Eugenics Powers IQ and AI
February 27, 2024 12:00 PM Subscribe
What kind of intelligence is valued in AI? Writing for Public Books in 2021, Natasha Stovall (previously) asked us to consider whether the claim that conceptually undergirds IQ—that "human intelligence is universal, hierarchical, measurable"—is reified in the development of AI. The answer seems clear from today's perspective; we use the same terminology to talk about AI advances as we do "gifted" individuals (e.g., verbal comprehension, perceptual reasoning, processing speed, working memory.) More provocatively, Stovall charges that such a "reductive definition of human ability" has a coherent lineage from eugenics through the popularization of IQ and on to today's version of AI—and that all of the above are rooted in whiteness.
Intelligence and cognitive testing in the US is mostly used by schools to fund special education services, which is currently budgeted at 15.5 billion dollars.
posted by Brian B. at 1:09 PM on February 27 [1 favorite]
posted by Brian B. at 1:09 PM on February 27 [1 favorite]
This seems like a pretty confused article, and the author has a pretty clear agenda. To pick apart but one piece of this:
> The instruments and practices derived from Terman’s work—IQ tests, standardized tests (including the SAT), Gifted and Talented programs—continue to promote and solidify a hierarchical distribution of power. ... We have cleansed the racial language from the study of intelligence—and eugenicists from the history of psychology and the sciences—but left the eugenic core unscathed in theory and practice.
If these practices are rooted in eugenics and "whiteness", as the article contends, what should we make of the fact that Asian-Americans consistently outperform all other ethnic groups in IQ and standardized testing?
posted by osmond_nash at 1:15 PM on February 27 [6 favorites]
> The instruments and practices derived from Terman’s work—IQ tests, standardized tests (including the SAT), Gifted and Talented programs—continue to promote and solidify a hierarchical distribution of power. ... We have cleansed the racial language from the study of intelligence—and eugenicists from the history of psychology and the sciences—but left the eugenic core unscathed in theory and practice.
If these practices are rooted in eugenics and "whiteness", as the article contends, what should we make of the fact that Asian-Americans consistently outperform all other ethnic groups in IQ and standardized testing?
posted by osmond_nash at 1:15 PM on February 27 [6 favorites]
If these practices are rooted in eugenics and "whiteness", as the article contends, what should we make of the fact that Asian-Americans consistently outperform all other ethnic groups in IQ and standardized testing?
It seems like you have a pretty clear agenda too!
Here's a question: if you haven't surveyed the easily-accessible research on this topic, why would you default to the eugenicist's position?
Hey, just askin' questions!
posted by klanawa at 1:27 PM on February 27 [19 favorites]
It seems like you have a pretty clear agenda too!
Here's a question: if you haven't surveyed the easily-accessible research on this topic, why would you default to the eugenicist's position?
Hey, just askin' questions!
posted by klanawa at 1:27 PM on February 27 [19 favorites]
Intelligence and cognitive testing in the US is mostly used by schools to fund special education services, which is currently budgeted at 15.5 billion dollars.
...but also still used to decide who gets to go to prestigious private schools and the hi-cap public programs.
I do think it's interesting and worthwhile to shine a light on the lineage from eugenics to modern day AI boosters. Come to think of it, it's a little surprising that Elon Musk hasn't announced a eugenics company, since he's exactly that kind of dipshit. I'm putting it on the 2025 bingo card.
posted by qxntpqbbbqxl at 1:37 PM on February 27 [5 favorites]
...but also still used to decide who gets to go to prestigious private schools and the hi-cap public programs.
I do think it's interesting and worthwhile to shine a light on the lineage from eugenics to modern day AI boosters. Come to think of it, it's a little surprising that Elon Musk hasn't announced a eugenics company, since he's exactly that kind of dipshit. I'm putting it on the 2025 bingo card.
posted by qxntpqbbbqxl at 1:37 PM on February 27 [5 favorites]
This argument leaps from "there have been people who have measured intelligence in ways that unfairly favored white people" to "measuring intelligence will necessarily favor white people".
At best, the logical leap is unjustified. At worst, it's one of the most offensive I've ever heard.
posted by foursentences at 1:41 PM on February 27 [6 favorites]
At best, the logical leap is unjustified. At worst, it's one of the most offensive I've ever heard.
posted by foursentences at 1:41 PM on February 27 [6 favorites]
man when did this place become reddit re discussions of structural racism?
the relentless 101-ness of the comments above is shocking.
posted by lalochezia at 1:57 PM on February 27 [18 favorites]
the relentless 101-ness of the comments above is shocking.
posted by lalochezia at 1:57 PM on February 27 [18 favorites]
Wow. Lotta IQ defenders here.
Did someone put up bat -signal but instead of a bat it was a set of head calipers or something?
posted by Artw at 1:59 PM on February 27 [22 favorites]
Did someone put up bat -signal but instead of a bat it was a set of head calipers or something?
posted by Artw at 1:59 PM on February 27 [22 favorites]
This argument leaps from "there have been people who have measured intelligence in ways that unfairly favored white people" to "measuring intelligence will necessarily favor white people".
No, the argument is that intelligence as Terman defined it and as the LLM folks define it when they claim to be talking about "generalized AI" is not a single real simple thing that can be measured and is not more important than a lot of other types of knowing and understanding of the world that are neglected by that model of intelligence (which in the US is inextricably connected to whiteness).
posted by hydropsyche at 2:07 PM on February 27 [11 favorites]
No, the argument is that intelligence as Terman defined it and as the LLM folks define it when they claim to be talking about "generalized AI" is not a single real simple thing that can be measured and is not more important than a lot of other types of knowing and understanding of the world that are neglected by that model of intelligence (which in the US is inextricably connected to whiteness).
posted by hydropsyche at 2:07 PM on February 27 [11 favorites]
"I hate the impudence of a claim that in fifty minutes you can judge and classify a human being’s predestined fitness in life"
OK, agreed. But after reading that, I don't know what currently-occurring type of judgement she objects to and what she proposes to replace it with. Both 1) there is too much judgement and 2) we're doing the judgement wrong are reasonable positions.
