People on Tumblr are discussing Eliezer Yudkowsky’s old essay The Correct Contrarian Cluster, and my interpretation was different enough that I thought it might be worth spelling out. So here it is: is there a General Factor of Correctness?
Remember, IQ is supposed to come from a General Factor Of Intelligence. If you make people take a lot of different tests of a lot of different types, people who do well on one type will do well on other types more often than chance. You can do this with other things too, like make a General Factor Of Social Development. If you’re really cool, you can even correlate the General Factor of Intelligence and the General Factor of Social Development together.
A General Factor Of Correctness would mean that if you asked people’s opinions on a bunch of controversial questions, like “Would increasing the minimum wage to $15 worsen unemployment?” or “Which interpretation of quantum mechanics is correct?” or “are artificial sweeteners safe?” and then somehow discovered the answers to these questions, people who did well on one such question would do well on other types more often than chance.
This is a surprisingly deep and controversial issue, but one with potentially big payoffs. Suppose you want to know whose economic theories are right, but you don’t want to take the time to learn economics. Consider some position that was once considered fringe and bizarre, but now known to be likely true – for example, pre-Clovis settlement of the New World. Find the economists who believed in pre-Clovis settlement of the New World back when doing so was unpopular. Those economists have demonstrated a proven track record of being able to winnow out correct ideas amidst a sea of uncertainty. Invest in whatever company they tell you to invest in and make a killing.
I’m sort of joking, but also sort of serious – shouldn’t something like this work? If there’s such a thing as reasoning ability, people who are good at sifting through a mess of competing claims about pre-Columbian anthropology and turning up the truth should be able to apply that same skill to sifting through a mess of competing claims about economic data. Right?
If this is true, we can gain new insight into all of our conundra just by seeing who believes what about New World migration. That sounds useful. The problem is, to identify it we have to separate it out from a lot of closely related concepts.
The first problem: if you just mark who’s right and wrong about each controversial issue, the General Factor Of Correctness will end up looking a lot like a General Factor of Agreeing With Expert Consensus. The current best-known heuristic is “always agree with expert consensus on everything”; people who follow this heuristic all the time are most likely to do well, but we learn nothing whatsoever from their success. If I can get brilliant-economist-points for saying things like “black holes exist” or “9-11 was not a government conspiracy”, then that just makes a mockery of the whole system. Indeed, our whole point in this exercise is to see if we can improve on the “agree with experts” heuristic.
We could get more interesting results by analyzing only people’s deviations from expert consensus. If you agree with the consensus about everything, you don’t get to play. If you disagree with the consensus about some things, then you get positive points when you’re right and negative points when you’re wrong. If someone ends consistently ends up with a positive score beyond what we would expect by chance, then they’re the equivalent of the economist who was surprisingly prescient about pre-Clovis migration – a person who’s demonstrating a special ability that allows them to outperform experts. This is why Eliezer very reasonably talks about a correct contrarian cluster instead of a correct cluster in general. We already know who the correct cluster is, and all of you saying “I have no idea what Clovis is, but whatever leading anthropologists think, I think that too” are in it. So what? So nothing.
The second problem: are you just going to rediscover some factor we already know about, like IQ or general-well-educatedness? I’m not sure. WHen I brought this up on Tumblr, people were quick to point out examples of very intelligent, very well-educated people believing stupid things – for example, Newton’s obsession with alchemy and Biblical prophecy, or Linus Pauling’s belief that you could solve health just be making everyone take crazy amounts of Vitamin C. These points are well-taken, but I can’t help wondering if there’s selection bias in bringing them up. Yes, some smart people believe stupid things, but maybe even more stupid people do? By analogy, many people who are brilliant at math are terrible at language, and we can all think of salient examples, but psychometrics has shown again and again that in general math and language skills are correlated.
