It seems to me that there's an oversimplification in the discussion of heritability and fitness.
The argument about the breeder's function and what that implies about historical IQ/EQ relies on two implicit assumptions: that what is being inherited is 'just' intelligence with no meaningful side effects, and that the fitness function of that is monotonically increasing.
There's no reason that either of those has to be true. At the end of the day, what is actually happening is that something in the wiring of the brain is being tweaked - a biological engineering optimization, if you will. And a key lesson from non-biological engineering is that changes almost always have side-effects, and the value of the whole package is almost never monotonically increasing out to infinity.
The clearest example of this I've seen is with racehorses. A faster racehorse is better, right? And one key way to make a horse faster is to make the legs lighter, particularly the lower leg (basic physics here - less mass is easier to accelerate). But as you do that, you end up shaving down the leg bones more and more, and at some point the incremental speed gain from a lighter leg is dramatically outweighed by the fact that a bone just snapped and it's now lying in a crumpled heap on the ground.
So what is evolution metaphorically 'shaving down' to increase g or GFP, and is there a point at which that fails catastrophically?
>It seems to me that there's an oversimplification in the discussion of heritability and fitness.
Many! It's kind of a toy model. What I wanted to get at was the tension between a) defining g as intelligence, the ability to accomplish goals, and an unalloyed good b) high heritability c) high intelligence of Ancient Greeks. Your racehorse analogy is good, and I think it would change what some believe about intelligence. There are many that treat g as an unalloyed good. Some talk of increased risk of autism, but that's about the only caveat I have heard. You suggest there is some phenotypic constraint (analogous to brittle legs), and that seems likely to me.
That is on the phenotypic level. It could also be that the genes that code for g also code for some other mysterious bad thing. That would also be really interesting, and I'd like to see that articulated. Particularly given the interest in using embryo selection to increase g. I actually wrote about embryo selection for GFP over here: https://parrhesia.substack.com/p/embryo-selection-and-our-stone-age
>So what is evolution metaphorically 'shaving down' to increase g or GFP, and is there a point at which that fails catastrophically?
In this case it seems we should be able to find deficiencies in high g people, right?
Yeah, that basically comes down to "heritable, purely beneficial, smart Greeks: pick two". As is probably clear, I think 'purely beneficial' is the most questionable one there.
In terms of what I'll call a 'true' downside, I'd be looking for one of two things: a) something along the lines of overfitting - a point where more 'intelligence' starts increasing false positives more than valid ones, or b) a trade-off between two different things the brain is supposed to do. There's only so many neurons available, after all, and you can't build infinite complexity into a finite system. Autism looks to me like it would fit one or both of those, and more obviously some symptoms of paranoia look an awful lot like overtuned pattern recognition. In this case, I wouldn't *necessarily* expect that there's a detectable flaw in high-g people in general. It could be a threshold effect of some sort: 'Whoops, you just crossed from fitting into overfitting'.
Secondly, though, it's important to remember that what matters here is *reproductive* fitness, not quality of life, so the effect doesn't need to be something that we'd automatically consider a negative at first glance - from the perspective of evolution, 'can effectively utilise birth control' is a major drawback. How many copies of a hypothetical 'smart' gene would have been removed from the gene pool because the carrier was too busy changing the world to reproduce?
Not claiming to have any hard answers here - I'm an interested amateur, not an expert. I haven't been able to track it down again, but I picked up most of this from a piece on 'brains on the cliff edge' (or something similar to that) which I think was linked by Rob Henderson on his Substack.
Lathspell, happy to see you calling out that any conception of "good" has to be clear whether it's referring to evolutionary fitness vs. quality of life. I read way too many essays and commentaries where you can see the author's train of thought oscillating between these two without them even realizing it.
That said, in a species like ours where there's been thousands of generations of sexual selection performed by individuals possessing highly culturalized minds, I challenge that evolution wouldn't have solely favored cognitive traits that yielded better survival skills, whether that be figuring out tools, making plans, or getting along with others. What if our psychologies also have a random assortment of peacock's tails evolved into them?
As the Cultural Imperative broke free from the Genetic Imperative that spawned it, any number of traits that don't impact survival might subconsciously become attractive in a sex partner, just because having that partner made one "feel better" -- i.e., enjoy a higher quality of life of mind, you could say. If so, seems that would introduce a lot of confounding noise into attempts to make sense of g and GFP in terms of their evolutionary fitness, where we think of fitness as just the ability to better survive and support children.
I was going to comment something along these lines. Except (correct me if I'm wrong as I'm not an expert here) it seems to me like the second assumption is even stronger: not of monotonically increasing fitness, but of linearly increasing fitness. It seems intuitive to me that there are effectively step discontinuities in this function: being just smart enough for agriculture is much better than being just below that threshold. But also it seems to me like this function should have a saturation point for individuals which is a function of the overall intelligence and resource availability of society - what is there in being smart enough to do quantum chromodynamics if your society lacks the means to build a particle accelerator? If you think along those lines, it's perfectly possible that the greats of antiquity were near the saturation point of their age, that this is still a decent level for modern standards, but that the fitness landscape of the time was such that much less intelligent individuals could be fitter if they were physically strong or just generally healthier (see hypothesis that increases in intelligence are related to certain genetic conditions which may lower fitness, especially if Hypocrates is the best medic around).
