of Minnesota but wasn't accepted.
James Lee: I was accepted to UCLA to do psychometrics. So psychometrics is a rather obscure, highly technical field having to deal with construction of tests, questionnaires, surveys, and the analysis of the data collected with those instruments. I actually find that a fascinating field. In the past there were some very impressive people like Frederick Lord worked in it. But I didn't really see myself as a psychometrician.
I felt that getting into psychometrics at that point would commit me to a certain specialization where at a point where I just sort of still wanted to be a generalist because I wasn't sure what direction I wanted to go. So I ended up going to the other place that accepted me, which was Harvard. I suppose, that, Steven Pinker didn't, wasn't put off by how this guy who'd like, seems like he'd been interested in all kinds of different things.
James Lee: And so he took a chance on me and I'm very grateful for that.
So, after Harvard you became a researcher and I think your primary focus was behavior genetics, which really is exploring the effect of genetics, individual genetics, on individual differences. So how people differ from each other in say intelligence or personality. Is that a fair characterization of what you're doing right after your PhD?
James Lee: More or less? I actually became, a postdoctoral researcher with, , with NIH, the National Institutes of Health. The unit where I was, I think it's changed its name over the years, today it might be called the mathematical biology something. I think back then it was called something different. Basically the people there.
They would just take up any question in biology, often having something to do with obesity or something health-related but sometimes not, not necessarily. Like they thought they could tackle, with quantitative methods, mathematical modeling. It turned out a lot of the things I worked on while there were of relevance to behavioral genetics.
James Lee: So yeah, I would say that, , what I did during my post-doc. I mean, some of it was not so specific to behavioral genetics, but it was, it was pretty seamless from my point of view, at least.
Steve Hsu: Amazingly, I know some of those people at a NH lab, which which is actually a story lab that, you know, has done a lot of important work in mathematical modeling of biological systems.
Steve Hsu: So from there you, am I correct, from there you went to Minnesota where he became professor?
James Lee: Yes, that's right. I think it goes to show the large role luck plays in, in academia. So Tom Bouchard, who I mentioned earlier, the twin researcher, he had retired in 2009. But I believe he was not replaced right away because of financial difficulties at the university.
James Lee: So that was held off on until for a number of years. Finally, when they looked for someone to fill that position, that was when I happened to be available in the job market. And in a lot of ways, it was the perfect fit. This very place, filling the shoes of the very man who inspired me to, in a way to get into this field in the first place.
James Lee: And so I applied for that job. I came for the interview. Tom himself, who was obviously retired emeritus, , himself, came out to, to, to watch my job talk. And I was fortunate to take the position.
Steve Hsu: But I think for those, that older generation of behavior geneticists. Who, I would say to, you know, to my satisfaction demonstrated lots of aspects about heritability of important human traits.
Steve Hsu: For them, what you're doing now is kind of a dream fulfilled where you're able to analyze large numbers of individuals, , with access both to their DNA sequence and to phenotype measurements or measurements about the individual differences. And so we're, to me we're sort of now in a kind of golden age for this kind of research.
Steve Hsu: So I imagine for them, for Tom Bouchard, it must be just, , extremely wonderful that you came along to carry the torch forward
James Lee: In some ways, yes. In other ways, Tom, and a lot of people from that generation are not happy about the things now, why maybe we can get into that later.But it's actually interesting, , at the time, when I interviewed for the job, there were a lot of people who were actually skeptical about genome-wide association studies. That is studies that aim to go beyond the classical twin and family studies, which show that yes, there is some genetic contribution to the familial resemblance and try to find the actual sites in the human genome at the DNA level, the G's, the C's, the T's, the A's where flipping from one to another will actually affect these traits.
James Lee: This was because at the time, the studies that have been done of, of intelligence and schizophrenia and various other traits and not yielded very much. There was a time when they would do these studies called candidate genes studies at the time, where in retrospect, pretty much everything was a false positive.
James Lee: And this left people scratching their heads. They felt like, well, we did a pretty powerful study. We had hundreds of people, maybe thousands, even in some cases, but it looks like we're not turning it up much. So what's, so what, so what is going on? There's something wrong with the twin studies?
James Lee: Now it was clear from studies of non behavioral traits, such as height. So those studies had found things, but everything they'd found was very small.
