This was all a really nice read, very informative, strong favorited. You did lose me in claiming a double exponent for economic growth resulting from increased intelligence. I don't have enough information to say you're right or wrong, but for such an extreme claim, I want to see more of your model. I also want to see more of your model because I just think it would be interesting!
Thanks!
Frankly the double exponential growth claim is one of the more speculative in the post. There ARE some similarities with a field like quantum computing, where Neven's law describes how quantum computers' performance at certain tasks improves at a doubly exponential rate.
The doubly exponential rate at which, according to Neven, quantum computers are gaining on classical ones is a result of two exponential factors combined with each other. The first is that quantum computers have an intrinsic exponential advantage over classical ones: If a quantum circuit has four quantum bits, for example, it takes a classical circuit with 16 ordinary bits to achieve equivalent computational power. This would be true even if quantum technology never improved.
My assertion of doubly exponential economic growth is premised on two things: economic growth being an exponential multiple of the abilities of the beings that create it, and intelligence itself increasing exponentially as we begin to adopt genetic modification.
This obviously could turn out to be wrong.
I've kind of used them interchangeably throughout the post, though I prefer to use "intelligence" to describe what we are actually aiming for: increasing the ability of individuals to accomplish goals given a set of constraints. There are highly g-loaded tests such as IQ tests or SAT scores that capture g well but don't always translate to real-world performance.
Part of the challenge for any genetic engineering program will be figuring out how to robustly measure traits we want to change. IQ test are highly g-loaded, but they can be gamed. I remember reading a study (I can't find it now) showing that if you trained people on IQ test you could increase their scores by like 15 points, which shows that IQ scores can become decoupled from g. Ideally we'd like a test that can't be gamed.
View my previous post on genetic engineering
In my last post I covered three different ways in which genetically engineering humans could be beneficial:
1. It could allow us to address the mismatch between our genes and our environment
2. It could allow us to reach states of higher fitness that evolution simply cannot
3. It could allow the adoption of traits that only have a benefit once a sufficiently large percentage of the population has them.
Today I will be covering a fourth benefit of genetic engineering that, although it could probably fall under one of the previous categories, deserves its own post due to the outsized impact it would likely have: increasing human intelligence.
Unusually high general intelligence is one of the human race's most distinguishing features. Our ability to cooperate flexibly and in large groups via language and our collective belief in fictitious identities is something no other species can match. Ant colonies cooperate in large groups, but they lack the flexibility of human organizations. Dolphins and chimpanzees have complex social relationships, but cannot cooperate at anywhere near the scale that humans can.
The exact characteristics that allow humans to do this are worthy of a whole book, but if you want a nice overview of the history of human civilization and the role of language in its development, I would recommend Sapiens: A Breif History of Humankind by Yuval Noah Harari. But before I make the case for purposefully engineering humans with greater intelligence, I think it is appropriate to explain some basic facts about intelligence that will help lay the groundwork.
Competence in abilities across multiple domains tends to correlate. An individual that is good at math will, more often than not, score highly in reading comprehension and deductive reasoning tests. This underlying factor that determines general ability has a name: g. The existence of the g factor was originally proposed by the English psychologist Charles Spearman in the early years of the 20th century. He observed that children's performance ratings, across seemingly unrelated school subjects, were positively correlated, and reasoned that these correlations reflected the influence of an underlying general mental ability that impacted performance on all kinds of mental tests. Spearman suggested that all mental performance could be conceptualized in terms of a single general ability factor, which he labeled g, and many narrow task-specific ability factors. Cognitive ability tests such as IQ tests are designed to measure this underlying g factor.