It does become more complicated if the fitness landscape is sometimes positive and negative. But the cumulative effect (ignoring second-order problems like completely removing genes from the gene pool), will be roughly the average selection gradient.
>but of linearly increasing fitness
Yes, this would change things, especially if there was a "smart enough" as a farmer vs a lawyer or merchant. However, one of the common claims about IQ is that more is always better, and it's not that you have to be "smart enough" to be a physicist or a doctor and additional IQ past the threshold is not as useful. They will cite IQ correlates in job performance like being a janitor. Could be that a janitor also has to be literate to pay taxes, etc, and that is simply a modern artefact. At any rate, adding this to the model would challenge some claims about g.
Thinking further on this, I argue that there's a fundamental qualitative difference between IQ and GFP, such that GFP should not be described as 'intelligence'. That doesn't mean it isn't an incredibly valuable mental quality, just that it should be categorised separately.
I'm here defining intelligence as the capacity to search a solution space an optimal solution via reasoning. Take inputs of available resources, known constraints, a set of values according to which results should be judged, etc, and then output the course of action that best suits the given scenario.
GFP, then, isn't an alternative technique for searching the solution space, it's a hardwired override on the pre-existing technique: all solutions in the subsection of the solution space known as 'being a dickhead' are to be discarded (/strongly devalued) *without* analysis or consideration.
It's a short-cut. The same result can be produced via standard intelligence as well, but it takes a fairly deep education on the iterated prisoner's dilemma, the two-box problem, and other similar game-theory constructs, plus the ability to relate such abstractions back to day-to-day life.
That's outside what has been factored in to human intelligence 'properly' but it's an important (and adaptive) result, so it's been built in to our cognition in another form. Calling it 'emotional intelligence' is like calling the aversion to rotten meat 'scent intelligence' or teenagers being horny 'sex intelligence'. It's not, but the word 'intelligence' is prestigious so people who want to argue that this other thing is also valuable are mangling the definition in order to try to claim some of that prestige for the concept they think is being underrated.
TLDR: 'Intelligent' is 'gets the right answer and can show your work'. GFP is one of a set of older/deeper mental processes that get the right answer but can't show the work, so should be classified separately.
>How often do you think people can show their work in real life?
Not particularly often - maybe 10%? It depends on which set of decisions we're talking about, though. Snap physical reactions in combat/sports would be basically zero. Same with what to have for lunch. Building a bridge would tend to be rather higher (at least if the bridge is going to be any good...).
How fit it was historically... that's hard to answer. Gut feeling is that it could be a group-level thing as much as individual: it's quite valuable to have at least one highly intelligent member of your tribe, but doesn't provide significant extra value having *all* the members being that smart (they just need to listen to the one who is in the areas where it matters most). For the average member, it's necessary to be intelligent enough to *follow* the explanation, but not to be able to come up with it first.
I should note that by 'show the work' I don't mean it has to be a 100% rigorous mathematical derivation from first principles - more that there are enough (consciously understood) substeps that the conclusion can be explained to someone else who doesn't immediately buy it, as opposed to something like a moral rule where the 'explanation' is repeating 'it just is!' in a steadily louder voice.
>Calling it 'emotional intelligence' is like calling the aversion to rotten meat 'scent intelligence' or teenagers being horny 'sex intelligence'.
This is an excellent line :) Though I really am a Lexical Hypothesis maximalist; I try to defer to the masses on definitions whenever possible. "Intelligent," "knowledgeable," and so on load strongly on the GFP, so I think that they must be related. Of course one can construct another definition of intelligence but IMO that is dangerous territory. Many have tried!
I'm curious how well you think g captures intelligence, and additionally the difference between intelligence and wisdom. GFP def of intelligence certainly bleeds into wisdom which some may wish to define as completely different.
One thing I'd like to note is that what excites me the most is that GFP could be *phylogenetically* fundamental to intelligence. I think recursive thought emerged when the self perceived the self. It would still be interesting to me if recursion and reason are separate from GFP, but came from the same processes which produce the GFP in language and personality tests.
In terms of deferring to the masses, I don't think the specific phrasing 'emotional intelligence' ever actually came from them in the first place - I'd argue that it's widely used because the general population is deferring to perceived experts. I also think it has too many poorly-specified and somewhat conflicting definitions, so I'm wary of any survey data based on it - I don't think there's a good way of knowing what question the respondent was actually answering, or if they were all answering the same question, or how the answers may relate to another question that also used that particular phrase. You'll notice that I've mostly talked about GFP for that side - it's because that's actually well-defined enough to be useful.
