I'm a software developer by training with an interest in genetics. I am currently doing independent research on gene therapy with an emphasis on intelligence enhancement.
Posts
Wiki Contributions
I'll give a quick TL;DR here since I know the post is long.
There's about 20,000 genes that affect intelligence. We can identify maybe 500 of them right now. With more data (which we could get from government biobanks or consumer genomics companies), we could identify far more.
If you could edit a significant number of iq-decreasing genetic variants to their iq-increasing counterpart, it would have a large impact on intelligence. We know this to be the case for embryos, but it is also probably the case (to a lesser extent) for adults.
So the idea is you inject trillions of these editing proteins into the bloodstream, encapsulated in a delivery capsule like a lipid nanoparticle or adeno-associated virus, they make their way into the brain, then the brain cells, and the make a large number of edits in each one.
This might sound impossible, but in fact we've done something a bit like this in mice already. In this paper, the authors used an adenovirus to deliver an editor to the brain. They were able to make the targeted edit in about 60% of the neurons in the mouse's brain.
There are two gene editing tools created in the last 7 years which are very good candidates for our task, with a low chance of resulting in off-target edits or other errors. Those two tools are called base editors and prime editors. Both are based on CRISPR.
If you could do this, and give the average brain cell 50% of the desired edits, you could probably increase IQ by somewhere between 20 and 100 points.
What makes this difficult
There are two tricky parts of this proposal: getting high editing efficiency, and getting the editors into the brain.
The first (editing efficiency) is what I plan to focus on if I can get a grant. The main issue is getting enough editors inside the cell and ensuring that they have high efficiency at relatively low doses. You can only put so many proteins inside a cell before it starts hurting the cell, so we have to make a large number of edits (at least a few hundred) with a fixed number of editor proteins.
The second challenge (delivery efficiency) is being worked on by several companies right now because they are trying to make effective therapies for monogenic brain diseases. If you plan to go through the bloodstream (likely the best approach), the three best candidates are lipid nanoparticles, engineered virus-like particles and adeno-associated viruses.
There are additional considerations like how to prevent a dangerous immune response, how to avoid off-target edits, how to ensure the gene we're targeting is actually the right one, how to get this past the regulators, how to make sure the genes we target actually do something in adult brains, and others which I address in the post.
What I plan to do
I'm trying to get a grant to do research on multiplex editing. If I can we will try to increase the number of edits that can be done at the same time in cell culture while minimizing off-targets, cytotoxicity, immune response, and other side-effects.
If that works, I'll probably try to start a company to treat polygenic brain disorders like Alzheimers. If we make it through safety trials for such a condition, we can probably start a trial for intelligence enhancement.
If you know someone that might be interested in funding this work, or a biologist with CRISPR editor expertise, please send me a message!
EDIT: The full post is now up
Oh boy do I have a response for you.
I think it may be possible to significantly enhance adult intelligence through gene editing.
The basic idea goes something like this:
- There are about 20,000 genetic variants that influence fluid intelligence
- Most of the variance among humans is determined by the number of IQ-decreasing minor alleles someone has.
- If you can flip a significant portion of those IQ-decreasing alleles to their IQ increasing counterparts, you can likely significantly increase someone's intelligence
- The effect size is going to be smaller than it would be if you made those same edits in an embryo because some of the genes you're targeting are only active during development. But my best guess at the moment is that we would still expect a gain of several standard deviations. However I am not very certain about this because I have not yet gotten access to SOTA genetic predictors of intelligence.
There are a million little details to get into, especially those related to the delivery of an editing vector, avoiding a negative immune response and avoiding off-target edits. But after researching this with a couple of collaborators for the last month and a half, I am starting to think this is going to be possible.
What's more, there are already several clinical trails underway right now that plan to use the same gene editing delivery platform that I have in mind for this kind of adult intelligence enhancement.
IF one could get this protocol to work, the actual experience of the procedure would be kind of magical: you'd literally get an intravenous injection (and possibly some medication to temporarily suppress your immune system) and your fluid intelligence would improve by a couple of standard deviations within about a week. I suspect it would take further months to years for the full benefits of the change to become clear, since crystallized intelligence is what really determines outcomes.
