What is intelligence?

In which I think out loud about some awful eugenics shit I was taught as a kid.

A bell curve with the international circle-bar ‘no’ symbol over it — Credit: Wikimedia user Dmcq, “IQ_distribution.svg” with modifications. Used and reshared under Creative Commons CC-BY-SA 3.0 Unported.

If you are reading this in English, then it is fair to say that you live in a culture that treats “intelligence” as one of the most prized traits of the human mind. But have you ever really thought about what it is?

Let’s consider some dialogues. Here’s the first:

You: Alice is 37 years old.
Me: I understand.
You: How old is Alice?
Me: Alice is 37 years old.

This is where ChatGPT is: it can repeat back the things it has been told or that it already knows from training. If I respond this way, does it demonstrate that I am intelligent? I don’t think so. I’m just regurgitating words.

Here’s another dialogue:

You: Alice is 37 years old.
Me: Got it.
You: Bob is two years younger than Alice.
Me: Understood.
You: How old is Bob?
Me: Bob is 35 years old, give or take a year since “two years younger” is a little ambiguous.

At first glance we might say that I showed more intelligence here than in the previous dialogue. But there’s still something lacking, because you asked me a question that could be easily worked out through symbolic manipulation: I parsed your words into concepts like “Alice”, “Bob”, “years”, “younger than” (a tricky but fully mechanical process), applied simple rules to transform your question into an algebra problem (mechanical), solved the algebra problem (mechanical and almost trivial), and finally translated it back to the original context to produce an answer (mechanical). I might have demonstrated “greater intelligence” than in the first scenario, but I don’t think a lot of us would still consider this to be truly “intelligent”, at least not in the way that people mean when they apply the word to themselves or to other adults. You can replace me with a machine, and that machine does not actually need to know what symbols like “year” or “younger” actually mean out here in the real world.

We’re still missing something. Let’s try a third dialogue:

You: Alice is 37 years old.
Me: Okay.
You: Bob is two years younger than Alice.
Me: Sure.
You: Can you tell me something about Alice and/or Bob that I haven’t told you already?
Me: Yes, Alice is 13514 days old or older, no less.
You: Can you tell me something else?
Me: Yes, Alice is strictly younger than 13880 days old.
Yes: Anything else?
Me: Yes, Bob is 12783 days old or older.
Yes: Can you put that in years?
Me: Yes, Bob is 35 years old, give or take about a year.

This seems like a step up, maybe, but it’s still purely mechanical! The scenario that was presented to me has very few moving parts, and so there are very few new facts that can be deduced from those moving parts. If I am a machine with no understanding of meaning, all I have to do is start enumerating facts until you tell me to stop, by just brute forcing through every possible fact I can deduce from existing facts plus the rules of logic. I combine every fact I know with every other fact I know, identify if the combination results in a new fact, and if so I tell you and I add it to my pool of known facts. Rinse and repeat. The only tricky part is that I have to list the facts in breadth-first order, so that every fact I’m capable of deducing will be deduced in finite time.

(Aside: The axioms of logic and the background facts themselves, such as sentence structures and leap year rules and the fact that a logical deduction is the correct response to your utterances, are themselves subjective and context-dependent… but that context is implied by the fact that we are both speaking English, on Earth, in a verbal register that conveys “we are using logical deduction”. This is true even if I am a mindless, wholly non-sentient large language model like ChatGPT, because such things are encoded into my model from the probabilities of certain words being near each other in my training data. It can all be done via mechanical manipulation of symbols that have no individual meaning. The only source of meaning is my training data: the choices you made in curating it, and the choices made by the humans who wrote it.)

To get any further, I must be able to choose an enumeration of facts: there are an infinite number of concrete orderings that will eventually produce every possible fact, just like the brute force ordering that I mentioned above, but the choice of order can be used to express the relevance of the facts. “Relevance” is the key word here: I need to start deciding that some facts are more important than others, that some are worth saying first and others are worth not saying until needed, and that requires knowing some context about (a) my own subjective goals, such as “correctly answering you”, and (b) the situation in which I personally happen to be in right here and right now, including the fact that you are the person asking me the question rather than someone else.