RE: intelligence and wisdom, I do have a very clear split between my definitions for those: in my view, intelligence is capacity to work with with theory and abstractions; while wisdom is condensed empirical understanding of the world that doesn't require a theoretical underpinning, often developed and transmitted over a multigenerational timespan. They have wildly different strengths - in a stable environment over the long term, wisdom can impart a level of detail and nuance in understanding that intelligence can't hope to match (particularly in low-tech societies), but it is also very fragile in the face of context changes or breaks in transmission. In a stable environment intelligence can use pre-existing theory as a significant force-multiplier, but it's fundamentally resilient to context changes that invalidate previous knowledge as it can bootstrap new theory far, far quicker than new wisdom can be produced in a scenario where the old rules either no longer apply or have been lost.
There's overlap, in that intelligence will tend to improve the generation, transmission and refinement of wisdom, but it's a causal link, not equivalence.
More fundamentally, wisdom is something you *know*, while intelligent is something you *are*.
>In terms of deferring to the masses, I don't think the specific phrasing 'emotional intelligence' ever actually came from them in the first place - I'd argue that it's widely used because the general population is deferring to perceived experts.
I hoped to avoid much of this by talking about GFP, which is defined on natural language. There is still some "pissing in the lexical pool" by experts as "egotistical" is coined by Freud and entered common parlance. But for the most part language is _pure_ (as we will get)
>I also think it has too many poorly-specified and somewhat conflicting definitions
Indeed, including in this article. I define GFP a bunch of different ways, and in truth am not sure. Definitely a weakness of the argument (though that applies just as much to IQ, IMO)
"Nor is it surprising that Jesus condensed the 613 rules in the Old Testament to “do unto others as you would have them do unto you.” One can even think of it as a dimensionality reduction of the Old Testament. What is the one latent rule that generates all the others? Don’t kill or commit adultery are merely object-level implementations of this higher law, the Golden Rule. But importantly for Jesus, this is deeper than a commandment; it is a divine spiritual truth. Your soul was forged such that you cannot be at peace while you fail to treat others the way you would like to be treated. This is the Living Water that He offers. And it is strikingly close to Darwin’s claim that our minds must have been forged by this foundational tenant of morality."
Golden Rule/GFP is real without doubt, but its evolutionary advantage seems more likely to me to have been purely defensive, no matter what Jesus and/or Darwin thought about it. After all, in the world of humanity, "There are many enemies" (Musashi). GFP modeling allows for their early identification -- they are the residuals who aren't predicted by that model.
I think we are biased by living in complex societies, where there are more opportunities to defect. Simple strategies relying on reciprocal altruism would have been more useful in the past when we lived with and depended on the same group of 50 people for a lifetime.
However, risk aversion works at all social scales, all the time. The supposed Golden Rule altruism of the past would have also been usefully risk averse in effect if used to constantly scan the environment -- to determine, for example, if all 50 dependable people remain reliably so.
This has also made me think (again) about my views on tests of things like the Big Five. They've always seemed to me like:
a) A version of zodiac signs for scientifically-minded people;
b) A version of those mathematical parlor tricks where you ask people to think of a number, do a few mathematical operations, and then surprise them by ending with the same number they began with.
By this I mean that a "good" horoscope or zodiac sign description will make you instinctively feel like they apply to you personally and tell you something deep and personal about your condition. Traditional horoscopes achieve this by either shotgun tactics or by framing universal aspects of the human experience in personal terms. These tests achieve this by a process analogous to (b), by asking you loads of questions which correlate with personality traits and then telling you things about your personality - how did they know! oh yes, I told them. Maybe good "seers" do an intuitive version of that too!
But here you are using this to try and investigate important features of the human condition. I guess in the face of it this doesn't contradict my prior view - after all, this is still a universal feature, and implies nothing about the trick stuff. But it feels like it should make me seriously consider whether this stuff is tautological less in the way of party tricks and more in the way of pure mathematics...
I divide Big Five into the lexical stuff and the instrument stuff. Lexical structure really is very robust (for at least the first three factors), but there is inherently going to be a bunch of noise when you start trying to place an individual on a factor and then explaining what that means. Even the idea of a point-estimate for personality is suspect, IMO. (In ML people used to make the same argument about word vectors, fwiw. Switching to transformers, which can attend to the context of a word, solved that for NLP in general though my work still treats words as point estimates.)
As an example of the point-estimate problem wrt horoscopes, people love being told they are an "introverted extrovert." And it's true, their desires are very context dependent. But it kinda contradicts the idea of giving someone a score on a factor.
>But here you are using this to try and investigate important features of the human condition. I guess in the face of it this doesn't contradict my prior view - after all, this is still a universal feature, and implies nothing about the trick stuff. But it feels like it should make me seriously consider whether this stuff is tautological less in the way of party tricks and more in the way of pure mathematics...
This is why it's so important to me that PC1 of the data supports what basically all the greats have said. Darwin, and many other evolutionary scientists, talk about morality in the evolution of humans. The golden rule independently pops up in basically every religious tradition. Importantly I don't think that is just a game-theoretically stable way to organize society. It is such a common teaching because we literally evolved that module in our brain and whenever people start trying to abstract morality they get to this idea.