Many factors seem to influence g including nutrition, social connections, genes, and maternal womb environment. A combination of genes and maternal womb environment explain a high amount of variance in g between individuals. "Monozygotic twins raised apart are more similar in IQ (74%) than dizygotic (fraternal) twins raised together (60%) and much more than parent-children pairs (42%); half-siblings (31%); adoptive siblings (29%-34%); virtual twins, or similarly aged but unrelated children raised together (28%); adoptive parent-child pairs (19%) and cousins (15%). Nothing but genes can explain this hierarchy." (Matt Ridley)
One last note: IQ is often conflated with intelligence, but it is not actually the same thing. IQ is basically just a test that is designed to measure g. But even well-designed and well-administered IQ tests (and many IQ tests are neither) have their limitations in assessing the type of real-world performance that we ultimately care about. However, as I will explain shortly, they are still quite useful for predicting a broad range of outcomes.
Let me now turn to the benefits of raising intelligence. From an individual perspective, we might pursue superior Intelligence for social reasons: a greater likelihood of occupying high paying jobs, better health outcomes due to better choices, etc. But at a societal level, I think the question we should be asking ourselves is whether raising the intelligence of one individual positively affects others. In other words, are the benefits of intelligence zero-sum or positive-sum?
From the research I've done, the answer seems to be positive-sum. A quick perusal of the Wikipedia page on g reveals that g correlates with job performance, particularly in highly skilled positions. Unusually intelligent people also seem to have a lower discount rate; that is to say, they are more willing to accept a smaller payout now in exchange for a larger one later. Given that many of the world's biggest problems today (such as climate change, nuclear war, and other types of even worse existential risk) would be better dealt with if the general public placed a higher value on the quality of human lives lived in the future, it seems like lowering time discount rates could be extremely valuable to the long term survival of the human species.
There is also research showing that more intelligent people are disproportionately likely to support systems that enforce beneficial cooperation. Here I'll quote from Garett Jones' book Hive Mind:
Another study by Brown University economist Louis Putterman and his coauthors found still more evidence that higher-IQ individuals are more likely to start off by playing nice, by being generous team players. In this game, known as the public goods game, players individually decide how much of their own money to put in a metaphorical pot, the money doubles or triples, and then it gets divided up among the group. When you give money, you’re directly contributing to the public good. The game was repeated for a few rounds with the same team so players would have a chance to learn from each other, a chance to find a path to cooperation.
As this was run at Brown University, an Ivy League school where one might expect that almost all students are raised in incredibly advantaged environments, it might seem that differences in IQ scores would be irrelevant. But in Putterman’s cooperation experiment, IQ mattered. He and his coauthors found that higher-IQ students at Brown put more money in the pot during the early rounds of the game: the higher-IQ students were more pleasant early on. That’s the smart thing to do, because extra money early on can send a signal of kindness, of cooperativeness, to the other players. And it’s worth noting that in another part of the experiment, when the students could vote on a way to penalize low contributors, higher-IQ students were more likely to vote for a rule that would penalize the non-cooperators: so higher-IQ students were pleasant, but not naive.
The list of the benefits of higher intelligence goes on and on and on. Here's a summary of a study done by Anne Roe examining the average IQs of the top scientists in the US in the 1950s. Spoiler alert: the average IQ of these groups is extremely high: 163 in the group of 49 studied by Roe. This suggests that unusually intelligent people are significantly more likely to contribute to scientific advancement.
In summary: raising the average intelligence seems very likely to have huge benefits and a large amount of those benefits are accrued by the society in which intelligent people participate rather than the individual themselves.
A question that inevitably comes up when discussing the effects of raising intelligence is the following: if the benefits of raising intelligence are so huge, why hasn't evolution selected for it already?
To this I would make several comments. First of all, evolution HAS been optimizing for higher intelligence among homo sapiens for several million years now. Brain sizes roughly trip tripled over the last 3 million years, and metabolic rates have likewise increased. The average sedentary human burns about 27% more calories than the average chimpanzee, and there is evidence to indicate that we burn more calories than our ancestors. It seems very likely that this increase in brain size corresponded to a increase in intelligence, since at some point during that process humans developed the ability to use language, create more sophisticated tools and form increasingly complex societies. This assertion is backed up by modern studies showing that skull size positively correlates with IQ (though at a fairly modest factor of 0.3).