We could reduce stratification by, for example, choosing Harvard students by lottery among qualified applicants, encouraging companies to recruit from a wider variety of schools, and hiring based more on experience than credentials. But there's always going to be some judgment needed, and there's strong evidence that standardized tests (which are a form of IQ tests) are less discriminatory than other methods of evaluation.
Full disclosure: When applying for jobs out of college, all my interviews were cut quite short, but I got a job with a company that screened via a test that echoed the work that they actually needed doing. That work was to analyze data and write about it, so it wasn't too different than an IQ test, but I could imagine jobs and tests that would differ more.
posted by Mr.Know-it-some at 2:11 PM on February 27 [3 favorites]
OK, agreed. But after reading that, I don't know what currently-occurring type of judgement she objects to and what she proposes to replace it with. Both 1) there is too much judgement and 2) we're doing the judgement wrong are reasonable positions.
We could reduce stratification by, for example, choosing Harvard students by lottery among qualified applicants, encouraging companies to recruit from a wider variety of schools, and hiring based more on experience than credentials. But there's always going to be some judgment needed, and there's strong evidence that standardized tests (which are a form of IQ tests) are less discriminatory than other methods of evaluation.
Full disclosure: When applying for jobs out of college, all my interviews were cut quite short, but I got a job with a company that screened via a test that echoed the work that they actually needed doing. That work was to analyze data and write about it, so it wasn't too different than an IQ test, but I could imagine jobs and tests that would differ more.
posted by Mr.Know-it-some at 2:11 PM on February 27 [3 favorites]
“Whiteness” is a useless and provincial as a way to think about AI. Dare I say it’s almost racist in the way it appears to center the preoccupations of white American leftists in thinking about an industry which is global and which is dominated by non-white engineers and scientists, in the US, to say the least of abroad. At best you are getting Asian-American engineers and executives to play along a bit, as in the extraordinary disaster of the bowdlerized Google AI agent … at worse they would regulate and ESG self-regulate American companies into irrelevance as people are forced to use Asian or Russia AI solutions that actually work.
posted by MattD at 2:39 PM on February 27 [2 favorites]
posted by MattD at 2:39 PM on February 27 [2 favorites]
As long as people who are writing the checks subscribe to a particular ideology, I don't think it matters who the employees who implement it are.
That said, there's a lot of "magic" around the coverage of LLMs, where people truly believe this is the future when in reality, it's the first large roll-out of a technology that is extremely mutable as costs go down and become more open source.
In other words, you can hem and haw about the current "AI" (when really discussing one or two firms LLMs) as being ideological, but you could have said the same thing about literature or videogames at the beginning, once it spreads to be in the hands of the many that will fade away. At leas that's the future I see.
Investors and owners of LLM "AI" firms prefer to see it differently, where they control and own the technology, become insanely rich and powerful, and we all pay them to use it and must except their ideology. I think that would be ahistorical for a technology of this general type.
posted by chaz at 3:02 PM on February 27 [3 favorites]
That said, there's a lot of "magic" around the coverage of LLMs, where people truly believe this is the future when in reality, it's the first large roll-out of a technology that is extremely mutable as costs go down and become more open source.
In other words, you can hem and haw about the current "AI" (when really discussing one or two firms LLMs) as being ideological, but you could have said the same thing about literature or videogames at the beginning, once it spreads to be in the hands of the many that will fade away. At leas that's the future I see.
Investors and owners of LLM "AI" firms prefer to see it differently, where they control and own the technology, become insanely rich and powerful, and we all pay them to use it and must except their ideology. I think that would be ahistorical for a technology of this general type.
posted by chaz at 3:02 PM on February 27 [3 favorites]
I'll just do a quick drive-by to point out that Asian American isn't usually hyphenated these days.
posted by TwoStride at 3:05 PM on February 27 [2 favorites]
posted by TwoStride at 3:05 PM on February 27 [2 favorites]
No, the argument is that intelligence as Terman defined it and as the LLM folks define it when they claim to be talking about "generalized AI" is not a single real simple thing that can be measured and is not more important than a lot of other types of knowing and understanding of the world that are neglected by that model of intelligence
It can simultaneously be true that the same metric
- makes *some* useful predictions about a person's ability to reason, and therefore should not be discarded by people who are studying cognition,
- but does not capture *everything* about that person's ability to reason, and therefore should not be elevated to the exclusion of everything else.
The predictive power of g is one of the best-replicated findings in the social sciences. And absolutely, it would be a madman who thought it was the only thing worth knowing about a mind.
I have no idea when any of this is supposed to have become controversial.
posted by foursentences at 3:35 PM on February 27 [7 favorites]
It can simultaneously be true that the same metric
- makes *some* useful predictions about a person's ability to reason, and therefore should not be discarded by people who are studying cognition,
- but does not capture *everything* about that person's ability to reason, and therefore should not be elevated to the exclusion of everything else.
The predictive power of g is one of the best-replicated findings in the social sciences. And absolutely, it would be a madman who thought it was the only thing worth knowing about a mind.