If we look for more general data, we get inconsistent results. Neither IQ nor educational attainment seems to affect whether you believe in climate change very much, though you can get slightly different results depending on how you ask and what you adjust for. There seems to be a stronger effect of intelligence increasing comfort with nuclear power. Other polls show IQ may increase atheism, non-racism, and a complicated cluster of political views possibly corresponding to libertarianism but also showing up as “liberalism” or “conservativism” depending on how you define your constructs and which aspects of politics you focus on. I am very suspicious about any of this reflecting real improved decision-making capacity as opposed to just attempts to signal intelligence in various ways.
The third problem: can we differentiate positive from negative selection? There are lots of people who believe in Bigfoot and ESP and astrology. I suspect these people will be worse at other things, including predicting economic trends, predicting world events, and being on the right side of difficult scientific controversies, probably in a way independent of IQ or education. I’m not sure of this. But I suspect it. If I’m right, then the data will show a General Factor of Correctness, but it won’t necessarily be a very interesting one. To give a reductio ad absurdum, if you have some mental disorder that causes you to live in a completely delusional fantasy world, you will have incorrect opinions about everything at once, which looks highly correlated, but this doesn’t necessarily prove that there are correlations among the people who are more correct than average.
The fourth problem: is there a difference between correctness and probability calibration? Suppose that Alice says that there’s a 90% chance the Greek economy will implode, and Bob has the same information but says there’s only an 80% chance. Here it might be tempting to say that one of either Alice or Bob is miscalibrated – either Alice is overconfident or Bob is underconfident. But suppose Alice says that there’s a 90% chance the Greek economy will implode, and Bob has the same information but says there’s only a 10% chance that it will. Now we’re more likely to interpret this in terms of them just disagreeing. But I don’t know enough about probability theory to put my finger on whether there’s a true qualitative difference.
This is important because we know calibration is a real thing and some people are good at it and other people aren’t but can improve with practice. If all we’re showing is that people who are good with probabilities are good with probabilities, then whatever.
But there are tantalizing signs that there might be something more here. I was involved in an unpublished study which I can’t upload because I don’t have the other authors’ permission, but which showed conclusively that people with poor calibration are more likely to believe in the paranormal (p < 0.001), even when belief in the paranormal was not assessed as a calibration question. So I went through the Less Wrong Survey data, made up a very ad hoc measure of total calibration skill, and checked to see what it did and didn't predict. Calibration was correlated with IQ (0.14, p = 0.01). But it was also correlated with higher belief in global warming (0.13, p = 0.01), with higher belief in near-term global catastrophic risk (-0.08, p - 0.01), increased support for immigration (0.06, p = 0.048) and with decreased support for the human biodiversity movement (0.1, p = 0.002). These were all independent of the IQ correlation. Notably, although warming and GCR were asked in the form of probabilities, immigration and HBD weren't, suggesting that calibration can be (weakly) correlated with opinions on a non-calibration task. Maybe the most intriguing evidence for a full-fledged General Factor of Correctness comes from Philip Tetlock and IARPA's Good Judgment Project, which got a few thousand average people and asked them to predict the probability of important international events like “North Korea launches a new kind of missile.” They found that the same small group of people consistently outperformed everyone else in a way incompatible with chance. These people were not necessarily very well-educated and didn’t have much domain-specific knowledge in international relations – the one profiled on NPR was a pharmacist who said she “didn’t know a lot about international affairs [and] hadn’t taken much math in school” – but they were reportedly able to outperform professional CIA analysts armed with extra classified information by as much as 30%.
These people aren’t succeeding because they parrot the experts, they’re not succeeding because they have more IQ or education, and they’re not succeeding in some kind of trivial way like rejecting things that will never happen. Although the article doesn’t specify, I think they’re doing something more than just being well-calibrated. They seem to be succeeding through some mysterious quality totally separate from all of these things.
But only on questions about international affairs. What I’d love to see next is what happens when you ask these same people to predict sports games, industry trends, the mean global temperature in 2030, or what the next space probe will find. If they can beat the experts in those fields, then I start really wondering what their position on the tax rate is and who they’re going to vote for for President.
Why am I going so into depth about an LW post from five years ago? I think in a sense this is the center of the entire rationalist project. If ability to evaluate evidence and come to accurate conclusions across a broad range of fields relies on some skill other than brute-forcing it with domain knowledge and IQ, some skill that looks like “rationality” broadly defined, then cultivating that skill starts to look like a pretty good idea.