James Lee: So if you do a twin study of height, it will tell you that 80%, something like that, 80, 85, 78%, something like that, of the individual differences in height are caused by genetic differences.
James Lee: But then when they did the DNA level studies and they got the sample sizes up to the tens of thousands, and then at one point in 2010, 180,000, now they were getting some replicable results. But the results said was that, well, the reason why this is hard is because of the DNA variations in the human genome with the biggest effects on height, at least common, common variants, , common variant mean that if you have a site in the human genome where there's two variants, we call these alleles, say one, a wheel is G and the other is T something like that.
James Lee: And then one of them increases your height. The other decreases it. A common variant means that you're talking about a site in the genome where both alleles are pretty common. That is one might be 43% and the other might be 57% or 12 and 88, something like that. It can't be that one of the alleles is only carried by, I don't know, like a 10th of 1% of all the chromosomes.
James Lee: So anyway, these common SNPs, which are easier to study, it looks like there is no common snip in the human genome with where the plus allele has more than say an effect of a 10th of an inch per allele. That is whichever allele is the height booster, for each additional copy you have, it just increases your height by a 10th of an inch.
James Lee: Something like that. And those are the comments SNPs with the biggest effects. So people found this puzzling too. They said, can that really be true? That you need a sample size of tens of thousands to detect even the biggest thing. And what about the rest of the, the, the heritable variation in height?
James Lee: If that's the biggest thing, then there must be a lot of things that have smaller effects, and then you just kind of crunch the numbers and there must be thousands of those to explain why 80% of individuals with differences in height are genetically caused. So people found these numbers a bit mystifying science fiction. And so people were scratching their heads.
James Lee: And those are the kinds of questions that some of the questions I feel good when I was interviewing for the job back in 2012.
Steve Hsu: Well, that the state of affairs you just described sort of coincides with my entry into this field. And since you mentioned height, I can point out that just recently in the last, I guess, week or so, this massive study using 5 million plus individual genomes was done, which finally confirms the effect of common SNPs.
Steve Hsu: You described what common SNPs are. So common SNPs of that 80% say heritability of height, maybe 50% of that is from common SNPs. And now it appears that effectively all the common SNPs, which are at genome-wide significance, that's a technical term have been found, using such a huge statistical power, 5 million people. And there are about 10,000 of those.
Steve Hsu: And about four years ago, we had already done some machine learning where we were able to predict height using about 10 or 20,000 SNPs, at, , an accuracy of, I think the standard error was right around a few centimeters. So I don't know how, how widely this whole, , this, these advances have percolated into fields like psychology, but for people who are really following the frontier, I think it's now been demonstrated that even a trait, as complicated as height, which is controlled by tens of thousands of individual genetic variants has really in effect, been solved by massive statistical power and machine learning.
Steve Hsu: I think maybe now is a good time to turn to educational attainment and cognitive ability, which is, which is what you have worked on professionally at Minnesota. And obviously the cognitive ability of an individual, or for example, what level of educational attainment they're able to achieve in their lifetime.
Steve Hsu: Those are also highly complicated traits and probably, well, we now know, controlled or influenced by many, many thousands of individual genetic variants. Maybe you can tell us a little bit about the consorti that you work with SSGAC and what you guys have discovered. And I think that the most recent thing I've heard of is something called EA4, which is educational attainment 4. You're the first author, I believe, of the paper in Nature, which described EA3, which was the third big release.
James Lee: Nature Genetics.
Steve Hsu: Nature Genetics. Great. Yeah. So maybe just tell us what SSGAC is and then tell us what have you guys discovered.
James Lee: So the SSGAC stands for Social Science Genetic Association Consorti.
James Lee: So this is a group that started to come together. , well, when I was in graduate school, actually. The reason I came together was because, well, basically they could see that the developments and the study of, say diseases or anthropometric traits had shown that, well, for whatever reason, it is like that. A typical trait is affected by many, many, many sites with small effects. And to detect any of them with a high degree of confidence does require massive sample sizes.
James Lee: And so this consorti came together to pool data from various groups to attain the statistical power, to identify these individual sites in the genome effecting behavioral traits. It was decided that the trait to study would be years of education, because this is something that every biomedical study has.
James Lee: They always asked us, check off the box in years of education do you have. So, , the SSGAC published its first paper about educational attainment in 2013, getting three hits with a sample size of about 125,000 people.