I have no idea when any of this is supposed to have become controversial.
posted by foursentences at 3:35 PM on February 27 [7 favorites]
I too tire of the 101 version of this conversation. Many people do not fully engage with the argument, and it gets reduced to a caricature—"now they're saying that time* is white."
*substitute whatever thing is being critiqued here. Generalized intelligence, economic theory, language rules.
In this case, I feel like there are so many interesting questions to explore if you engage a little further:
*substitute whatever thing is being critiqued here. Generalized intelligence, economic theory, language rules.
In this case, I feel like there are so many interesting questions to explore if you engage a little further:
- If world history had unfolded differently, and Malawi (chosen with a random country generator) had the hegemonic position that the US does, how would the content of the internet be different?
- In that alternate universe, would the LLMs be any different?
- If the alternate universe LLM and our universe's LLM both accurately captured the intelligence of their internet and they both took an intelligence test from our universe, what would that finding mean?
Oh cool I get to talk about testing validity at home and not just at work with people trapped in a small break room with me. (True story. Happened today. It was about the Connors Continuous Performance Test and the TOVA.)
But there's always going to be some judgment needed, and there's strong evidence that standardized tests (which are a form of IQ tests)
No they aren’t. They’re achievement tests. IQ and achievement testing is vastly different. While there is debate about whether achievement testing predicts success, it is basically agreed upon that they do measure what they’re trying to measure—academic abilities. (Just because you have the academic abilities doesn’t mean you’ll succeed at college, though. Many other factors.)
But regarding IQ, most psychologists trained in assessment agree that it is terribly flawed and not really measuring what we actually think of as “intelligence.” I was so baffled to see “IQ testing is structurally racist and not a useful measure of what we consider intelligence” presented as a new and shocking idea that I looked up the author thinking she must not be trained in the psychology field—this was like, shit we went over in first year assessment classes. She is a clinical psychologist, apparently, but she was trained in Russia and I admittedly have zero idea what Russian psychologist training is like.
If these practices are rooted in eugenics and "whiteness", as the article contends, what should we make of the fact that Asian-Americans consistently outperform all other ethnic groups in IQ and standardized testing?
Asian Americans do not usually outperform on IQ testing, only on achievement testing. There is an entire body of research dedicated to the fact that Asian Americans regularly test far about their IQ on achievement measures (which is part of the whole argument about whether IQ is measuring what we think it is measuring).
Anyone trained in measuring IQ can see it’s obviously culturally loaded—some tests more than others, but the most common IQ testing measure has questions like “who wrote Sherlock Holmes?” and “who wrote Alice in Wonderland?” I’m not joking, your knowledge of Old White Guy canon is part of one of the most widely used measures (APA don’t come handcuff me about test security…), and your answer to this weighs into your intelligence score. I hope I don’t need to explain how that’s wildly culturally specific and why, perhaps, children whose access to reading materials varies significantly from the traditional middle class nuclear family’s might not know the answer despite knowing many other things just as relevant to intelligence.
makes *some* useful predictions about a person's ability to reason, and therefore should not be discarded by people who are studying cognition,
For some people. For people who aren’t middle class white children with no disabilities, it’s wildly incorrect in ways that are actively harmful. It’s not a penny jar where every little bit helps; sometimes you’re pouring molten metal in the jar and now it’s all useless.
posted by brook horse at 4:32 PM on February 27 [28 favorites]
But there's always going to be some judgment needed, and there's strong evidence that standardized tests (which are a form of IQ tests)
No they aren’t. They’re achievement tests. IQ and achievement testing is vastly different. While there is debate about whether achievement testing predicts success, it is basically agreed upon that they do measure what they’re trying to measure—academic abilities. (Just because you have the academic abilities doesn’t mean you’ll succeed at college, though. Many other factors.)
But regarding IQ, most psychologists trained in assessment agree that it is terribly flawed and not really measuring what we actually think of as “intelligence.” I was so baffled to see “IQ testing is structurally racist and not a useful measure of what we consider intelligence” presented as a new and shocking idea that I looked up the author thinking she must not be trained in the psychology field—this was like, shit we went over in first year assessment classes. She is a clinical psychologist, apparently, but she was trained in Russia and I admittedly have zero idea what Russian psychologist training is like.
If these practices are rooted in eugenics and "whiteness", as the article contends, what should we make of the fact that Asian-Americans consistently outperform all other ethnic groups in IQ and standardized testing?
Asian Americans do not usually outperform on IQ testing, only on achievement testing. There is an entire body of research dedicated to the fact that Asian Americans regularly test far about their IQ on achievement measures (which is part of the whole argument about whether IQ is measuring what we think it is measuring).
Anyone trained in measuring IQ can see it’s obviously culturally loaded—some tests more than others, but the most common IQ testing measure has questions like “who wrote Sherlock Holmes?” and “who wrote Alice in Wonderland?” I’m not joking, your knowledge of Old White Guy canon is part of one of the most widely used measures (APA don’t come handcuff me about test security…), and your answer to this weighs into your intelligence score. I hope I don’t need to explain how that’s wildly culturally specific and why, perhaps, children whose access to reading materials varies significantly from the traditional middle class nuclear family’s might not know the answer despite knowing many other things just as relevant to intelligence.
makes *some* useful predictions about a person's ability to reason, and therefore should not be discarded by people who are studying cognition,
For some people. For people who aren’t middle class white children with no disabilities, it’s wildly incorrect in ways that are actively harmful. It’s not a penny jar where every little bit helps; sometimes you’re pouring molten metal in the jar and now it’s all useless.
posted by brook horse at 4:32 PM on February 27 [28 favorites]
If anyone asks though my favorite IQ testing measure is the KBIT-2, because one of the questions asks people to define “discord” and “What, like, in the Discord channels?” is my favorite answer to any test question I’ve ever given.