Enrico Fermi said he was fascinated by the question of extraterrestrial life because whether it existed or it didn’t, either way was astounding. Maybe a paradox, but the same paradox seems true of the General Factor of Correctness.
Outside the Laboratory is a post about why the negative proposition – no such General Factor – should be astounding:
“Outside the laboratory, scientists are no wiser than anyone else.” Sometimes this proverb is spoken by scientists, humbly, sadly, to remind themselves of their own fallibility. Sometimes this proverb is said for rather less praiseworthy reasons, to devalue unwanted expert advice. Is the proverb true? Probably not in an absolute sense. It seems much too pessimistic to say that scientists are literally no wiser than average, that there is literally zero correlation.
But the proverb does appear true to some degree, and I propose that we should be very disturbed by this fact. We should not sigh, and shake our heads sadly. Rather we should sit bolt upright in alarm. Why? Well, suppose that an apprentice shepherd is laboriously trained to count sheep, as they pass in and out of a fold. Thus the shepherd knows when all the sheep have left, and when all the sheep have returned. Then you give the shepherd a few apples, and say: “How many apples?” But the shepherd stares at you blankly, because they weren’t trained to count apples – just sheep. You would probably suspect that the shepherd didn’t understand counting very well.
If, outside of their specialist field, some particular scientist is just as susceptible as anyone else to wacky ideas, then they probably never did understand why the scientific rules work. Maybe they can parrot back a bit of Popperian falsificationism; but they don’t understand on a deep level, the algebraic level of probability theory, the causal level of cognition-as-machinery. They’ve been trained to behave a certain way in the laboratory, but they don’t like to be constrained by evidence; when they go home, they take off the lab coat and relax with some comfortable nonsense. And yes, that does make me wonder if I can trust that scientist’s opinions even in their own field – especially when it comes to any controversial issue, any open question, anything that isn’t already nailed down by massive evidence and social convention.
Maybe we can beat the proverb – be rational in our personal lives, not just our professional lives.
And Correct Contrarian Cluster is about why the positive proposition should be equally astounding. If it’s true, you can gain a small but nonzero amount of information about the best economic theories by seeing what their originators predicted about migration patterns in pre-Columbian America. And you can try grinding your Correctness stat to improve your ability to make decisions in every domain of knowledge simultaneously.
I find research into intelligence more interesting than research into other things because improvements in intelligence can be leveraged to produce improvements in everything else. Research into correctness is one of the rare other fields that shares this quality, and I’m glad there are people like Tetlock working on it.
Discussion questions (adapted from Tumblr):
1. Five Thirty Eight is down the night before an election, so you search for some other good sites that interpret the polls. You find two. Both seem to be by amateurs, but both are well-designed and professional-looking and talk intelligently about things like sampling bias and such. The first site says the Blue Party will win by 5%; the second site says the Green Party will win by 5%. You look up the authors of the two sites, and find that the guy who wrote the first is a Young Earth Creationist. Do you have any opinion on who is going to win the election?
2. On the bus one day, you sit next to a strange man who mumbles about how Bigfoot caused 9-11 and the Ark of the Covenant is buried underneath EPCOT Center. You dismiss him and never see him again. A year later, you see on TV that new evidence confirms Bigfoot caused 9-11. Should you head to Florida and start digging?
3. Schmoeism and Anti-Schmoeism are two complicated and mutually exclusive economic theories that you don’t understand at all, but you know the economics profession is split about 50-50 between them. In 2005, a survey finds that 66% of Schmoeist economists and 33% of anti-Schmoeist economists believe in pre-Clovis settlement of the New World (p = 0.01). In 2015, new archaeological finds convincingly establish that such settlement existed. How strongly (if at all) do you now favor one theory over the other?
4. As with 3, but instead of merely being the pre-Clovis settlement of America, the survey asked about ten controversial questions in archaeology, anthropology, and historical scholarship, and the Schmoeists did significantly better than the anti-Schmoeists on 9 of them.