I might contribute some angles that I believe to be relevant, both with respect to IQ/EQ over time and within the nature/nuture debate, that I have not seen frequently raised.
Assume for a moment that we are all born as high performance cars, or high IQ if you will. While all of us could theoretically race at identical laps, this would not tend to happen in reality. Some would just not get fuel in them, some the wrong fuel, some use the clutch non-optimally or gas or brakes or not handle drifting etc etc. Over time it would also increase the discrepancies, as some will do more laps, pay more attention, develop more of a desire to lower their lap times. Some will not change tires or do services at proper intervals etc.
That whole analogy serves to point out the simple fact that our mental "machine" is simply used or understood at different levels both across time, across cultures and families.
This will, I believe create significant differences in performance (on the narrow ranges measured), even assuming equal potential, analogous to genetically identical twins where one exercises and the other does not. One will clearly lift more than the other over time.
What happens in practice, I believe, is the introduction or removal of bad "code" in our minds. We can hold contradictory beliefs that are mutually incompatible, and function almost optimally when we do not place much weight on these believes. Once more weight is placed, eg attention, we spend much energy basically running loops or what we might see as syntaxt errors. This reduces our ability to deliver with respect to our potential.
Any machine works best in the absence of friction, or friction that results in energy expenditure on "things" that do not directly contribute to the objective goal of the machine. (Now please do not take all my analogies as anything but serving to make a point). This is where we might engage in a debate on the central point of most religions, how cultures and parents impact their inhabitants/childrens ability to reach their potential, and if that is even desirable.
I generally think of IQ as the ability process abstractions and conclude the logical next step/former step of the system that the abstraction represents. Often this implies active "thinking" or using the brain actively. Whereas EQ is a particular subset that would align more with understanding the workings of your own mind and that of other minds (more wisdom, if you like). The discrepancy though is manifested through how "attention" is directed, with IQ it tends to be smaller/focused and EQ wider while still picking up the small clues/information that arise constantly.
Will end here in case you do not find it relevant...
I think the complaints about sky-hooks are overrated. Like imagine you had a cheeseburger; how would you know whether the cheeseburger is real?
Well, you could try to look at it. If it is bright, then light will bounce off the cheeseburger and into your eyes, and they will send nerve signals to your brain to give you an image of it.
You could also try to smell it. Particles from the cheeseburger disperse in the air, enter your nose, and activate the olfactory system which sends nerve signals to your brain.
Or you could try to touch it. Electrons in the cheeseburger will repel electrons in your skin, which in turns will activate your nerve signals.
Even if you eat it, that leads to it going into your stomach, where nerve signals for satiating get activated, and the burger gets dissolved and various complex (hormonal?) signals get sent around.
No matter what you do, you cannot observe the cheeseburger directly. Rather, the cheeseburger is a latent variable which through chains of cause and effect eventually influences your brain, making you learn about it.
Of course I shouldn't overstate the case. The cheeseburger can be observed in much much much greater fidelity than all psychological variable. But one shouldn't *understate* the indirectness of the material world either.
Maybe not, but then I think it is philosophically enlightening to think about what sense it is not a latent variable. It seems like it matches the formal definition (an entity which can only be indirectly observed through its effects on other things), and people working in AI often desire to construct networks that have what they call "latent spaces" with direct representations of semantically meaningful concepts such as cheeseburgers.
But yes I can think of at least one sense in which you are very correct: the point of terms like "latent variable" is to distinguish tangible things like cheeseburgers from wishy-washy things like g. What I am getting at is that most of this distinction seems to be down to the quality of measurement.
There's no sense at all in which physical objects are latent variables. If g was a cheeseburger or Ford F-150 there would be no mystery about its definition.
I guess I don't really know what your definition of "latent variable" is or how it differs from my definition which is "an entity which can only be indirectly observed through its effects on other things", so I don't really have any way to argue with you or parse out your position.
I use it the way everyone else uses it. From chatGPT:
"A latent variable is a variable that is not directly observed but is inferred or estimated from other variables that are observed or directly measured. The concept of latent variables is commonly used in many fields, including statistics, econometrics, psychometrics, and machine learning."
The idea that a cheeseburger is not directly observed is...certainly not the colloquial usage of "direct" or "observe."
A related idea from Hoel's recent book is that in neuroscience a common goal is to correlate neural excitement with physical actions (eg. moving the arm). But the neurons are so noisy it's hard to show anything. Therefore, it became standard to do the same action many times to create many samples, then average the neuronal response. But this is problematic because the average is just a construct. The brain never had access to the average and doesn't *do* anything with the average. There is a real divide between the observed data and the constructed average variable. Similarly, between observed data and PC1.
It seems to me that there's an oversimplification in the discussion of heritability and fitness.
The argument about the breeder's function and what that implies about historical IQ/EQ relies on two implicit assumptions: that what is being inherited is 'just' intelligence with no meaningful side effects, and that the fitness function of that is monotonically increasing.