Second, only a fraction of the benefits created by an intelligent individual are captured by that individual. Take tool creation: tool creation is undeniably a product of intelligence (and not just human intelligence). Yet it is far more difficult to create a new tool than to copy its design once it has been created. Since evolution optimizes for inclusive reproductive fitness, and new tool creation is a fairly rare event, there is not much of a differential reproductive fitness advantage for individuals that are unusually adept at creating new and useful tool designs. Many of the benefits of intelligence are like this. Thus many of the benefits of intelligence that we actually care about do not provide sufficient evolutionary fitness advantages for natural selection to spread them through the population.
And the third reason we might expect raising intelligence to have large benefits is the environment in which we live today is much different than that of our ancestors. Factors that may have been constraints in the past are probably not constraints today. Two of the most notable are the ability of an infant head to fit through the birth canal and the caloric requirements on large energy-hungry brains. C-sections have been around since the mid 19th century. With the ability to perform C-sections, infant head size at birth is no longer the same kind of constraining factor that it once was.
Calorie expenditure is also becoming increasingly irrelevant, particularly in the developed world. Champion strongmen like Hafþór Júlíus Björnsson and Brian Shaw consume around 10,000 calories per day. And even those who can't make a living off of strongman competitions and sponsorship can afford to do this working fairly normal middle-class jobs (though obviously with difficulty). Calorie consumption was obviously a limiting factor in the past, but the expected cost of needing additional calories is significantly lower today than it was in the past.
We now come to the heart of the matter: is it actually possible to ethically increase human intelligence with modern techniques or those that are likely to be created in the near future? I plan to give a more detailed answer in a future post, but I will give a quick summary here because this post would be incomplete without showing that this is at least possible.
To genetically engineer humans in a desired direction, one needs two things: a method to modify genes (or at the very least to produce variance) and a means of telling how specific genetic variants influence trait expression.
There are many methods to do so, best summarized in this really fantastic post Gwern put together back in 2017. The most viable near term method for increasing intelligence is embryo selection. In this method, two parents produce a number of embryos via In Vitro Fertilization (IVF). Biopsies are performed on the resulting embryos, and the DNA from the extracted cells are sequenced. An embryo that scores well on a test designed to predict the embryo's expression of certain desired traits is implanted in the mother or surrogate.
There are two crucial limitations on the ability to use embryo selection to facilitate genetic engineering of desired traits: the ability to produce a large number of embryos and the knowledge of which genes (or regions of the chromosome) cause certain outcomes.
With current techniques, it is only safe to extract around ten eggs from the ovaries of a female during IVF. We would therefore expect only a very modest increase in IQ from embryo selection among 10 eggs: probably on the order of ~3 points.
The second limitation has to do with the most common method to measure the expected expression of a trait using genome sequencing: a Genome-Wide Association Study (GWAS). A GWAS is how we tease out the contribution of different genes to highly polygenic traits like intelligence, height, and heart disease risk. With so many genes influencing height, and moral considerations preventing us from using randomization techniques to determine which genes cause which traits, these studies are the next best way to determine causation.
These studies look at extremely large data sets of individuals with different expressions of certain traits and identify regions of DNA that are strongly associated with trait expression. Despite the massive increase in available data over the last decade, these studies still cannot usually identify the exact single nucleotide polymorphism (SNP) that causes the observed trait. Instead, a GWAS will tell you the section of 10 or so base pairs within which the causal variant is located, but can't currently tell you exactly which one is responsible for the observed effect.
This will be the least empirically supported section of my post because there are so many unknowns here. So treat this as a very informal and at least partially incorrect assessment of the likely outcomes.
Human intelligence is in one way or another responsible for modern civilization, for destruction of natural environments around the world, and for most good and bad things we experience on a daily basis. To say that dramatically increasing human intelligence would have a large impact on the world is an understatement.