James Lee: Now, this study was in my opinion, very well done. These results were very reliable. Nevertheless, I think there was some skepticism. it's it seems like a funny trait to study to get three hits with 120,000. Well, , well, anyway. Well, let's just say a GWAS with three hits often doesn't contain much information.
James Lee: Like if you could try to go on with, the number of hits that is SNPs, we call these variants SNPs [Single Nucleotide Polymorphisms], you can be very confident are associated with the trait is sort of an index of how much signal there is in the GWAS as a whole.
How well can you predict the phenotype in new samples. , what biological information can you extract from, , the results as a whole? So three hits doesn't really get you all that much. , that is, , that probably is telling you that there's not much you can do with the information that GWAS.
James Lee: So in some ways people thought this was interesting, but in others they thought it was maybe not so interesting. But then subsequently the consortium increased the sample size to, well, depending on how you count, but, , so let's say roughly 300,000 people. Now this led to the identification of many more sites in the genome.
James Lee: We could be very confident were associated with use of education. At this point, various tools that come along to calculate various quantities of interest. For example, , the quantity called the genetic correlation. So the genetic correlation between two traits just means the correlation between their heritable parts.
James Lee: So if you think of a trait as it has a genetic part, plus an environmental part, if we could lop off the environmental parts and calculate the correlation between whatever's left, , the genetic parts that that's just the genetic correlation.
James Lee: A genetic correlation is often a bit higher than the normal correlation between the two traits because often it seems that we can think of the environmental part is just noise that obscures the, the connection between the genetic parts.
By the time of our second paper about educational attainment, , we were able to calculate that that there are, , moderate to high genetic correlations between, , years of education, and IQ, and intracranial vole.
James Lee: Also some surprising ones like bipolar disorder. Thanks to a tool by Brendan Sullivan, Hilary Connie, and a number of other people, at least square regression. , we had a fair number of hits. , we could start to get into the biological basically the biology that the GWAS results point to.
James Lee: And I would say that this was a big milestone when we published this paper in 2016.
Steve Hsu: And EA3 was, what year? Was that EA3 that you were referring to?
James Lee: I was talking about EA2, as we call it. EA3 came out two years later. By then we had increased the sample size to roughly 1.1 million individuals.
James Lee: And yeah, I mean, it was pretty much more the same, if there's any specific findings of that paper you want to talk about?
Steve Hsu: Yeah, let me, let me just, because I think this the way we're discussing it right now, it could be a little bit abstract for the average person. So let me give a concrete result, which is from EA3, it's a figure from your paper.
Steve Hsu: And after I summarize the figure, I promise I will faithfully summarize the figure, but then after I do that, I'd like to discuss what you think EA4 would give for us in a similar figure, which is even better powered. And, what you think the consequences of this are for our next topic, which is the editorial that you wrote for the Wall Street Journal on embryo selection.
Steve Hsu: So I'm looking at a figure from your paper in Nature Genetics, the vertical axis is mean prevalence of college completion. Conditional on a particular score, polygenic score, which you can read out just from the DNA of an individual. You could say, what is roughly their probability of completing college?
Steve Hsu: And it's a validation calculation. So it's, you're using longitudinal data from cohorts called one is called add health, for example. These are cohorts that were assembled many years ago for different purposes, but the individuals in the cohorts have been tracked through their lives and their genotypes are available for analysis.
Steve Hsu: So they're not used in training the predictor. And so this is a validation of the predictor.
Steve Hsu: For people, right at the mean, so they sort of have an average EA polygenic score. It looks like their college completion percentages is typical for the population, maybe for their generation, it's maybe something like 20 to 30% completion of college.
Steve Hsu: If you're in the highest quintile. So say you are in the top 20% of the population, top fifth of the population for polygenic score. You have roughly a double, twice as large probability of completing college as the middle group. And if you're in the bottom 20% on the polygenic score, just reading off your graph, it looks like you have a little bit less than half the probability of completing college.
Steve Hsu: So these are rather striking in the sense that they're pretty much what we would say in physics, macroscopic effects, right? Not, not some little effect like, oh, I got one 10th of an inch taller or something. But if I told the mother that her child had a pretty typical for the population chance of completing college in her life, she might react one way.