(I gave IQ measures a lot as part of my training; I don’t use them now except when wanting to look for “spiky profiles” and not as a comparison to other individuals.)
posted by brook horse at 4:35 PM on February 27 [3 favorites]
(I gave IQ measures a lot as part of my training; I don’t use them now except when wanting to look for “spiky profiles” and not as a comparison to other individuals.)
posted by brook horse at 4:35 PM on February 27 [3 favorites]
Oh, to engage with the article though. The fact that we measure LLM intelligence based on verbal reasoning*, and specifically verbal reasoning as expressed through white American cultural standards, leads to some interesting problems. I would be particularly curious whether LLMs understand and communicate well in AAVE, for example. I entered an example sentence from a paper discussing AAVE structure into ChatGPT and it understood it well; however, it responded in standard American English. I asked it to respond in AAVE and it agreed to this (honestly wasn’t sure it would). I entered the same sentence and this is what I got:
*Especially because verbal reasoning doesn’t necessarily equate to high IQ. I’ve got a very high verbal reasoning score but my overall IQ is average. Which is enough to get you through a PhD; meanwhile, I know multiple people with bonkers high IQ who beat me across the board on all IQ sub-tests, and they all struggled to get through college.
posted by brook horse at 5:22 PM on February 27 [6 favorites]
Yo, if this doctor keeps leaving us on read, he gon’ catch hands, on gang.I don’t know enough about AAVE to say how accurate it is, though I frequently work with individuals who use it and my instinctual reaction was that it sounds wrong. But I don’t know if that’s because it’s pulling (necessarily) from online communication (and I’m chronically offline), which varies significantly from in-person. Would welcome any linguist opinions.
I feel you, it's mad frustrating when the doc keeps ghostin' us. On gang, communication should be on point, especially in healthcare. Maybe hit 'em up again or check with the office. Keeping it real, violence ain't the answer, but making sure they hear you is. Stay cool, fam.
*Especially because verbal reasoning doesn’t necessarily equate to high IQ. I’ve got a very high verbal reasoning score but my overall IQ is average. Which is enough to get you through a PhD; meanwhile, I know multiple people with bonkers high IQ who beat me across the board on all IQ sub-tests, and they all struggled to get through college.
posted by brook horse at 5:22 PM on February 27 [6 favorites]
I know multiple people with bonkers high IQ who beat me across the board on all IQ sub-tests, and they all struggled to get through college.
I am only guessing that at least some of those bonkers high IQ people were neurodiverse. College can be an unhealthy and overwhelming place for neurodiverse people.
posted by Cardinal Fang at 10:46 PM on February 27
I am only guessing that at least some of those bonkers high IQ people were neurodiverse. College can be an unhealthy and overwhelming place for neurodiverse people.
posted by Cardinal Fang at 10:46 PM on February 27
[Intelligence metrics make useful predictions] for some people. For people who aren’t middle class white children with no disabilities, it’s wildly incorrect in ways that are actively harmful. It’s not a penny jar where every little bit helps; sometimes you’re pouring molten metal in the jar and now it’s all useless.
What? So you use the best available statistical methods to unconfound the data, and you discuss any given metric only as to populations for which it's valid. Probably you limit your conclusions to within-group comparisons. Certainly you don't throw away valuable information because of some weird metallurgical metaphor.
the most common IQ testing measure has questions like “who wrote Sherlock Holmes?”
Which IQ test are you referring to?
posted by foursentences at 10:54 PM on February 27 [3 favorites]
What? So you use the best available statistical methods to unconfound the data, and you discuss any given metric only as to populations for which it's valid. Probably you limit your conclusions to within-group comparisons. Certainly you don't throw away valuable information because of some weird metallurgical metaphor.
the most common IQ testing measure has questions like “who wrote Sherlock Holmes?”
Which IQ test are you referring to?
posted by foursentences at 10:54 PM on February 27 [3 favorites]
This is why Metafilter stays so white. This constant sea-lioning of racism is tiresome as fuck. Its funny too, 'cause I work in a white environment. Y'all never pop off about this bullshit in my face. It's only online that you feel safe and comfortable enough to play this dumb shit. This is not only ignorant and racist. Its cowardly.
posted by anansi at 2:30 AM on February 28 [14 favorites]
posted by anansi at 2:30 AM on February 28 [14 favorites]
Oh cool I get to talk about testing validity at home and not just at work with people trapped in a small break room with me.
If you're trapped in a room with people at work, it's not a break.
posted by Cardinal Fang at 3:08 AM on February 28 [1 favorite]
If you're trapped in a room with people at work, it's not a break.
posted by Cardinal Fang at 3:08 AM on February 28 [1 favorite]
I read an article (can't find it) a while ago about how the problem is that people think LLMs are 'intelligent' because they're using the same incorrect metric they use to evaluate people's intelligence: language proficiency. This adversely affects the percieved intelligence (and trustworthiness, etc) of people with different kinds of disabilities, people whose first language isn't the one they're being evaluated in, people who speak non-prestige variations of their language, etc.
This is of course replicated in LLM's training, and it forms a tech-loop with society's pre-existing biases, reinforcing them and being reinforced by them.
posted by signal at 5:33 AM on February 28 [3 favorites]
This is of course replicated in LLM's training, and it forms a tech-loop with society's pre-existing biases, reinforcing them and being reinforced by them.
posted by signal at 5:33 AM on February 28 [3 favorites]
What? So you use the best available statistical methods to unconfound the data, and you discuss any given metric only as to populations for which it's valid. Probably you limit your conclusions to within-group comparisons. Certainly you don't throw away valuable information because of some weird metallurgical metaphor.