There's no reason that either of those has to be true. At the end of the day, what is actually happening is that something in the wiring of the brain is being tweaked - a biological engineering optimization, if you will. And a key lesson from non-biological engineering is that changes almost always have side-effects, and the value of the whole package is almost never monotonically increasing out to infinity.
The clearest example of this I've seen is with racehorses. A faster racehorse is better, right? And one key way to make a horse faster is to make the legs lighter, particularly the lower leg (basic physics here - less mass is easier to accelerate). But as you do that, you end up shaving down the leg bones more and more, and at some point the incremental speed gain from a lighter leg is dramatically outweighed by the fact that a bone just snapped and it's now lying in a crumpled heap on the ground.
So what is evolution metaphorically 'shaving down' to increase g or GFP, and is there a point at which that fails catastrophically?
>It seems to me that there's an oversimplification in the discussion of heritability and fitness.
Many! It's kind of a toy model. What I wanted to get at was the tension between a) defining g as intelligence, the ability to accomplish goals, and an unalloyed good b) high heritability c) high intelligence of Ancient Greeks. Your racehorse analogy is good, and I think it would change what some believe about intelligence. There are many that treat g as an unalloyed good. Some talk of increased risk of autism, but that's about the only caveat I have heard. You suggest there is some phenotypic constraint (analogous to brittle legs), and that seems likely to me.
That is on the phenotypic level. It could also be that the genes that code for g also code for some other mysterious bad thing. That would also be really interesting, and I'd like to see that articulated. Particularly given the interest in using embryo selection to increase g. I actually wrote about embryo selection for GFP over here: https://parrhesia.substack.com/p/embryo-selection-and-our-stone-age
>So what is evolution metaphorically 'shaving down' to increase g or GFP, and is there a point at which that fails catastrophically?
In this case it seems we should be able to find deficiencies in high g people, right?
Yeah, that basically comes down to "heritable, purely beneficial, smart Greeks: pick two". As is probably clear, I think 'purely beneficial' is the most questionable one there.
In terms of what I'll call a 'true' downside, I'd be looking for one of two things: a) something along the lines of overfitting - a point where more 'intelligence' starts increasing false positives more than valid ones, or b) a trade-off between two different things the brain is supposed to do. There's only so many neurons available, after all, and you can't build infinite complexity into a finite system. Autism looks to me like it would fit one or both of those, and more obviously some symptoms of paranoia look an awful lot like overtuned pattern recognition. In this case, I wouldn't *necessarily* expect that there's a detectable flaw in high-g people in general. It could be a threshold effect of some sort: 'Whoops, you just crossed from fitting into overfitting'.
Secondly, though, it's important to remember that what matters here is *reproductive* fitness, not quality of life, so the effect doesn't need to be something that we'd automatically consider a negative at first glance - from the perspective of evolution, 'can effectively utilise birth control' is a major drawback. How many copies of a hypothetical 'smart' gene would have been removed from the gene pool because the carrier was too busy changing the world to reproduce?
Not claiming to have any hard answers here - I'm an interested amateur, not an expert. I haven't been able to track it down again, but I picked up most of this from a piece on 'brains on the cliff edge' (or something similar to that) which I think was linked by Rob Henderson on his Substack.
Lathspell, happy to see you calling out that any conception of "good" has to be clear whether it's referring to evolutionary fitness vs. quality of life. I read way too many essays and commentaries where you can see the author's train of thought oscillating between these two without them even realizing it.
That said, in a species like ours where there's been thousands of generations of sexual selection performed by individuals possessing highly culturalized minds, I challenge that evolution wouldn't have solely favored cognitive traits that yielded better survival skills, whether that be figuring out tools, making plans, or getting along with others. What if our psychologies also have a random assortment of peacock's tails evolved into them?
As the Cultural Imperative broke free from the Genetic Imperative that spawned it, any number of traits that don't impact survival might subconsciously become attractive in a sex partner, just because having that partner made one "feel better" -- i.e., enjoy a higher quality of life of mind, you could say. If so, seems that would introduce a lot of confounding noise into attempts to make sense of g and GFP in terms of their evolutionary fitness, where we think of fitness as just the ability to better survive and support children.
I was going to comment something along these lines. Except (correct me if I'm wrong as I'm not an expert here) it seems to me like the second assumption is even stronger: not of monotonically increasing fitness, but of linearly increasing fitness. It seems intuitive to me that there are effectively step discontinuities in this function: being just smart enough for agriculture is much better than being just below that threshold. But also it seems to me like this function should have a saturation point for individuals which is a function of the overall intelligence and resource availability of society - what is there in being smart enough to do quantum chromodynamics if your society lacks the means to build a particle accelerator? If you think along those lines, it's perfectly possible that the greats of antiquity were near the saturation point of their age, that this is still a decent level for modern standards, but that the fitness landscape of the time was such that much less intelligent individuals could be fitter if they were physically strong or just generally healthier (see hypothesis that increases in intelligence are related to certain genetic conditions which may lower fitness, especially if Hypocrates is the best medic around).