Using embryo selection, it is likely that we would not "max out" desirable trait expression for a dozen generations or more depending on how many embryos were used for each generation. This occurs for a simple reason: there are roughly 3 billion base pairs in human DNA, and each generation of embryo selection only increase the frequency of desirable variants by a small amount.
There would therefore be gradually accruing inequality between generations as older generations found themselves outclassed by a new generation of smarter, better looking, healthier humans. While it is certainly possible to genetically engineer adult humans (such as is done during immunotherapy for cancer), it is significantly harder. And there are likely some changes that will be extremely hard to make in adults because their primary effect is to change the manner in which development occurs.
This problem will be compounded if Aubrey De Grey anyone from the dozens of startups working on life extension technology actually gets anti-aging technology to work and humans start to live to ages of 100 or more. My hope is that extremely smart humans will be better equipped to solve the difficult technical challenges involved in genetically altering adult humans in ways that would allow them to fully participate in a society of genetically enhanced people.
However, I do not suspect this will actually be a problem for more than a few generations. It is likely that either much more advanced techniques such as Iterated Embryo Selection (IES) will allow us to max out the gains from embryo selection in one generation, or we will reach the singularity and digital superintelligence will vastly out-class the performance of human brains with several billion years of technical design debt.
Increasing human intelligence is likely to have a substantial impact on our collective ability to mitigate existential risk.
As shown in the Brown University study, highly intelligent people also seem to have a discount factor closer to 1; they value the future more than people of lower intelligence and are therefore more likely to make tradeoffs that increase the quality of the future in exchange for making the present slightly worse. They also seem to have better models of how the world works.
Existential risk mitigation of all types requires exactly those traits to be properly addressed. They require researchers to do the actual work, and societies to support those researchers even though the benefits are not likely to be realized for decades at least. Both of these would be positively affected by increasing average human intelligence.
Take research on artificial intelligence safety, as an example: the work is highly technical and the major payoffs are not likely to occur for several decades. Unusually intelligent people are much more likely to solve major problems in the field, and intelligent people not working in the field are more likely than average to recognize the threat posed by unaligned powerful AI support increased research funding.
This is one area where things are likely to get really crazy. Human intelligence HAS been increasing over evolutionary timescales, but the rate of improvement has been glacial. All modern economic growth has been achieved by a brain whose basic hardware has received few upgrades. We're made a few gains from reducing mental stunting due to malnutrition and from eliminating leaded gasoline. But apart from those, almost all gains of the last ten thousand years have been object-level optimizations of the world itself, the number of humans living in it, and the accumulated knowledge of our species.
However, if we begin to use technology to increase intelligence, the equation for economic growth starts to look quite different. The unchanging optimizer, which was previously producing exponential economic growth is now itself being optimized. While no one knows exactly what the trajectory of intelligence improvement will look like, it is likely that until we truly max out the physical limits of computation, its improvement will be exponential (since the ability to design better intelligence is itself a function of the intelligence of those optimizing it).
I think it's therefore reasonable to believe that until we truly max out all possible gains, economic growth will resemble a double exponential of the form . Plot out the result of such a function and compare it to exponential growth and you will see just how dramatic of a difference this will make.
In my upcoming posts I plan to cover several more facets of human genetic engineering. Human genetic engineering brings up many moral questions. From what I have read so far there do not seem to be any huge showstopping problems that would prevent us from ever undertaking human genetic engineering, but there is certainly the potential for this technology to be abused and that potential is worthy of discussion. There is a long and sordid history of eugenics programs to be dealt with and I do not feel that I can really cover this topic without explaining all the ways we should NOT use this technology. I also think it's important to explain how techniques like embryo selection can avoid the toxic exclusionary policies of eugenics movements so that we can make sure when we do finally get around to genetic engineering it will benefit all and not simply widen the gap between the haves and the have-nots.
I also plan to discuss in detail the technical barriers preventing us from doing serious genetic engineering today and how those obstacles might be overcome.