”I am only going to study white Americans with no disabilities” has been the modus operandi of research for decades and we are trying to get away from that. Saying you’re doing it because the measures aren’t valid for anyone else doesn’t make it any better. You could spend your time coming up with valid measures but if you would prefer to only study the same sub-population we’ve always studied, and shrug your shoulders when inevitably that gets extrapolated to mean this is How Behavior Works, you’re welcome to. But I don’t have to like it.
Which IQ test are you referring to?
Wechsler.
If you're trapped in a room with people at work, it's not a break.
Also none of us were on break. The break room and the room for doing documentation outside of patient view is the same. Hospitals are great!
posted by brook horse at 6:36 AM on February 28 [3 favorites]
”I am only going to study white Americans with no disabilities” has been the modus operandi of research for decades and we are trying to get away from that. Saying you’re doing it because the measures aren’t valid for anyone else doesn’t make it any better. You could spend your time coming up with valid measures but if you would prefer to only study the same sub-population we’ve always studied, and shrug your shoulders when inevitably that gets extrapolated to mean this is How Behavior Works, you’re welcome to. But I don’t have to like it.
Which IQ test are you referring to?
Wechsler.
If you're trapped in a room with people at work, it's not a break.
Also none of us were on break. The break room and the room for doing documentation outside of patient view is the same. Hospitals are great!
posted by brook horse at 6:36 AM on February 28 [3 favorites]
I have no idea when any of this is supposed to have become controversial.
You're about 50 years behind the cutting edge at this point. Maybe ask yourself how you managed to get so out of touch? Can you think of any kind of systematic bias that might be clouding your judgement here?
posted by grog at 6:51 AM on February 28 [3 favorites]
You're about 50 years behind the cutting edge at this point. Maybe ask yourself how you managed to get so out of touch? Can you think of any kind of systematic bias that might be clouding your judgement here?
posted by grog at 6:51 AM on February 28 [3 favorites]
Yeah, to re-iterate, the basic points the author is touching on are taught at graduate level psychology classes across the US. It’s not actually not controversial, it’s the standard understanding of IQ testing. This is part of why anyone with a high school degree can be a psychometrician—you don’t need any further education to administer an IQ test. Interpreting the results of an IQ test, however, has to be done by a PhD-level clinician because of this very point. You can’t just take the numbers at face value. You need clinical judgement. Which also means that the numbers, at face value, are not inherently meaningful. Can they provide some useful information in the broader context of a larger assessment? Sure, sometimes. But the numbers themselves don’t mean anything like what we think they mean.
posted by brook horse at 8:08 AM on February 28 [3 favorites]
posted by brook horse at 8:08 AM on February 28 [3 favorites]
I always wanted to devise an index of general athleticism a and use it to argue that Refrigerator Perry could win the Tour de France and Nairo Quintana could win the superbowl. Can't see any problem with that.
posted by klanawa at 8:18 AM on February 28 [8 favorites]
posted by klanawa at 8:18 AM on February 28 [8 favorites]
Klanawa if I hadn’t given up teaching I would be using that example in every psychology class I ever taught (citing you, of course). That’s beautiful.
posted by brook horse at 8:35 AM on February 28 [5 favorites]
posted by brook horse at 8:35 AM on February 28 [5 favorites]
i love being asian diaspora with high iq as someone who attended an elite american university because so many white people like trying to use those like me as cudgels to protect whiteness and the moneyed classes even in "liberal" and "left-leaning" environments.
iq is hollow bullshit and, like the sats/acts, measures for a very specific definition of intelligence that is favored in a manner that works reasonably well for an industrialized, capitalist society if there aren't any other confounding factors like neurodivergence, gender, sex, race, class, networking ability, luck, motivation...
and as far as claiming that dei/esg is going to wreck american/western ai efforts is not just a dogwhistle but a fucking racist blowhorn and it's shocking that a vile comment like that is allowed to stand. just go ahead and claim that critical race theory and gender ideology is destroying our competitiveness while you're at it.
posted by i used to be someone else at 8:38 AM on February 28 [11 favorites]
iq is hollow bullshit and, like the sats/acts, measures for a very specific definition of intelligence that is favored in a manner that works reasonably well for an industrialized, capitalist society if there aren't any other confounding factors like neurodivergence, gender, sex, race, class, networking ability, luck, motivation...
and as far as claiming that dei/esg is going to wreck american/western ai efforts is not just a dogwhistle but a fucking racist blowhorn and it's shocking that a vile comment like that is allowed to stand. just go ahead and claim that critical race theory and gender ideology is destroying our competitiveness while you're at it.
posted by i used to be someone else at 8:38 AM on February 28 [11 favorites]
If I were to pick a common thread between IQ and LLM hype, it would be the very strong risk of mistaking the map for the territory.
In relation to whiteness, I will go with the take that systems developed under some dominant culture are predisposed to reinforcing the dominance of that culture.
posted by RobotHero at 9:58 AM on February 28 [4 favorites]
In relation to whiteness, I will go with the take that systems developed under some dominant culture are predisposed to reinforcing the dominance of that culture.
posted by RobotHero at 9:58 AM on February 28 [4 favorites]
...but also still used to decide who gets to go to prestigious private schools and the hi-cap public programs.