Steve Hsu: If I said your child is twice, as likely as the average kid to complete college, they would probably be very happy. And if I said your kid has less than well, under half the probability of completing college as the average kid, probably the mother would not be very happy to receive that news. So I promise I'm, I'm describing the graph verbally, accurately because I have it right in front of me.
Steve Hsu: How do you think this will have changed due to EA4 or maybe even in the fullness of time where, , we have, you know, much larger data sets and other measures, you know, measurements on people like IQ scores, things like this. How, how much more powerful will this tool become?
James Lee: Well, first of all, I should point out that the polygenic score for years of education. So earlier I was talking about the effects of individual SNPs, individual sites in the genome, a single letter. Those tend to be small, but if you aggregate, every site that you have genotypes are imputed across the whole genome, hundreds of thousands of them, millions of them, the predictive power also aggregates.
James Lee: And if you have a pretty well powered GWAS, instead of these little things, like, you know, a week of education or a 10th of an inch, , you can predict whatever the trait is, height or years of education, pretty accurately. But it should be pointed out that user education is a funny trait in the sense that it does have a moderate large-ish environmental component. Well, every trade is somewhat environmentally influenced, which is to say that, , the genetics cannot perfectly predict it. , even education. That was a little different in that some of the environmental influence has a large systematics, you know, source. It's something to do with the parental influence.
James Lee: For example, if you look at comparing monozygotic twins who are genetically identical to dizygotic twins, who are just like ordinary siblings except born at the same time. The monozygotic twins are a little more similar in their educational attainment, but not by much, despite being a lot more similar genetically.
James Lee: So this shows that being reared with someone in the same household does have a large effect on years of education. , this has also shows up in adoption studies. So adopted offspring do resemble their adoptive parents in years of education. Although the resemblance is not as strong as a biological family without with siblings, likewise do resemble each other in years of education.
James Lee: That is if you're a sibling goes to college, you're slightly more likely to as well, even though you might not be biologically related.
James Lee: Again, this resemblance not being as strong as his biological families, but still present.
James Lee: Presumably the mechanism of this, as we say in science is that, you know, parents for higher in social class, income, education, IQ themselves, are somehow able to help their kids go to college. College or education being the one component of class that's probably the easiest to, let's say game or manipulate.
James Lee: And it seems that parents want to influence this and they do.
James Lee: So the likely result of this is that, the associations between, SNPs in the human genome and use of education is confounded. Suppose that you have some genetic potential due to innate ability or a willingness to work hard, to attain educational credentials, but you inherited that genetic disposition from your parents.
James Lee: So your parents, you know, more than average, more than expected by chance to share a similar disposition, they themselves might be highly educated, wealthy or what have you. And they will help you obtain more education. So in addition to this direct causal path from your genome to your own education, you have this sort of a loop going over the top from your genetics, your parents' genetics to their own status or achievements, which then help them help you.
James Lee: It's not a hundred percent clear that this explains, well, let me explain the phenomenon sort of, went into the spiel without making clear. What am I trying to explain here? What I'm trying to explain is that, if you try to use the polygenic score for years of education to make predictions within families, that is within the same biological family, if there's two or more siblings, we can ask the question does the sibling with the higher polygenic score also tend to get more years of education? , we can ask this question for any trait. We can ask you for height, IQ, disease risk for some disease.
James Lee: It turns out that for most traits it seems that, , that the answer is yes. The genetic score does predict which sibling will be taller or whatever, pretty well. Almost as well as, when you asked this question, not within a family, but just picking people from the population at random.
James Lee: So that shows that the GWAS of height say is not confounded in this way that I was just talking about. That is it's not the case that taller parents can make their kids taller through some environmental path. However, for years of education, when we try to predict education within families, there's a sibling with the higher polygenic score also tend to get more years of education.
James Lee: , the answer is yes, but there's a substantial attenuation of the predictive power. That is if you have two brothers say who has have some difference in their polygenic score, that difference does not on average correspond to as big a difference in their education as when you're comparing two people plucked from the population at random.
James Lee: The attenuation is pretty substantial. Something like 40%, 45%.
James Lee: I mean, we're not a hundred percent sure, but part of it's explained by a sort of meaning for years of education. The reason why I sort of mating attenuates this thing within families, the technical, , we can get into it later if you want.
James Lee: But the main reason we think we observe this attenuation is because of this environmental confounding that I just mentioned.