It's difficult to predict how long it will take to roll out something like this in an actual human trial, but I think it's plausible we could have something working within 5 years, which might be soon enough to significantly impact the trajectory of AI.
I'm working on a longer post about this, so I'll ping you when it goes up.
1
1OpenAI's continued practice of publishing the blueprints allowing others to create more powerful models seems to undermine their claims that they are worried about "bad actors getting there first".
If you were a scientist working on the Manhattan project because you were worried about Hitler getting the atomic bomb first, you wouldn't send your research on centrifuge design to german research scientists. Yet every company that claims they are more likely than other groups to create safe AGI continues to publish the blueprints for creating AGI to the open web.
Is there any actual justification for this other than "The prestige of getting published in top journals makes us look impressive?"
TL:DR: If you're female you should consider freezing your eggs and if you're male with a female partner you should consider talking to them about freezing their eggs. You should probably do this regardless of whether you want to wait for the technology to improve. The process will cost about $40k-50k for the first kid with today's prices, and probably $10k/kid after that. The benefit will be at least a year or so of increased life expectancy per kid, a decrease of heart disease, diabetes, and various cancers on the order of 10%-40%, and possibly increased IQ of somewhere between 0 and 10 points even if you don't directly select for it (due to positive pleiotropy).
Here are some more details:
A BASIC PRIMER
So right now we have a bunch of Genome Wide Associate Studies (GWAS) that look at single letters in the genome and how strongly changes in those letters are associated with some trait of interest. These GWAS can usually explain 10-15% of the variance in a given trait, with some notable exceptions such as height, where we can explain >40% of the variance.
I think the two potential benefit of waiting to have kids would be seeing an improvement in the percentage of variance explainable by polygenic scores and having a broader set of traits from which to choose.
WHAT IS AVAILABLE NOW
The only company I know of actually offering polygenic screening available to the general public is Genomic Prediction. Their trait panel is entirely focused on common diseases like heart disease, cancer, diabetes and a couple of others. Let me first give a summary of the cost-effectiveness of this type of "disease reduction" screening.
The implied "variance explained" by the reductions shown in their genomic index is actually quite impressive for some of these diseases. Let's use their original preprint from here: https://www.mdpi.com/2073-4425/11/6/648/htm
I used Carmi et al's code from "Utility of polygenic embryo screening for disease depends on the selection strategy" to estimate the implied variance explained given those reductions and come up with predictors able to explain about 40-50% of variance for Type 2 Diabetes, Heart Attack and Coronary Artery Disease, and slightly lower for Hypertension and the others.
Those are very impressive numbers. Most stand-alone predictors explain less than 15% of variance. This implies that either Genomic Prediction's numbers are wrong, or there's something really amazing going on in genomic indexing: somehow selecting against multiple diseases is straight up better than selecting for a single disease, even if you only care about a single disease.
Part of this might just be a result of sample size: when your coronary artery disease predictor is trained on one population and your hypertension predictor is trained on another, there's probably some kind of pooling effect going on. But given that most of the data for these predictors seems to come from UK Biobank, there's also a more profound implication to the reductions shown in their panel: it seems likely that most of these clinically distinct disease are all manifestations of some underlying "health factor", and that health factor has a strong genetic basis. Some genetic variants increase your risk of many many diseases. If that was not true, you would not see simultaneous reductions of this size across so many diseases. And my bet is there are reductions to diseases not even shown on the panel. What a crazy thing to discover while researching a LessWrong post reply. This is probably worth a whole post.
EDIT: I found a study that replicated the strong positive pleiotropy effect shown in Genomic Prediction's index: https://www.researchgate.net/publication/323614487_Improving_genetic_prediction_by_leveraging_genetic_correlations_among_human_diseases_and_traits
"For six additional phenotypes in the UK Biobank data, we find increases in prediction accuracy ranging from 0.7% for height to 47% for type 2 diabetes, when using a multi-trait predictor that combines published summary statistics from multiple traits, as compared to a predictor based only on one trait."