For humans, the “correct” way for me to order my facts in this scenario is to guess your state of mind and try to reproduce your beliefs about which facts are relevant. But that doesn’t really solve the mystery here, because how on Earth did you decide which facts were relevant? How did you decide that “Bob is 35 years old” is a better answer than “Bob is 12783 days old or older”, or that either of those is a better answer than anything to do with Alice’s age?

Well, on one hand, you probably had to make the same guess at some point when you were a kid, trying to figure out what kind of answer your parents or teachers wanted. Once you made the right guess, you remembered it for life. In other words, culture.

On the other hand, you probably had some sort of neurological hard-wired assumptions about which kinds of facts are likely to be relevant / important in general situations vaguely like this. In other words, biology.

Note very carefully, though, that culture and biology are themselves subjective. Both of them are determined by the context in which you live, you personally, not by universal truths that all cultures and all biologies must share across all of space and time. To be unique is to be subjective and contextual, rather than objective and universal. The only way to achieve objectivity and universality is to destroy uniqueness and individuality.

Let’s change things up a little. Imagine that you are a teacher, and you are putting each of your students into dialogues similar to the ones I was showing you before. The students don’t get to hear what’s said between you and any other students; you take each one individually into a closed, sound-proof, windowless room to state your hypotheticals and ask your questions. Of your ten students, the first eight picked exactly the same fact as the one to tell you first. The other two picked different facts, different both from the majority and from each other.

Who is “more intelligent”: the eight students who behaved identically to each other, the ninth student, or the tenth student?

Well, it depends. On a lot of things.

If the eight identical students all gave you exactly the answer you expected, that means they did an equally good job of exactly reproducing your beliefs about “relevance” of some facts over others. If your personal beliefs about relevance were somehow “objectively true”, then those eight would be the students who were “most intelligent” in the class. Easy peasy!

Let’s say that the dialogue with student 9 went on for twenty or thirty prompts, before they finally stumbled across the answer you were looking for. This means they did a poor job of guessing what your beliefs were about relevance of the facts. But is it a lack of intelligence, or a simple difference of context? Maybe the ninth student is an immigrant from another country, for instance, and wasn’t familiar with some of the English words you were using. Or maybe they grew up in a weird family with a lot of private slang.

Is “low intelligence” always just a lack of context? Could “being stupid” and “not knowing something” actually just be different names for the same thing? Maybe the ninth student is just three words away from realizing how different your expectations were from their own, but only if you can figure out which words those are… in which case you are the one who is “less intelligent” here, because you are failing to model the ninth student’s mind just as much as they are failing to model yours, because you do not understand their subjective cultural context. How can you both be less intelligent than each other?

And finally, let’s say that student 10 surprised you with their first fact, but quickly caught on that you were interested in hearing something specific that wasn’t that fact, and then were able to guess the one you were fishing for in just a few more tries. Is this student “more” intelligent than the first eight, or “less” intelligent? Is this student “more” intelligent than the ninth, or “less” intelligent? You were surprised to hear their first answer, but was it a better answer than the one you wanted, or a worse one?

I think the thing that would most stand out to you, that would be most inclined to make you label the tenth student as the “most intelligent”, would be if you realized that your own beliefs about which fact was “most relevant” was wrong according to your existing goals. Maybe the fact that you were looking for wasn’t the most relevant fact that you should have been quizzing your students on!

But maybe your belief was correct for the hypothetical as it existed in your head, but you left out certain things as implied that the tenth person didn’t pick up on, because they had a slightly different upbringing, and in the hypothetical that they imagined, their first choice was better than the one you wanted to hear. Maybe their brain is wired differently, so that even though they were exposed to the same culture that you were, they made different assumptions about the relevance of a fact… but, who’s to say that your assumptions are better than their assumptions, or vice versa? If your brains aren’t wired the same way, doesn’t that mean your situations are different and thus your goals are not identical?