James Lee: If you have a certain genotype, this not only has like direct causal effect on your own education, but, there's this confounding with parental genetics and then parental phenotype, which then affects your own phenotype.
James Lee: So when we talk about, you know, the bottom 20% in the polygenic score, having a 5% probability of going to college and the top 20% of the polygenic score having a 50% probability, something like that, it should be realized that a lot of that difference is not due to the causal effect of the polygenetics of the genetic split, but it's due to this confounding.
Steve Hsu: I think study of this kind of confounding, which might be due to shared family environment, is super important. In our own work, we did this with many different phenotypes and traits. I think we had something like tens of thousands of sibling pairs who were raised in the same family.
Steve Hsu: And I think what we found agrees with what you just said. So for diseases and traits like height, there's very little difference in the amount of predictive power you have when comparing two siblings versus two random people in the population. For EA, for educational attainment, there's a pretty significant reduction in predictive power if the siblings also share the same environment growing up, versus two random people.
Steve Hsu: But for IQ, so a predictor, which is specifically trained on an IQ score, we found something kind of in the middle. So there was a reduction, but not nearly as much as for EA. And so the, the power to predict which of the two siblings would have a higher intelligence score, that most of that remained, even if the siblings had experienced the same family environment growing up.
Steve Hsu: Does that sound like what you guys find?
James Lee: Yes, we, a number of people and I have a pre-print, , you can find that on bio archive, , where we, for a number of phenotypes, about 25 of them, we do, well, almost exactly what you just said. We compare the GWAS results. Which are usually obtained in a population of unrelated people to special subset of GWAS results, where we're basically comparing siblings in the same family.
James Lee: For IQ, well, they give you an idea. We see that in the case of height, there's a reduction to 90%. So within family, you can predict roughly speaking, 90%, as well as, as naively, just, just taking the population GWAS results at face value.
James Lee: For years of education, and it's more like 55% something like something like that. And then IQ seems to be like in the middle, like maybe like 80, 82%, something like that.
Steve Hsu: Yeah. Your results are identical to what we, what we found.
James Lee: So, so IQ is a score you take on a test. They tally up how many you get right. Depending on the test, and maybe an hour to administer. Here, we do not see nearly the attenuation we see with years of education.
James Lee: And this is actually consistent with the classical twin and family studies that, that we don't see too much effects of family environment on test scores, IQ. You get it for years of education, but apparently not, not test scores.
Steve Hsu: Yeah. I think for me personally, I see this is very plausible because it seems easier for say wealthy parents or determined parents to buy effectively extra years of education for their kid.
Steve Hsu: But it is not as easy, despite what test prep people say to raise the cognitive score of your kid, maybe after a lifetime of effort, you could, but not just buying them a summer tutor or summer training course for the SAT. I don't think you get huge gains from that. So it's just much harder for parents to use their family environment, resources, to increase the test score, the cognitive test score, as opposed to buying an extra year of education.
James Lee: Yeah, that's right.It's consistent with that. Yes.
Steve Hsu: So let's shift now. Cause we've been, I've kept you already pretty long. So we got to get to the end of this interview, although I've enjoyed it a lot. And I, I of sometimes wish I could do Joe Rogan, like interviews, which lasts three hours, but I'm not sure my interviewees would put up with it.
Steve Hsu: So let's talk about your Wall Street Journal editorial. So you wrote an editorial in the Wall Street Journal, and I believe the title. I don't know if you chose this or...
James Lee: I did not.
Steve Hsu: Okay. The editor chose the title, Imagine a Future Without Sex. I have the article in front of me or opinion piece in front of me. So let me just give you a kind of summary of it and you, you can, you can fill in or correct what I say. So you start out by giving the specific example of IVF, embryo selection, using IQ, polygenic prediction of IQ as an example of what you might do.
Steve Hsu: And you say, oh, we might get a gain, , maybe five points in the next generation. Let's say, just ballpark, you had five embryos to choose from and you use the predictor and you pick the best one. And on average you get a gain of maybe five points and then you say, oh, in two generations, that would be 10 points, which is starting to get significant.
James Lee: Four is significant, but yes. Ten, ten even more.
Steve Hsu: Yeah. And you talk about the cost of this maybe $10 or $20,000 for an IVF cycle. Even if you have all the add-ons and all the kinds of genetic testing that, for example, my company Genomic Prediction can do.