This is actually incredible. My interpretation is that there's not only a general factor g for intelligence across cognitive tasks, but also an h factor for health across multiple diseasees
Anyways, the implication for you question here is that current DISEASE predictors are already very very strong. Explaining 40-50% of variance from a predictor is incredible. That's probably getting close to the limit of heritability for some of these. So for heart disease, diabetes, and some types of cancer, we're probably nearing the limit of what polygenic predictors can explain and there is not much point waiting for them to get better. Right now you could probably simultaneously decrease the risk of many of these diseases by 70-80% by selecting among 10 embryos.
WHAT ARE THE BENEFITS OF WAITING?
Disease predictors are not nearly as good for non-European populations. I believe they the second best predictors are for South Asian, followed by east asian and then African. If you or your spouse trace your primary ancestor to one or multiple of those groups, it makes more sense to wait. Predictors for those of African ancestry in particular have substantial room for improvement.
The second caveat is about selecting for non-disease traits. This community has expressed particular interest in selecting for intelligence, though there are obviously other non-disease traits such as conscientiousness or mental energy that are also important.
There is substantial room for improvement in our intelligence predictors. Right now you could likely pay a PHD student <$10,000 to construct an intelligence predictor for you based on the Education Attainment Study #3 that would probably explain about 20% of variance in intelligence. If you had 14 euploid embryos to choose from and 70% of those implant, you would expect your first child to have an IQ about 4-5 points higher than the average of you and your spouse/partner.
Steve Hsu, one of the leading researchers in this field, has estimated that we would be able to explain 50-60% 30-40% of the variance in cognitive ability if the UK biobank simply offered their existing intelligence test to the 90% of BioBank participants who haven't taken it. That would raise the expected IQ gain from selection among 14 embryos to ~9.5 points, which would perhaps be worth waiting for, though it's not clear when or even if UK Biobank will do that. And since most of the biobank participants are European, the benefit might be somewhat smaller for other ethnicities.
So if you and your spouse are both European, you used normal IVF with multiple rounds of egg extractions and improved predictors would be a gain of about 13 IQ points. And since you probably wouldn't select exclusively for IQ (disease are important too), I'd guess a more realistic gain would be about 10 points.
Also paying that PHD student to make the intelligence predictor might get all research into the genetic roots of intelligence banned, so consider that a major possible downside. Though if it wasn't banned you could distribute it to anyone who wanted it and everyone doing IVF could have children 3-10 IQ points above their parents.
Then there's the question of all these other important traits that we don't even have predictors for, like conscientiousness, mental energy, performance while sleep-deprived and whatever else you value. I haven't researched these other traits in depth too much, but it seems like there's a lot of other important stuff that fall into this bucket.
Here's a GWAS looking at neuroticism that found 190 genes associated with the trait at 2.5*10^-5. https://www.nature.com/articles/s41598-021-82123-5#Sec2
Funny anecdote from the study: the associated genes were found to modulate behavioral response to cocaine. The authors don't say what percentage of variance is explained by those 190 genes, but my guess is it's in the neighborhood of 5%. So if you waited 5 years to have kids, these predictors of personality traits would almost certainly improve, probably to somewhere between 15% and 40%.
I can't find a single GWAS on mental energy. Why has no one looked into that yet?
A similar improvement is likely to happen for many of the other predictors, particularly those for which people have already done GWAS.
Of course there's one more question you'd have to answer even if you did have great predictors: which of these personality traits should be selected for and how strongly? All else held equal, more intelligence seems to pretty much always be better, and high disease risk seems to pretty much always be worse. Of course you can't necessarily hold all else equal when selecting a for a finite set of traits, but most of the literature I've read about plieotropy suggests that unless you have extremely powerful selection techniques (i.e. iterated embryo selection, gene editing or whole genome synthesis), these are unlikely to be a concern.
But with personality traits I don't yet have a clear mental model of which traits should be selected for and how strongly. I think most parents mostly want to give their child a happy productive life more than anything else, and besides the no-brainers like reducing predisposition to depression and anxiety, it's not entirely clear how to do that.