Hopefully, it’s now clear that “intelligence” is actually this fuzzy thing with a deeply ill-defined meaning. What’s more, to the extent that there is a meaning of “intelligence”, it’s really about drawing analogies back and forth between symbolic / mechanical facts (which are free of meaning) and meanings drawn from past, present, and potential future subjective experiences. Simply put, “intelligence” is necessarily contextual because it’s about “doing the best thing in a situation”, but situations are always contextual and subjective. It’s perfectly normal and expected that the same person could have “high” intelligence in one situation but “low” intelligence in another, because “intelligence” seems to actually be about making guesses about which symbol-meaning analogies are useful and which are not, and using those guesses to communicate meaning between subjective viewpoints.

From a math / logic / computer science perspective, this is perfectly expected! Questions in zeroth-order (propositional) logic are always “decidable”, but getting a correct answer generally requires a number of steps that grows exponentially with the number of facts that are relevant to the scenario. (That’s the brute force ordering I was talking about before. In Computer Science, we call it P≠NP.) Meanwhile, questions in first-order logic and higher are outright undecidable: there is no procedure, none whatsoever that any inventor could ever invent or that evolution could ever stumble upon, that can answer every first-order question with a definite yes-or-no answer, or even to identify for an arbitrary question if the question even has a definite yes-or-no answer. Some questions are true under one assumption but false under another assumption, yet there’s not always a good reason why one assumption should be considered “safer” or “more correct” than the other. And if there is… say it with me… it’s contextual!

So of course there is no such thing as “general”, context-free, objective intelligence. It’s actually been irrefutably known for a century that such a thing is forbidden by logic itself!

We’ve just been kinda pigheaded about accepting it.

So what the flying fuck are eugenicists doing to their kids, then?

In addition to the obvious topic of eugenics, this section has a big ol' content warning about racism, antisemitism, and other forms of bloodline bigotry, and just dehumanizing ideologies in general.

The modern version of eugenics originated in the mid-1800s in the UK, quickly spreading to the rest of the British Empire and its current and former subjects, and then more slowly spreading to other countries (like Nazi-era Germany) as the ideas were translated. It was inspired primarily by two things:

A deep misunderstanding of Darwinian evolution propagated by people like Herbert Spencer and Sir Francis Galton, which was readily embraced by the British nobility and gentry because the misunderstanding perfectly matched their preconceived ideas about heritability
The culture which had developed in English-speaking countries around the race-centered chattel slavery system, which was created by Britain for use in its colonies (most famously the United States), which itself developed out of the same European hereditary feudalism that also led to the misunderstanding of Darwin

While heritability of traits in humans, crops, and domestic animals had been observed and remarked on since antiquity, there were no systematic explanations for how it worked until folks like Darwin and Mendel cracked it in the 1800s.

The early-modern concept of “race” predates that era by many centuries. In a nutshell, “Jew” was the first race in the modern European sense of the word: the Inquisition, which had the goal of forcibly converting European Jews and European Muslims to Christianity by threatening them with suffering and death if they did not, noticed that Jews were particularly hard to convert and tended to go back to their old practices when no one was looking. Instead of trying to understand the cultural differences that made the Christian message fall flat — a process that would have required interrogating the “obvious” supremacy of Christianity itself — the inquisitors came to believe that Jews were passing some heritable personality trait to their children through their bloodline, and that this personality trait made Jews lost causes who should just be put to death rather than converted.

After all, the principles of selective breeding in crops and livestock were very well known. Applying it to people, especially to people they saw as ideologically inferior to Christians, was not a huge leap. If a negative trait started appearing in your herd, you culled the individuals with the negative trait. Simple as that. No need to understand a Punnett square or to otherwise have a sense of the mathematics or logic that governed the process.