It does become more complicated if the fitness landscape is sometimes positive and negative. But the cumulative effect (ignoring second-order problems like completely removing genes from the gene pool), will be roughly the average selection gradient.
>but of linearly increasing fitness
Yes, this would change things, especially if there was a "smart enough" as a farmer vs a lawyer or merchant. However, one of the common claims about IQ is that more is always better, and it's not that you have to be "smart enough" to be a physicist or a doctor and additional IQ past the threshold is not as useful. They will cite IQ correlates in job performance like being a janitor. Could be that a janitor also has to be literate to pay taxes, etc, and that is simply a modern artefact. At any rate, adding this to the model would challenge some claims about g.
Thinking further on this, I argue that there's a fundamental qualitative difference between IQ and GFP, such that GFP should not be described as 'intelligence'. That doesn't mean it isn't an incredibly valuable mental quality, just that it should be categorised separately.
I'm here defining intelligence as the capacity to search a solution space an optimal solution via reasoning. Take inputs of available resources, known constraints, a set of values according to which results should be judged, etc, and then output the course of action that best suits the given scenario.
GFP, then, isn't an alternative technique for searching the solution space, it's a hardwired override on the pre-existing technique: all solutions in the subsection of the solution space known as 'being a dickhead' are to be discarded (/strongly devalued) *without* analysis or consideration.
It's a short-cut. The same result can be produced via standard intelligence as well, but it takes a fairly deep education on the iterated prisoner's dilemma, the two-box problem, and other similar game-theory constructs, plus the ability to relate such abstractions back to day-to-day life.
That's outside what has been factored in to human intelligence 'properly' but it's an important (and adaptive) result, so it's been built in to our cognition in another form. Calling it 'emotional intelligence' is like calling the aversion to rotten meat 'scent intelligence' or teenagers being horny 'sex intelligence'. It's not, but the word 'intelligence' is prestigious so people who want to argue that this other thing is also valuable are mangling the definition in order to try to claim some of that prestige for the concept they think is being underrated.
TLDR: 'Intelligent' is 'gets the right answer and can show your work'. GFP is one of a set of older/deeper mental processes that get the right answer but can't show the work, so should be classified separately.
How often do you think people can show their work in real life? How fit do you think this was historically?
I don't have a definition of intelligence, but in general think it should include adaptively reaching goals, leveraging abstractions along the way.
> GFP is one of a set of older/deeper mental processes
In line with the pylogenetic argument!
>How often do you think people can show their work in real life?
Not particularly often - maybe 10%? It depends on which set of decisions we're talking about, though. Snap physical reactions in combat/sports would be basically zero. Same with what to have for lunch. Building a bridge would tend to be rather higher (at least if the bridge is going to be any good...).
How fit it was historically... that's hard to answer. Gut feeling is that it could be a group-level thing as much as individual: it's quite valuable to have at least one highly intelligent member of your tribe, but doesn't provide significant extra value having *all* the members being that smart (they just need to listen to the one who is in the areas where it matters most). For the average member, it's necessary to be intelligent enough to *follow* the explanation, but not to be able to come up with it first.
I should note that by 'show the work' I don't mean it has to be a 100% rigorous mathematical derivation from first principles - more that there are enough (consciously understood) substeps that the conclusion can be explained to someone else who doesn't immediately buy it, as opposed to something like a moral rule where the 'explanation' is repeating 'it just is!' in a steadily louder voice.
>Calling it 'emotional intelligence' is like calling the aversion to rotten meat 'scent intelligence' or teenagers being horny 'sex intelligence'.
This is an excellent line :) Though I really am a Lexical Hypothesis maximalist; I try to defer to the masses on definitions whenever possible. "Intelligent," "knowledgeable," and so on load strongly on the GFP, so I think that they must be related. Of course one can construct another definition of intelligence but IMO that is dangerous territory. Many have tried!
I'm curious how well you think g captures intelligence, and additionally the difference between intelligence and wisdom. GFP def of intelligence certainly bleeds into wisdom which some may wish to define as completely different.
One thing I'd like to note is that what excites me the most is that GFP could be *phylogenetically* fundamental to intelligence. I think recursive thought emerged when the self perceived the self. It would still be interesting to me if recursion and reason are separate from GFP, but came from the same processes which produce the GFP in language and personality tests.
In terms of deferring to the masses, I don't think the specific phrasing 'emotional intelligence' ever actually came from them in the first place - I'd argue that it's widely used because the general population is deferring to perceived experts. I also think it has too many poorly-specified and somewhat conflicting definitions, so I'm wary of any survey data based on it - I don't think there's a good way of knowing what question the respondent was actually answering, or if they were all answering the same question, or how the answers may relate to another question that also used that particular phrase. You'll notice that I've mostly talked about GFP for that side - it's because that's actually well-defined enough to be useful.