WHAT SHOULD I DO?
If you would be willing to pay ~$40k to substantially decrease your child's risk of common diseases and increase their lifespan by ~1 year, you should consider doing freezing eggs and doing IVF. And if you're not ready to have kids yet or you want to wait for polygenic predictors to improve, you should freeze your eggs (or talk with your partner about freezing their eggs).
Why freeze eggs? Well unfortunately a woman's production of chromosomally normal eggs gets substantially lower with age. The percentage of eggs that will be "euploid" (chromosomally normal) first increases in the late teens and early 20's before reaching its max around 25. It then slowly declines starting around 30 and really accelerating after age 35. By the early 40's, 80%+ of eggs produced will be aneuploid. The more euploid eggs available for freezing, the bigger a gain you'll get from polygenic screening.
A woman's capacity to actually carry a pregnancy to term on the other hand, lasts well into the post-menopausal period. The oldest mother to giver birth via donor eggs was 74! So by freezing eggs, you can preserve fertility for as much as 40 years.
If you're single and a guy, then there are not really many action items for you. Sperm quality doesn't really seem to decline until about 40, at which point it drops off slowly. The only direct option here would be to get eggs from a donor bank, but if you do that you'd likely have to face the challenges of single parenting. Plus donor eggs cost a few tens of thousands, so it would be quite a bit more expensive.
HOW DO I ACTUALLY DO THIS?
If you're seriously considering doing IVF for polygenic screening, the first step is comparing IVF clinics. Some IVF clinics are 3x the cost of others for essentially the same service. Some IVF clinics have poor implantation cryopreservation and low implantation success rates. So choosing the right clinic will have a big effect on your cost/benefit analysis. Egg retrieval usually takes 3-6 visits from what I've heard, so it may actually be worth flying to another state (or perhaps even another country) to lower the price.
You then have to consider the IVF funnel to figure out how much it's going to cost to achieve a certain reduction in disease risk/increase in healthspan. I really wish there was a tool for this because a lot of factors can substantially affect loss rates. But the basic gist is this: at each step in the IVF process, fewer eggs/embryos come out than go in. The three most important factors affecting the number of embryos you have to choose from are the IVF clinic, the genetic testing company, and the age of the mother.
Here are all the steps that have to be done.
-
Medication is taken stimulating egg production
-
Eggs are extracted
-
Eggs are frozen and unfrozen at a later date (optional but necessary for polygenic screening)
-
Eggs are fertilized, turning them into embryos
-
Embryos grow to day 5 blastocysts, at which point they are biopsied
-
Day 5 blastocysts are biopsied for polygenic screening (and to see if they're chromosomally normal)
-
The euploiod embryo with the highest polygenic score is implanted.
-
A baby is born
At every single one of these steps, fewer eggs/embryos come out than went in.
If you're 23-28 you'll probably get around 15 eggs per cycle of IVF. According to some random news articles I looked up, 40%-50% of those will grow to day 5 blastocysts (this might be higher if you go to a good clinic and/or don't have fertility issues)
If you're 23-28, about 80% of the embryos that reach this stage will be euploid, meaning they have the potential to implant and turn into a healthy child. The others will either result in miscarriage or have a condition like Down Syndrome if implanted.
When you choose an embryo to implant, there's a roughly 70% chance it will lead to a live birth (lower if you have fertility issues).
So roughly 30% of eggs extracted will lead to a live birth (though it should be noted that the above numbers may not be accurate since my numbers might be wrong a bunch of factors influence the percentage).
That means you need 3-4x as many eggs extracted as you want to select from. At 15 eggs per IVF cycle in good conditions, that's 2-3 rounds of egg extraction if you want 10 embryos to choose from (taking implantation rates into account).
I think egg freezing is about $6k/cycle with genetic testing included. So for 3 cycles, that's about $20k. Then IVF itself is I think like $15k. So maybe $35k-45k all-in cost not including the cost of childbirth, which is stupidly exensive but usually covered by insurance.