This was enhanced by the fact that European royalty and nobility were already starting to think of themselves as having superior bloodlines. When you consider the explanations that nobles told each other and their peasants, the old “might makes right” era of Gilgamesh and Achilles had long since given way to the “divine right of kings” era, but people were starting to question it, and a new excuse was needed for a new era. Although noble inbreeding originally began as a way of keeping wealth concentrated on just one branch of the bloodline, “good breeding” became the new just-so story that explained why the villagers needed to obey the Habsburg dude with the funny jaw who gained power by killing another Habsburg dude with a slightly different funny jaw. Some people were just better at ruling, after all, and the hemophilia and insanity were just some minor side effects of keeping those “good ruler” traits pure and concentrated in the bloodline.

Centuries later, those cultural tidbits spread to slaveowners in the Americas, who used it to invent the broad-brush races “Black” and “White”. (In contrast, European races had originally been very specific things like “Saxon”, “Breton”, “Pict”, and “Frank”, and were only just starting to coalesce into nation-race identities like “French” and “Italian”.) “White” was needed in the United States because the US had a mix of European-sense “races” as the landed gentry, such as Saxons, Franks, Scots, and so on, while “Black” was coined because they were sure that all of Africa was just one big race anyway because Europeans couldn’t be bothered to tell African faces and ethnicities apart.

Eugenics, then, was really just putting a cap in all of this “race” stuff. Basically, the idea was to breed humans for specific roles in the same way we breed dogs. Hunting dogs, herding dogs, pest control dogs, show dogs, and so on. The specific trait of “intelligence” was one of first that eugenicists wanted to breed humans for, because eugenicists had convinced themselves that “intelligence” was the quality which royals and nobles had but peasants lacked. So the first thing the eugenicists did was to outlaw “miscegenation”, which would undo their selective breeding efforts, and the second thing they did was to start planning how to breed intelligent people with other intelligent people.

Now, the eugenicists were wrong but they weren’t stupid. They had seen what happened to the Habsburgs and other inbred royal families, and they knew well what happens to domestic animals when you inbreed them too tightly. With Darwin, they finally had the key insight: you need random mutation to provide the basic material that selection can then act upon, which means you can’t be too aggressive with the culling step. Instead, they realized they would need to find peasants who had more of the “intelligence” trait due to random mutations, which meant they needed to develop a test for it.

The result was the IQ test. The IQ test they invented — I’m mostly thinking of my own childhood experience with the WISC, a later test, but almost all IQ tests use the same principles — is basically just the teacher-student dialogue I presented earlier, except that the “teacher” is a set of pre-written questions on a piece of paper, and the student has to guess what kind of answer is expected. The test is then “graded”: sorted into breeding eligibility scores, in the same way that meats and eggs are graded for sale. The grading measures how quickly the test-taker arrives at the “correct” answer, where correctness is determined entirely by the culture and language of the test-maker and assumed to be “approximately objective” in some way.

In short, IQ tests will tell you that kids one through eight are smarter than kids 9 and 10, and they will sort kids 1 through 8 to find the “best” by using a stopwatch and figuring out which kid buzzed in the predetermined “correct” answer the fastest.

This is, to put it mildly, not a good measure for selective breeding.

Remember back to our earlier discussion, where we concluded that what really matters is how “useful” an answer is, and “usefulness” is subjective. We kind of wondered if student 10 might be the “most” intelligent, since they were the only student of ten who didn’t know what the teacher wanted ahead of time yet was able to puzzle it out, but it seemed more likely that “intelligence” just flat out doesn’t exist.

But the guessing process used by student 10 is random: he tries throwing stuff at the wall, starting with the stuff he thinks is most likely to work. He guesses at what the teacher’s implied question might be, and then he guesses at which facts might conceivably lead to an answer, and then he guesses at the steps needed to combine the facts into an answer. Each and every step here involves a cascade of tiny, almost-but-not-quite-meaningless random decisions, and when a step is failing to pan out, he backtracks and tries a different guess in an earlier step. And yet, it’s entirely possible that the only reason student 10 did better than student 9, made better guesses than student 9, was that student 10 had more of the cultural context, even if it was less than the other eight.