Steve Hsu: And you imagine that this is a pretty appealing prospect for people to sort of get these improved IQ.
James Lee: Appealing to some.
Steve Hsu: Yeah. To some, at least. Our company, of course, I should point out doesn't do IQ related stuff. We only do health-related stuff and we can quantify how many, for example, extra years of expected life, healthy life that you get for your kids through embryo selection.
Steve Hsu: So that's another thing that might attract people. But I think then later in your essay, I think you're anticipating some very long-term consequences of what might happen if this embryo selection technology really becomes widely adopted.
Steve Hsu: Is that, is that a fair characterization of your, of your piece?
Steve Hsu: So I'm, I'm curious, maybe you could say the initial assumptions you make in the op-ed you think they're pretty reasonable. They're not speculative extrapolations of what's likely to be possible or likely to happen.
I don't remember exactly, but I assume that embryo selection you're picking best of 10, best of 12, something like that.
James Lee: , there was a paper in New England Journal of Medicine earlier this year by some of my colleagues, I was not part of it, that made the same assumption. I think, best of 12. I've learned that this varies a lot. That some women, some cycles, it might be best of four, but if you really try to max it out, you can do best of 30.
James Lee: I heard there's a lot of variation, but anyway, I assed best of 12. That the correlation between the polygenic score for IQ and actual IQs, like 0.23 or something. Anyway, I just took the R squared from the most recent published GWAS of IQ from 2018.
Steve Hsu: Right. So you, you didn't extrapolate into the future here. These are today's numbers.
James Lee: Yep. And then I just assumed that the parents pick the best of 12 or whatever it was.
Steve Hsu: Yup. Comment on that number of embryos people have to choose from. It really varies a lot partially because of embryo freezing. So it's been found that embryos frozen in liquid nitrogen work just as well when thawed out, , as fresh so-called fresh embryos.
Steve Hsu: And so women who are, for example, pursuing a career at Apple and they have this as a benefit as part of their job benefits, employment benefits. They can start freezing eggs when they're quite young. And so you will encounter people in the IVF industry who do literally they have 30 eggs and 30 fertilized, , when they're fertilized, they become 30 embryos. So that's not unusual, especially if women start early in thinking about their fertility, future fertility.
Steve Hsu: But let's, let's assume then that it's appealing enough that large numbers of people start doing this. You focused in the essay on some negative, potential negative consequences in the long-term.
Steve Hsu: Maybe, maybe you could elaborate on that.
James Lee: Well, I don't claim to have special expertise in predicting the future. Someone I know says I'm not good at predicting the future. It's hard enough to understand the present, especially nowadays. So I'm not really sure what people will do. Let me, let me lay out some possible timelines.
James Lee: , one is where, you know, some people do this, but a lot of others don't. And then, you know, what will happen is that the minority that does this, and they're likely to be somewhat strange people already. If it's just a small minority, and they do it consistently, like, it's, it's clear that within two generations, three, that there'll be very different from everyone else.
James Lee: I originally in my op-ed argued that focusing on IQ, but it doesn't really matter what trait you're talking about. Separation in multidimensional space, even if at any dimension, it's differences trivial and the overlap is huge. Nevertheless, you can have a huge separation in the full dimensions.
James Lee: Yeah, I mean, it's hard to predict because we seem to be getting away from the idea of like, well, whatever job we need to fill, find the best person. But as long as someone is needed to keep the show on the road and we're not happy to just let the wheels fall off the wagon, it seems inevitable that as long as its minority is not like numerically super tiny, 1% of the population, 2%, it'll be clear that these people are sort of dominating everything. Government, large corporations, academia, everything, everything. Everyone will come from this background. And it seems to me that at that point, that there will be very strong pressure on everyone else to follow suit.
James Lee: Another timeline, which might just follow the first one I laid out, but, , somewhat different, , is everyone does this. They, they, they realize, hey, it's, , you know, why not?
James Lee: So what are the consequences of this? I mean, you could imagine a timeline where this is made mandatory. The reasoning might be well, you're imposing healthcare care costs on the rest of us, if you don't do it. So this is the only allowed way.
James Lee: What would happen to that timeline? Well, almost surely total fertility will drop a lot. It's almost like it's like this pre-print I saw where they said that once they made car seats mandatory, people having started having fewer kids.