It should be noted that there's actually a pretty big gain from selecting from just 2 embryos. Going up to 10 increases the benefit by about 80%, but the gains are still pretty noticable from any selection at all.
Anyways, I hope this was helpful. Let me know if you want me to write a more in-depth post about how to do IVF for polygenic selection.
I think people underestimate the degree to which hardware improvements enable software improvements. If you look at AlphaGo, the DeepMind team tried something like 17 different configurations during training runs before finally getting something to work. If each one of those had been twice as expensive, they might not have even conducted the experiment.
I do think it's true that if we wait long enough, hardware restrictions will not be enough.
4Billionaires read LessWrong. I have personally had two reach out to me after a viral blog post I made back in December of last year.
The way this works is almost always that someone the billionaire knows will send them an interesting post and they will read it.
Several of the people I've mentioned this to seemed surprised by it, so I thought it might be valuable information for others.
It's not clear whether that will mean the end of humanity in the sense of the systems we've created destroying us. It's not clear if that's the case, but it's certainly conceivable. If not, it also just renders humanity a very small phenomenon compared to something else that is far more intelligent and will become incomprehensible to us, as incomprehensible to us as we are to cockroaches.
Q: That's an interesting thought. [nervous laughter]
Hofstadter: Well, I don't think it's interesting. I think it's terrifying. I hate it. I think about it practically all the time, every single day. [Q: Wow.] And it overwhelms me and depresses me in a way that I haven't been depressed for a very long time.
I don't think I've ever seen a better description of how I feel about the coming creation of artificial superintelligence. I find myself returning over and over again to that post by benkuhn about "Staring into the abyss as a core life skill" I think that is going to become a necessary core life skill for almost everyone in the coming years.
It has been morbidly gratifying to see more and more people develop the same feelings about AI as I have had for about a year now. Like validation in the worst possible way. I think if people actually understood what was coming there would be a near total call to ban improvements in this technology and only allow advancement under very strict conditions. But almost no one has really thought through the consequences of making a general purpose replacement for human beings.
This reminds me a bit of my own hiring process. I wanted to work for a company doing polygenic embryo screening, but I didn't fit any of the positions they were hiring for on their websites, and when I did apply my applications were ignored.
One day Scott Alexander posted "Welcome Polygenically Screened Babies", profiling the first child to be born using those screening methods. I left a comment doing a long cost-effectiveness analysis of the technology, and it just so happened that the CEO of one of the companies read it and asked me if I'd like to collaborate with them.
The collaboration went well and they offered me a full-time position a month later.
All because a comment I left on a blog.
Man, what a post!
My knowledge of alignment is somewhat limited, so keep in mind some of my questions may be a bit dumb simply because there are holes in my understanding.
It seems hard to scan a trained neural network and locate the AI’s learned “tree” abstraction. For very similar reasons, it seems intractable for the genome to scan a human brain and back out the “death” abstraction, which probably will not form at a predictable neural address. Therefore, we infer that the genome can’t directly make us afraid of death by e.g. specifying circuitry which detects when we think about death and then makes us afraid. In turn, this implies that there are a lot of values and biases which the genome cannot hardcode…
I basically agree with the last sentence of this statement, but I'm trying to figure out how to square it with my knowledge of genetics. Political attitudes, for example, are heritable. Yet I agree there are no hardcoded versions of "democrat" or "republican" in the brain.
This leaves us with a huge puzzle. If we can’t say “the hardwired circuitry down the street did it”, where do biases come from? How can the genome hook the human’s preferences into the human’s world model, when the genome doesn’t “know” what the world model will look like? Why do people usually navigate ontological shifts properly, why don’t people want to wirehead, why do people almost always care about other people if the genome can’t even write circuitry that detects and rewards thoughts about people?”.
This seems wrong to me. Twin studies, GCTA estimates, and actual genetic predictors all predict that a portion of the variance in human biases is "hardcoded" in the genome. So the genome is definitely playing a role in creating and shaping biases. I don't know exactly how it does that, but we can observe that such biases are heritable, and we can actually point to specific base pairs in the genome that play a role.