If that were only the case, then it would directly compete with other methods of family influence. Many prestigious colleges gave up admissions testing requirements for liability reasons apparently, having to defend in court why so many non-legacy applicants weren't admitted with the highest scores. As for AI, it will do far more damage to knowledge workers connected by college degrees to the economy than anyone else.
posted by Brian B. at 10:17 AM on February 28
If that were only the case, then it would directly compete with other methods of family influence. Many prestigious colleges gave up admissions testing requirements for liability reasons apparently, having to defend in court why so many non-legacy applicants weren't admitted with the highest scores. As for AI, it will do far more damage to knowledge workers connected by college degrees to the economy than anyone else.
posted by Brian B. at 10:17 AM on February 28
iq is hollow bullshit
Here are some statements taken from an article arguing against the usefulness of IQ. There's plenty to argue against, but it's, to steal a phrase, "hollow bullshit" to simply dismiss it as meaningless.
"It is empirically true that the cognitive abilities that are measured on IQ tests are positively correlated with each other, giving rise to a 'general intelligence factor.' It is also true that the cognitive skills that are most strongly related to IQ (e.g., abstract reasoning, working memory, vocabulary, visual-spatial mental rotation) are highly general in the sense that they facilitate the speed and efficiency of learning novel and complex information across a wide range of contexts. It's difficult to imagine a situation in which one is conscious and processing information that doesn't draw whatsoever on these cognitive skills.
'IQ is related, probably more so than any other single measurable human trait, to many important educational, occupational, economic, and social outcomes.'
This is statistically true. ... IQ correlates positively with family income, socioeconomic status, school and occupational performance, military training assignments, law-abidingness, healthful habits, illness, and morality. In contrast, IQ is negatively correlated with welfare, psychopathology, crime, inattentiveness, boredom, delinquency, and poverty.
IQ tests do not simply index family background. IQ and SAT scores are still correlated with important academic and societal outcomes even after taking into account socioeconomic status."
posted by Mr.Know-it-some at 12:49 PM on February 28 [3 favorites]
Here are some statements taken from an article arguing against the usefulness of IQ. There's plenty to argue against, but it's, to steal a phrase, "hollow bullshit" to simply dismiss it as meaningless.
"It is empirically true that the cognitive abilities that are measured on IQ tests are positively correlated with each other, giving rise to a 'general intelligence factor.' It is also true that the cognitive skills that are most strongly related to IQ (e.g., abstract reasoning, working memory, vocabulary, visual-spatial mental rotation) are highly general in the sense that they facilitate the speed and efficiency of learning novel and complex information across a wide range of contexts. It's difficult to imagine a situation in which one is conscious and processing information that doesn't draw whatsoever on these cognitive skills.
'IQ is related, probably more so than any other single measurable human trait, to many important educational, occupational, economic, and social outcomes.'
This is statistically true. ... IQ correlates positively with family income, socioeconomic status, school and occupational performance, military training assignments, law-abidingness, healthful habits, illness, and morality. In contrast, IQ is negatively correlated with welfare, psychopathology, crime, inattentiveness, boredom, delinquency, and poverty.
IQ tests do not simply index family background. IQ and SAT scores are still correlated with important academic and societal outcomes even after taking into account socioeconomic status."
posted by Mr.Know-it-some at 12:49 PM on February 28 [3 favorites]
Mr. Know-it-some, are you trained in providing IQ testing? Because I am, and in fact there is a WISC kit in front of me and a WAIS kit behind me right now. Your points are understandable but missing a significant amount of information, which I will provide here. Brief note for all: full-scale IQ is what people usually mean when they say “IQ.” Index scores are the scores given for performance in verbal comprehension, visual spatial reasoning, fluid reasoning, working memory, and processing speed domains—these all combine to form the FSIQ. Subtest scores are the scores for individual tests in each domain—for example, verbal comprehension is made up of Similarities and Vocabulary. So you combine subtests into domain scores and then domain scores into the FSIQ.
1) Subtests on the WISC are positively correlated, but none achieve a correlation higher than 0.55. That correlation is between Visual Puzzles and Block Design, the two visual spatial reasoning subtests. Generally within primary index scales, the subtests correlate with each other. Similarities and Vocabulary, the two verbal reasoning subtests, correlate at 0.54. But neither of the verbal reasoning subtests correlate above 0.35 with either of the visual spatial reasoning subtests. This is the pattern across subtests. Within index scores, there is a moderate (but not high) correlation, but across indices the correlation is low. Source
2) Processing speed correlates particularly poorly with other index scores, and in fact is the least correlated with full scale IQ at 0.56; working memory and visual spatial reasoning come second at 0.68. This is in comparison to fluid reasoning (understanding conceptual relationships between pictures) at 0.80 and verbal reasoning at 0.78. This is contrary to what even the article assumes would be the most “necessary” factors for IQ. Source: same as above
3) Over 80% of the sample used to standardize the WISC-III had a variable profile, meaning that their index scores were statistically significantly different from each other and did not correlate with their full-scale IQ. Less than 20% had a flat profile where all of their index scores clustered together such that it reflected a generalized intelligence. For everyone else, the full-scale IQ was not an accurate representation of their abilities. Source
The evidence for generalized intelligence as a concept is low. This is not even taking into account the cultural issues. The actual test data itself does not support the idea that generalized IQ is a trait, or at least not one that we are actually appropriately measuring.