RE: intelligence and wisdom, I do have a very clear split between my definitions for those: in my view, intelligence is capacity to work with with theory and abstractions; while wisdom is condensed empirical understanding of the world that doesn't require a theoretical underpinning, often developed and transmitted over a multigenerational timespan. They have wildly different strengths - in a stable environment over the long term, wisdom can impart a level of detail and nuance in understanding that intelligence can't hope to match (particularly in low-tech societies), but it is also very fragile in the face of context changes or breaks in transmission. In a stable environment intelligence can use pre-existing theory as a significant force-multiplier, but it's fundamentally resilient to context changes that invalidate previous knowledge as it can bootstrap new theory far, far quicker than new wisdom can be produced in a scenario where the old rules either no longer apply or have been lost.
There's overlap, in that intelligence will tend to improve the generation, transmission and refinement of wisdom, but it's a causal link, not equivalence.
More fundamentally, wisdom is something you *know*, while intelligent is something you *are*.
>In terms of deferring to the masses, I don't think the specific phrasing 'emotional intelligence' ever actually came from them in the first place - I'd argue that it's widely used because the general population is deferring to perceived experts.
I hoped to avoid much of this by talking about GFP, which is defined on natural language. There is still some "pissing in the lexical pool" by experts as "egotistical" is coined by Freud and entered common parlance. But for the most part language is _pure_ (as we will get)
>I also think it has too many poorly-specified and somewhat conflicting definitions
Indeed, including in this article. I define GFP a bunch of different ways, and in truth am not sure. Definitely a weakness of the argument (though that applies just as much to IQ, IMO)
"Nor is it surprising that Jesus condensed the 613 rules in the Old Testament to “do unto others as you would have them do unto you.” One can even think of it as a dimensionality reduction of the Old Testament. What is the one latent rule that generates all the others? Don’t kill or commit adultery are merely object-level implementations of this higher law, the Golden Rule. But importantly for Jesus, this is deeper than a commandment; it is a divine spiritual truth. Your soul was forged such that you cannot be at peace while you fail to treat others the way you would like to be treated. This is the Living Water that He offers. And it is strikingly close to Darwin’s claim that our minds must have been forged by this foundational tenant of morality."
Golden Rule/GFP is real without doubt, but its evolutionary advantage seems more likely to me to have been purely defensive, no matter what Jesus and/or Darwin thought about it. After all, in the world of humanity, "There are many enemies" (Musashi). GFP modeling allows for their early identification -- they are the residuals who aren't predicted by that model.
I think we are biased by living in complex societies, where there are more opportunities to defect. Simple strategies relying on reciprocal altruism would have been more useful in the past when we lived with and depended on the same group of 50 people for a lifetime.
However, risk aversion works at all social scales, all the time. The supposed Golden Rule altruism of the past would have also been usefully risk averse in effect if used to constantly scan the environment -- to determine, for example, if all 50 dependable people remain reliably so.
This has also made me think (again) about my views on tests of things like the Big Five. They've always seemed to me like:
a) A version of zodiac signs for scientifically-minded people;
b) A version of those mathematical parlor tricks where you ask people to think of a number, do a few mathematical operations, and then surprise them by ending with the same number they began with.
By this I mean that a "good" horoscope or zodiac sign description will make you instinctively feel like they apply to you personally and tell you something deep and personal about your condition. Traditional horoscopes achieve this by either shotgun tactics or by framing universal aspects of the human experience in personal terms. These tests achieve this by a process analogous to (b), by asking you loads of questions which correlate with personality traits and then telling you things about your personality - how did they know! oh yes, I told them. Maybe good "seers" do an intuitive version of that too!
But here you are using this to try and investigate important features of the human condition. I guess in the face of it this doesn't contradict my prior view - after all, this is still a universal feature, and implies nothing about the trick stuff. But it feels like it should make me seriously consider whether this stuff is tautological less in the way of party tricks and more in the way of pure mathematics...
I divide Big Five into the lexical stuff and the instrument stuff. Lexical structure really is very robust (for at least the first three factors), but there is inherently going to be a bunch of noise when you start trying to place an individual on a factor and then explaining what that means. Even the idea of a point-estimate for personality is suspect, IMO. (In ML people used to make the same argument about word vectors, fwiw. Switching to transformers, which can attend to the context of a word, solved that for NLP in general though my work still treats words as point estimates.)
As an example of the point-estimate problem wrt horoscopes, people love being told they are an "introverted extrovert." And it's true, their desires are very context dependent. But it kinda contradicts the idea of giving someone a score on a factor.
>But here you are using this to try and investigate important features of the human condition. I guess in the face of it this doesn't contradict my prior view - after all, this is still a universal feature, and implies nothing about the trick stuff. But it feels like it should make me seriously consider whether this stuff is tautological less in the way of party tricks and more in the way of pure mathematics...
This is why it's so important to me that PC1 of the data supports what basically all the greats have said. Darwin, and many other evolutionary scientists, talk about morality in the evolution of humans. The golden rule independently pops up in basically every religious tradition. Importantly I don't think that is just a game-theoretically stable way to organize society. It is such a common teaching because we literally evolved that module in our brain and whenever people start trying to abstract morality they get to this idea.