Steve Hsu: Right.
James Lee: You just make it more difficult and inconvenient. I don't know what it is about us humans, but it seems the drive to preserve ourselves, it's pretty weak.
Steve Hsu: So just, just to comment on that observation, if you made it, I don't know about making it mandatory, but if you made it free for young women to freeze their eggs, it's a little bit of an inconvenience, but, , it, they came just sort of socially accepted that people do it.
Steve Hsu: Then the downstream part of getting pregnant and having the kid is not really any harder. It could be even easier because it's easier to get a surrogate mother. If you're already working frozen eggs. So I'm not sure that that the consequence for overall fertility wouldn't necessarily be that great, just because of, because of these work arounds.
James Lee: Yeah. Maybe. Well, anyway, another consequence is that. Well, generally we see in nature that if something is not needed it tends to disappear. One example is, , I is in cave, dwelling animals, fish, reptiles, amphibians, , they frequently lose their eyes entirely.
James Lee: Now this is not intuitive to people. I've discovered. People are surprised to hear this, but it turns out that if you're in a situation where a function that had been all important before, for example vision, is no longer needed, a mutation knocking out that function can rapidly sweep through the entire population.
James Lee: In fact, reaching selection coefficients of five or even 10%, it's estimated, that is that a little frog or fish or whatever that can't see as a five to 10% advantage of survival and reproduction, then those who still can. With that hugest selective advantage, it's just quickly all over.
James Lee: I mean, the, the principle is that if there's something you don't need, like not nothing is ever like neutral. Like nothing is ever like a different whether you have it or not. That, that, that, that, that is a point of zero measure almost. So something is either helping or hurting. In the case of vision, it might be that nutrients resources that you need to revert to, , fluids and so on, maintain your eyes, to keep your eyes free from being a source of infection, that this does now all cost and no benefit.
James Lee: And so, , your eyes are selected, I guess. Now the example of losing vision might also involve migration. It might be that the animals that still have some residual vision, I can see sort of the light at the end of the tunnel.
James Lee: So they leave, , and those who don't, they stay. But even if the selection coefficient is down to 1%, that's the order of magnitude reduction from 10. Yeah, that's still enough to completely knock out the function.
Steve Hsu: Yeah. I think the general observation, that functional traits like vision, you know, they're expensive.
Steve Hsu: You have to maintain those organs. And even at the genetic molecular level, you have to maintain your genome against mutations that, you know, would weaken your vision. And once the selection pressure against those things is gone, then you would expect over time, entropy would take over and the function would be lost.
Steve Hsu: I think your inference is completely right there. Here, maybe we're talking about fertility or sperm motility, or some aspects of reproduction that once you people start using IVF a lot or other methods of artificial methods of fertility, that some of these basic functions that are so important now, maybe start to be vitiated by this entropy.
Steve Hsu: But I do want to point out that you were talking about very long timescales here, at least in terms of human generations. And my feeling is this embryo selection thing, it may predominate for a generation or two at most, because progress both in multiplex gene editing and in our elucidation of genetic architectures are moving so fast.
Steve Hsu: That it's hard for me to imagine that 30 or 50 years from now embryo selection, won't be considered quaint old technology and active editing of embryo genomes. Won't be a very advanced and safe and effective procedure. And so well before any longterm evolutionary consequences from embryo selection come along, you'll have a whole different, much more formidable set of technologies to think about.
Steve Hsu: So that's kind of my perspective on this. I'd like to hear your reaction.
James Lee: I think though that even if has come along that, don't reduce, the conceiving of babies to, you know, here here's some porn, here's a cup, fill it up big guy. I think we still have to think about, well setting off a crazy arms race. Well, for example this might sound pretty selfish, but, I kind of enjoy doing science and I get a lot of meaning out of that. , not, not everything, but a lot. This kind of thing has the potential to like basically make you superannuated and worthless within your own lifetime.
James Lee: I mean, for example, if you could just flip SNPs at will you take a embryo that's already been deceived, at some early stage of the pregnancy, just sending the nanobots box or something to flip everything you want.