Somehow, the plan has to be coherent, integrating several conflicting shards. We find it useful to view this integrative process as a kind of “bidding.” For example, when the juice-shard activates, the shard fires in a way which would have historically increased the probability of executing plans which led to juice pouches. We’ll say that the juice-shard is bidding for plans which involve juice consumption (according to the world model), and perhaps bidding against plans without juice consumption.
Wow. I'm not sure if you're aware of this research, but shard theory sounds shockingly similar to Guynet's description of how the parasitic lamprey fish make decisions in "The Hungry Brain". Let me just quote the whole section from Scott Alexander's Review of the book:
How does the lamprey decide what to do? Within the lamprey basal ganglia lies a key structure called the striatum, which is the portion of the basal ganglia that receives most of the incoming signals from other parts of the brain. The striatum receives “bids” from other brain regions, each of which represents a specific action. A little piece of the lamprey’s brain is whispering “mate” to the striatum, while another piece is shouting “flee the predator” and so on. It would be a very bad idea for these movements to occur simultaneously – because a lamprey can’t do all of them at the same time – so to prevent simultaneous activation of many different movements, all these regions are held in check by powerful inhibitory connections from the basal ganglia. This means that the basal ganglia keep all behaviors in “off” mode by default. Only once a specific action’s bid has been selected do the basal ganglia turn off this inhibitory control, allowing the behavior to occur. You can think of the basal ganglia as a bouncer that chooses which behavior gets access to the muscles and turns away the rest. This fulfills the first key property of a selector: it must be able to pick one option and allow it access to the muscles.
Spoiler: the pallium is the region that evolved into the cerebral cortex in higher animals.
Each little region of the pallium is responsible for a particular behavior, such as tracking prey, suctioning onto a rock, or fleeing predators. These regions are thought to have two basic functions. The first is to execute the behavior in which it specializes, once it has received permission from the basal ganglia. For example, the “track prey” region activates downstream pathways that contract the lamprey’s muscles in a pattern that causes the animal to track its prey. The second basic function of these regions is to collect relevant information about the lamprey’s surroundings and internal state, which determines how strong a bid it will put in to the striatum. For example, if there’s a predator nearby, the “flee predator” region will put in a very strong bid to the striatum, while the “build a nest” bid will be weak…
Each little region of the pallium is attempting to execute its specific behavior and competing against all other regions that are incompatible with it. The strength of each bid represents how valuable that specific behavior appears to the organism at that particular moment, and the striatum’s job is simple: select the strongest bid. This fulfills the second key property of a selector – that it must be able to choose the best option for a given situation…
With all this in mind, it’s helpful to think of each individual region of the lamprey pallium as an option generator that’s responsible for a specific behavior. Each option generator is constantly competing with all other incompatible option generators for access to the muscles, and the option generator with the strongest bid at any particular moment wins the competition.
You can read the whole review here or the book here. It sounds like you may have independently rederived a theory of how the brain works that neuroscientists have known about for a while.
I think this independent corroboration of the basic outline of the theory makes it even more likely shard theory is broadly correct.
I hope someone can work on the mathematics of shard theory. It seems fairly obvious to me that shard theory or something similar to it is broadly correct, but for it to impact alignment, you're probably going to need a more precise definition that can be operationalized and give specific predictions about the behavior we're likely to see.
I assume that shards are composed of some group of neurons within a neural network, correct? If so, it would be useful if someone can actually map them out. Exactly how many neurons are in a shard? Does the number change over time? How often do neurons in a shard fire together? Do neurons ever get reassigned to another shard during training? In self-supervised learning environments, do we ever observe shards guiding behavior away from contexts in which other shards with opposing values would be activated?
Answers to all the above questions seem likely to be downstream of a mathematical description of shards.
I've started a gene therapy company, raised money, opened a lab, hired the inventor of one of the best multiplex gene editing techniques to be our chief scientific officer, and am currently working on cell culture experiments with the help of a small team.
I may write a post about what's happened at some point. But things are moving.