Thank you Andrew, I enjoyed reading this.
I might contribute some angles that I believe to be relevant, both with respect to IQ/EQ over time and within the nature/nuture debate, that I have not seen frequently raised.
Assume for a moment that we are all born as high performance cars, or high IQ if you will. While all of us could theoretically race at identical laps, this would not tend to happen in reality. Some would just not get fuel in them, some the wrong fuel, some use the clutch non-optimally or gas or brakes or not handle drifting etc etc. Over time it would also increase the discrepancies, as some will do more laps, pay more attention, develop more of a desire to lower their lap times. Some will not change tires or do services at proper intervals etc.
That whole analogy serves to point out the simple fact that our mental "machine" is simply used or understood at different levels both across time, across cultures and families.
This will, I believe create significant differences in performance (on the narrow ranges measured), even assuming equal potential, analogous to genetically identical twins where one exercises and the other does not. One will clearly lift more than the other over time.
What happens in practice, I believe, is the introduction or removal of bad "code" in our minds. We can hold contradictory beliefs that are mutually incompatible, and function almost optimally when we do not place much weight on these believes. Once more weight is placed, eg attention, we spend much energy basically running loops or what we might see as syntaxt errors. This reduces our ability to deliver with respect to our potential.
Any machine works best in the absence of friction, or friction that results in energy expenditure on "things" that do not directly contribute to the objective goal of the machine. (Now please do not take all my analogies as anything but serving to make a point). This is where we might engage in a debate on the central point of most religions, how cultures and parents impact their inhabitants/childrens ability to reach their potential, and if that is even desirable.
I generally think of IQ as the ability process abstractions and conclude the logical next step/former step of the system that the abstraction represents. Often this implies active "thinking" or using the brain actively. Whereas EQ is a particular subset that would align more with understanding the workings of your own mind and that of other minds (more wisdom, if you like). The discrepancy though is manifested through how "attention" is directed, with IQ it tends to be smaller/focused and EQ wider while still picking up the small clues/information that arise constantly.
Will end here in case you do not find it relevant...
Cheers.
I see EQ as a particular use of IQ similar to shape rotating. Not everyone with a high IQ has high EQ but everyone with high EQ has high IQ.
I think the complaints about sky-hooks are overrated. Like imagine you had a cheeseburger; how would you know whether the cheeseburger is real?
Well, you could try to look at it. If it is bright, then light will bounce off the cheeseburger and into your eyes, and they will send nerve signals to your brain to give you an image of it.
You could also try to smell it. Particles from the cheeseburger disperse in the air, enter your nose, and activate the olfactory system which sends nerve signals to your brain.
Or you could try to touch it. Electrons in the cheeseburger will repel electrons in your skin, which in turns will activate your nerve signals.
Even if you eat it, that leads to it going into your stomach, where nerve signals for satiating get activated, and the burger gets dissolved and various complex (hormonal?) signals get sent around.
No matter what you do, you cannot observe the cheeseburger directly. Rather, the cheeseburger is a latent variable which through chains of cause and effect eventually influences your brain, making you learn about it.
Of course I shouldn't overstate the case. The cheeseburger can be observed in much much much greater fidelity than all psychological variable. But one shouldn't *understate* the indirectness of the material world either.
A cheeseburger isn’t a latent variable
Maybe not, but then I think it is philosophically enlightening to think about what sense it is not a latent variable. It seems like it matches the formal definition (an entity which can only be indirectly observed through its effects on other things), and people working in AI often desire to construct networks that have what they call "latent spaces" with direct representations of semantically meaningful concepts such as cheeseburgers.
But yes I can think of at least one sense in which you are very correct: the point of terms like "latent variable" is to distinguish tangible things like cheeseburgers from wishy-washy things like g. What I am getting at is that most of this distinction seems to be down to the quality of measurement.
There's no sense at all in which physical objects are latent variables. If g was a cheeseburger or Ford F-150 there would be no mystery about its definition.
I guess I don't really know what your definition of "latent variable" is or how it differs from my definition which is "an entity which can only be indirectly observed through its effects on other things", so I don't really have any way to argue with you or parse out your position.
I use it the way everyone else uses it. From chatGPT:
"A latent variable is a variable that is not directly observed but is inferred or estimated from other variables that are observed or directly measured. The concept of latent variables is commonly used in many fields, including statistics, econometrics, psychometrics, and machine learning."
The idea that a cheeseburger is not directly observed is...certainly not the colloquial usage of "direct" or "observe."
A related idea from Hoel's recent book is that in neuroscience a common goal is to correlate neural excitement with physical actions (eg. moving the arm). But the neurons are so noisy it's hard to show anything. Therefore, it became standard to do the same action many times to create many samples, then average the neuronal response. But this is problematic because the average is just a construct. The brain never had access to the average and doesn't *do* anything with the average. There is a real divide between the observed data and the constructed average variable. Similarly, between observed data and PC1.