James Lee: This is also so that people find it unintuitive. But if you could do that, that is flipped positions at will, you know where all the consultants are. You could bring out about changes of like 10 IQ points, a hundred, 500, 800. It might seem like I'm pulling these numbers out of, out of out of my butt, I kind of have, because once you're way out there, it's not clear what these numbers even mean. But you could easily create a human being, who was as far above Newton, as Newton was above an ordinary person today.
James Lee: So I think about, will happen if this, this power is delivered into, into human hands.
James Lee: These arms races scenarios you can think of. I think in my own children who are not, as far as I can tell super geniuses, they seem pretty within the normal range to me.
James Lee: Yeah, I, I would, I would ask people to think carefully about just doing this because for the sake of something, because you want to see something exciting happen within your lifetime or whatever.
Steve Hsu: Well, I agree with your scientific analysis of what's possible. And I understand the concerns that you're raising.
James Lee: Worrying about the average IQ of our country or in the world getting lower. That is a serious thing to worry about. But we have, we have a means of taking care of this problem. It's the tried and true method of just having more kids.
Steve Hsu: Well I don't know if that helps your average.
James Lee: Well, if you and I had more kids that would definitely help the average, , I, whatever I say, this people react with incredulity. They say, ah, yeah, well, good luck with that, James.
But there are other reasons why fertility needs to increase. I don't think it's going to be politically possible for a shrinking base of young people who don't identify much with the older generations of their country, to support this top heavy age structure.
James Lee: I mean, it's true. It would require what Charles Murray in a book called a Moral and Spiritual Revival. But, , it seems to me that we, we sort of need that anyway, besides this problem.
James Lee: I mean, we look around and we see that our society is becoming increasingly dominated by not virtue, but the it's opposite. Everything is being corrupted by constant, unremitting deceitfulness, cowardice, which are then used as covers for outright malice and greed. We're basically running ourselves into the ground.
Steve Hsu: Well, let me just throw out. So I think we're, we're way over our time allotment here, but let me just throw out one more thing, which people, an aspect of this discussion that people often introduce, which is that if you talk to AI researchers, they think the timescale for things like AGI, even the more concise, I would say, not the super pessimistic ones, but the ones that are more pessimistic than the average AI researcher would say 50 to a hundred years is a plausible timescale for producing general AI, AGI human level or superhan level AGI.
Steve Hsu: And so the point has been raised that if humans are to remain competitive, with these machines that are coming just in a few generations, probably, , maybe we need to improve ourselves. And maybe, even though these future genetically engineered humans may be somewhat different than me, I might still prefer their survival to a civilization that's completely dominated by machines. Yeah. Do you have any thoughts on that?
James Lee: A lot of people are thinking and talking openly about that, because of various taboos and the sort of orthogonality of this to certain political issues. No one seems to want to talk about what we're talking about now or, or if they have an opinion they're not willing to talk about.
James Lee: So, for now, yeah I'm willing to leave the AI problem to others.
James Lee: I sort of sort of feel like it boils down to this. Do you, like, want to be human or not? I think it boils down to that.
James Lee: Like for example, it's, I mean, AI and genetic engineering have the potential to bring about so much change so rapidly that, you know, within, within the short time, you know, people that are older are not going to recognize anything they see around them.
James Lee: And there there's some people who embrace that. To me, I feel like, well, all these things, people talk about some, some ideas that are implicit in them like, you're only, you're only really your brain really. Or not even all of your brain. All you are is your intellect. So all these features of human nature that are not so intellectual, perhaps, some people would see them as encbrances, things we should happily husk off and leave behind.
James Lee: Me, I sort of feel like, , as tough as being human is often I sort of feel comfortable in my own skin. Now change is inevitable. We're not the same as we were a million years ago or even 10,000 years ago. So change happens. But I feel like letting nature do it. The old fashioned way is best for that kind of reason.
Steve Hsu: All right. Well, I feel like you've articulated your perspective. For this new podcast, because the old one was localized within the university we deliberately did not allow comments, for example, on the YouTube videos of the earlier podcast episodes, some of which have like 50 or 100,000 views now.
Steve Hsu: But I anticipate for this new podcast, which isn't really associated with the university at all, I anticipate allowing comments. So I really look forward to seeing the comments and how people react to all the interesting things you've said today, James.
James Lee: Okay. All right.
Steve Hsu: Thanks a lot. Thanks a lot for your time. You've been very generous.
James Lee: All right.
Steve Hsu: Okay. Bye.