“IQ scores do not simply index family background” is entirely true. But so is “IQ scores don’t actually measure what we think of as intelligence” and so is “generalized intelligence as a concept isn’t supported.”
posted by brook horse at 1:59 PM on February 28 [13 favorites]
1) Subtests on the WISC are positively correlated, but none achieve a correlation higher than 0.55. That correlation is between Visual Puzzles and Block Design, the two visual spatial reasoning subtests. Generally within primary index scales, the subtests correlate with each other. Similarities and Vocabulary, the two verbal reasoning subtests, correlate at 0.54. But neither of the verbal reasoning subtests correlate above 0.35 with either of the visual spatial reasoning subtests. This is the pattern across subtests. Within index scores, there is a moderate (but not high) correlation, but across indices the correlation is low. Source
2) Processing speed correlates particularly poorly with other index scores, and in fact is the least correlated with full scale IQ at 0.56; working memory and visual spatial reasoning come second at 0.68. This is in comparison to fluid reasoning (understanding conceptual relationships between pictures) at 0.80 and verbal reasoning at 0.78. This is contrary to what even the article assumes would be the most “necessary” factors for IQ. Source: same as above
3) Over 80% of the sample used to standardize the WISC-III had a variable profile, meaning that their index scores were statistically significantly different from each other and did not correlate with their full-scale IQ. Less than 20% had a flat profile where all of their index scores clustered together such that it reflected a generalized intelligence. For everyone else, the full-scale IQ was not an accurate representation of their abilities. Source
The evidence for generalized intelligence as a concept is low. This is not even taking into account the cultural issues. The actual test data itself does not support the idea that generalized IQ is a trait, or at least not one that we are actually appropriately measuring.
“IQ scores do not simply index family background” is entirely true. But so is “IQ scores don’t actually measure what we think of as intelligence” and so is “generalized intelligence as a concept isn’t supported.”
posted by brook horse at 1:59 PM on February 28 [13 favorites]
One of the most unquestioned aspects of "intelligence" is the fact that it's defined and measured on individuals.
There's no fundamental reason that should be the case: doing so is a choice. And it's a choice that inherently prioritises a certain set of values. I'd argue that choosing to define and test intelligence as a property of individuals is actually a pretty extreme position.
Because in general, people don't exist as individuals in the real world. They exist as part of groups: families, teams, villages, companies, tribes, couples, gangs, congregations, you name it. Humans are social creatures.
But we don't test the intelligence of groups. We almost exclusively test the ability of individuals to perform alone. And in doing so we don't value people's ability to co-operate; to function as a social unit that transcends the capabilities of its individual members.
It's actually an extremely artificial choice. For a few hours of people's lives, during which things work completely differently to the whole of the rest of their lives, we sit them down in exam halls where they may speak to nobody. And then we decide their futures based on how they perform in that completely artificial scenario.
What are the effects on society of that choice - what types of people does it favour? How much of "intelligence" is actually just measuring how much of an independent, self-centred loner you are?
And why do we choose to value that?
posted by automatronic at 4:47 PM on February 28 [8 favorites]
There's no fundamental reason that should be the case: doing so is a choice. And it's a choice that inherently prioritises a certain set of values. I'd argue that choosing to define and test intelligence as a property of individuals is actually a pretty extreme position.
Because in general, people don't exist as individuals in the real world. They exist as part of groups: families, teams, villages, companies, tribes, couples, gangs, congregations, you name it. Humans are social creatures.
But we don't test the intelligence of groups. We almost exclusively test the ability of individuals to perform alone. And in doing so we don't value people's ability to co-operate; to function as a social unit that transcends the capabilities of its individual members.
It's actually an extremely artificial choice. For a few hours of people's lives, during which things work completely differently to the whole of the rest of their lives, we sit them down in exam halls where they may speak to nobody. And then we decide their futures based on how they perform in that completely artificial scenario.
What are the effects on society of that choice - what types of people does it favour? How much of "intelligence" is actually just measuring how much of an independent, self-centred loner you are?
And why do we choose to value that?
posted by automatronic at 4:47 PM on February 28 [8 favorites]
I always wanted to devise an index of general athleticism a and use it to argue that Refrigerator Perry could win the Tour de France and Nairo Quintana could win the superbowl. Can't see any problem with that.
I'd use it to argue that a guy with only half a foot could kick the longest field goal in history.
posted by Cardinal Fang at 5:23 AM on February 29 [1 favorite]
I'd use it to argue that a guy with only half a foot could kick the longest field goal in history.
posted by Cardinal Fang at 5:23 AM on February 29 [1 favorite]
Mod note: Comment removed. There's no need to "fix" another commenter's writing, so please avoid doing that. Allow others to speak for themselves and be considerate and respectful of their words.
posted by Brandon Blatcher (staff) at 6:10 AM on February 29
posted by Brandon Blatcher (staff) at 6:10 AM on February 29
But we don't test the intelligence of groups. We almost exclusively test the ability of individuals to perform alone.
Individuals perform as individual talent on delegated tasks inside a team or society. Achievement tests were pioneered by ancient China's difficult civil service exam, to replace military rule or bribes to aristocrats. Reporting test scores as social groupings is what leads to the harms of cognitive testing. When society does education testing it is usually as a social intervention on individuals, for both emergency aid and gifted scholarships, as specialized learning opportunities.
posted by Brian B. at 12:44 PM on February 29
Individuals perform as individual talent on delegated tasks inside a team or society. Achievement tests were pioneered by ancient China's difficult civil service exam, to replace military rule or bribes to aristocrats. Reporting test scores as social groupings is what leads to the harms of cognitive testing. When society does education testing it is usually as a social intervention on individuals, for both emergency aid and gifted scholarships, as specialized learning opportunities.
posted by Brian B. at 12:44 PM on February 29
I'm not terribly versed in the 'science' and study of IQ, and in general I object to the viewpoint that has been seems to be put forward by some in this thread that deep academic understanding of a subject is necessary to participate meaningfully in a MetaFilter discussion. Why is it 'shocking' that some commenters exhibit what you may think is a '101' level of understanding of a subject? Are you suggesting that those who have not taken upper level university courses in this subject should not comment for fear of you mocking them? That, to me, seems elitist and offensive?