Something I neglected to tell you until now: student 10 was autistic. Me as a kid, in fact.

My best understanding of autism, as it relates to me and my fellow autists, is that each child with autism is assigned a random miswiring of interconnections between various brain modules, with a bias toward the connections being unusually strong. Evolution has carefully tuned our gene pool to create a “neurotypical” brain pattern that is very effective at achieving gene propagation (the one and only “goal” of biological evolution) under the most probable subjective experiences that a human being might be born into. Autism and other neurodivergences, then, are basically an insurance policy, a hedge against black swan events that could otherwise cause a 100%-neurotypical population to die out.

How does that work?

When you start crossing breeding pairs that express the same trait, what you’re doing is you’re putting all of these mutations into the same bloodline. A mother who’s a little unusual in direction A has kids with a father who’s a little unusual in direction B, and the result is a kid who’s weird in both directions at the same time.

If you’re a gene pool in a black swan event, that random cross-breeding is not a bad thing: population bottlenecks naturally cause inbreeding, and inbreeding increases the odds that two rare alleles will encounter one another in the same person, so population bottlenecks turbocharge the power of natural selection. You temporarily get more mutant trait variation than you would have gotten from the larger ancestral population, and that’s true even if the initial depopulation event was completely uncorrelated with any such variant. It’s good for you-the-gene-pool, even though most of those weird new allele combinations are extremely bad for the individuals who have them. The bottleneck gives the gene pool one last chance to invent a solution — a shocking change in habits or lifestyle or ecological niche that will escape extinction. It’s a temporary solution at best, because your accumulated random mutations are a finite resource that you are now burning through quickly, but it’s a rational gamble to make if you are a threatened gene pool.

But if you’re not going through an extinction event, you might just be creating new neurological disorders that the kids are then forced to live with.

This is very relevant to me personally. I am reasonably sure that my own paternal grandmother was a eugenicist, based on how vocal she was about her Mensa membership and her trivia knowledge and her Jeopardy appearance and her quick wits and her French-Nordic heritage, and also on how she had six (!!!) kids with a violent animal-torturing alcoholic husband who happened to be highly intelligent, and also on how proud she was of her two most “intelligent” grandchildren — me and my cousin Eric — to the exclusion of our siblings, whom she only expressed vague disappointment toward. As a result of her meddling in my father’s mate choice, I got dealt a weird-ass funky hand in the genetic poker game that no one knows what to do with, least of all me or my doctor, as I try to navigate all the major health problems that came for the ride: endocrine weirdness that has damaged my adrenal glands, major allergy / asthma / immune system problems, and two connective tissue disorders (one from each parent) that almost-but-not-quite cancel each other out in all joints except my hips. To boot, even though I might be more “intelligent” on some of the measures she cared about, I’m slow, which is something that she hated. Some things that others figure out in seconds, I take minutes or weeks or even years to do the same… and the common understanding of “intelligence” says that slow is less intelligent than fast. On the other hand, there are other things that are just so obvious that I don’t think they’re worth mentioning, that most other people would need minutes or weeks or years to figure those out. It’s not really that I’m any smarter! I’m just weird, and sometimes I’m lucky that my weird brain is exactly what my team or group needs to get out of a mental rut or blind spot, and other times I’m unlucky and everyone else runs intellectual rings around me.

The moral of the story is: for fuck’s sake, selective breeding sucks, that’s how you get Habsburgs, you don’t want Habsburgs. And if you do selective breeding for the sake of intelligence, the human beings resulting from it will be miserable weirdos whose selectively bred traits will have almost nothing to do with the “intelligence” you were looking for, and may even be “stupider” on the things you care about. (Fuck Jeopardy.) Let people do their thing, because evolution-provided instincts are better at it than you are.