With that out of the way I found this article odd and confusing for a few reasons:
- This is a minor point, so I hesitate to put this first, but it was the first thing that jumped out at me. Why is there a reference to astrology with no other context or explanation (something about the confluence of Pluto and Saturn). Does anyone know why that's in there? Seemed like a total non sequitur. Is this like a thing now? It's cool to put in random astrological references within a totally different context? Am I out of the loop here?
- I don't see how an IQ test can be meaningfully used to measure an AI's 'intelligence'? Like if I were to give ChatGPT or Bard or whatever an IQ test wouldn't all of the current gen good ones measure as 'off the charts'? They have all the answers!
- It seems to me that the author's thesis is:
Silicon Valley deifies intelligence and is run by white men who use their intelligence to amass incomprehensible wealth and power. Silicon Valley is really good at AI, and the people in charge of Silicon Valley want to create an AI that exhibits high intelligence. General intelligence as a concept has racist roots and is still being used to perpetuate racism and structural inequalities. Therefore AI will perpetuate and heighten oppression of those who are already oppressed and give more power to those who already have it.
Which seems... obvious? Like why is whiteness specifically being linked to this? I imagine the AI tech coming out of China (using China as an example as I believe China is another AI hotspot) will continue to accrete power to the wealthy and powerful there, who are probably mostly Han Chinese?
The overall vibe I got from this article is: Let's rehash commonly known history and problems with IQ tests and general intelligence but let's try to jam in AI in there to make it relevant, and let's do it in a unnecessarily florid style with weird references so people feel smart when they read it. Basically clickbait for a middle to high brow audience.
posted by sid at 8:40 PM on February 29 [3 favorites]
With that out of the way I found this article odd and confusing for a few reasons:
- This is a minor point, so I hesitate to put this first, but it was the first thing that jumped out at me. Why is there a reference to astrology with no other context or explanation (something about the confluence of Pluto and Saturn). Does anyone know why that's in there? Seemed like a total non sequitur. Is this like a thing now? It's cool to put in random astrological references within a totally different context? Am I out of the loop here?
- I don't see how an IQ test can be meaningfully used to measure an AI's 'intelligence'? Like if I were to give ChatGPT or Bard or whatever an IQ test wouldn't all of the current gen good ones measure as 'off the charts'? They have all the answers!
- It seems to me that the author's thesis is:
Silicon Valley deifies intelligence and is run by white men who use their intelligence to amass incomprehensible wealth and power. Silicon Valley is really good at AI, and the people in charge of Silicon Valley want to create an AI that exhibits high intelligence. General intelligence as a concept has racist roots and is still being used to perpetuate racism and structural inequalities. Therefore AI will perpetuate and heighten oppression of those who are already oppressed and give more power to those who already have it.
Which seems... obvious? Like why is whiteness specifically being linked to this? I imagine the AI tech coming out of China (using China as an example as I believe China is another AI hotspot) will continue to accrete power to the wealthy and powerful there, who are probably mostly Han Chinese?
The overall vibe I got from this article is: Let's rehash commonly known history and problems with IQ tests and general intelligence but let's try to jam in AI in there to make it relevant, and let's do it in a unnecessarily florid style with weird references so people feel smart when they read it. Basically clickbait for a middle to high brow audience.
posted by sid at 8:40 PM on February 29 [3 favorites]
The comments about '101' level refer to people jumping into threads that talk about racism, xenofobia, homofobia, transfobia, and/or other kinds of bigotry, and making comments that come from a position of privilege and ignorance about the subject matter, either dismissing the concerns of the discriminated-against people or demanding that these people explain very basic, easy to google concepts for their satisfaction, or both.
It's exhausting.
Don't be that guy/gal/non-binary pal.
posted by signal at 2:40 AM on March 1 [4 favorites]
It's exhausting.
Don't be that guy/gal/non-binary pal.
posted by signal at 2:40 AM on March 1 [4 favorites]
Reporting test scores as social groupings is what leads to the harms of cognitive testing.
(I don't believe automatronic is talking about forming averages of arbitrary groups after the fact by pooling measurements taken of individual people; they're talking about taking measurements of specific real-world teams as teams.)
posted by a faded photo of their beloved at 1:54 PM on March 1 [2 favorites]
(I don't believe automatronic is talking about forming averages of arbitrary groups after the fact by pooling measurements taken of individual people; they're talking about taking measurements of specific real-world teams as teams.)
posted by a faded photo of their beloved at 1:54 PM on March 1 [2 favorites]
signal: I read an article (can't find it) a while ago about how the problem is that people think LLMs are 'intelligent' because they're using the same incorrect metric they use to evaluate people's intelligence: language proficiency. This adversely affects the percieved intelligence (and trustworthiness, etc) of people with different kinds of disabilities, people whose first language isn't the one they're being evaluated in, people who speak non-prestige variations of their language, etc.
The article was Language Is a Poor Heuristic for Intelligence by Karawynn Long (link is to its new hosting location). [previously]
posted by heatherlogan at 7:45 AM on March 2 [4 favorites]
The article was Language Is a Poor Heuristic for Intelligence by Karawynn Long (link is to its new hosting location). [previously]
posted by heatherlogan at 7:45 AM on March 2 [4 favorites]
Thank you!
posted by signal at 8:09 AM on March 2 [1 favorite]
posted by signal at 8:09 AM on March 2 [1 favorite]
« Older To the Moon (eventually) but with great food! | I know you will probably put it up again Newer »
This thread has been archived and is closed to new comments
posted by McBearclaw at 12:45 PM on February 27 [4 favorites]