Connection Theory FAQ
An FAQ-style critique of Connection Theory, a putative comprehensive theory of human psychology developed by Geoff Anders at Leverage Research, a Bay Area organization that drew talent and funding from the LessWrong rationalist community in the early 2010s. Leverage claimed Connection Theory could predict and explain human belief and action; this document collects the case that it cannot.
1. Introduction 1.1 What is Connection Theory?
Connection Theory (CT) is a theory of psychology, developed by an organization called Leverage Research , founded and led by Geoff Anders. It claims to be a “simple, comprehensive theory of human belief and action”.
1.2 Where can I read about CT?
Leverage Research has published some material on CT on their website .
1.3 Is CT correct?
No. It is not correct in any way, shape, or form. This FAQ explains why.
1.4 Is CT worth spending time on?
No. The purpose of this FAQ is to collect all the arguments about CT in one place, so we can dismiss it and get on with our lives. Creationism is wrong, but that doesn’t mean we should spend all day refuting it. Science progresses by accepting established conclusions and building on them, not rehashing whether the Earth revolves around the Sun.
1.4.1 So why are you spending time on it? Hasn’t this all been said before?
Not really. Parts have been published, and other parts have been used for in-person debates. But in-person debates take hours or days, and only reach handfuls of people. Hence, an FAQ is better suited to our main goal of not wasting time.
1.5 If a point is refuted, will you update the FAQ?
Yes. Any such updates are noted in comments below.
2. Meta-Criticism 2.1 So what if point X is wrong? You can’t expect a new theory to get everything right.
This is a form of logical rudeness — making bold general claims, and immediately jumping back when discussing specifics. This logical rudeness seems endemic to any imprecise field — one whose subject matter is inherently complex. Any theory of psychology (or economics, or sociology…) makes many predictions. If you make a thousand unrelated predictions, at least one will be wrong, even if the theory is very good. Hence, one mistake doesn’t necessarily refute a theory, especially one with many previous successes.
But trouble comes when one is tempted to rudely deflect every criticism on this basis. One miss and a hundred hits is fine, but you must add up the misses — every miss counts against you more. A bold claim must be supported with bold evidence, and “comprehensive theory of human action” is really bold.
2.2 But didn’t Leverage agree CT was false?
Yes.
This is trying to avoid falsification by showing fake humility. By claiming that CT is false, without specifying how , one can deflect any counterargument by nominally agreeing, and then continuing to act as if CT was true — not updating one’s actions based on the evidence.
2.2.1 Does Leverage act like CT is true?
Yes. Leverage has produced a great deal of material on CT , much of which is “evidence” for CT, arguments for CT, or responses to criticism, just like any advocate for a theory would. This isn’t reasonable behavior if one believes his theory is false. If, say, a physicist had an interesting theory that violated Conservation of Energy, he’d immediately (a) make a narrower claim that didn’t violate CoE, like saying “these points fit this equation” without specifying a mechanism, or (b) resolve the contradiction. He’d at least put up big red labels explaining the violation. If he made a whole website arguing for the theory, while putting in a little footnote saying “ah, this also violates CoE”, he’d rightly be labeled a crackpot.
2.3 Even if CT has serious problems, what’s wrong with discussing it?
The threshold for mentioning an idea is low. I’ve brought up many ideas that were almost certainly crazy. But Leverage claims CT is worth, not just discussion, but full-time effort on the part of many highly capable people, a seven-figure annual budget, and the time and attention of some of the brightest minds in technology. To be worth that level of effort, an idea must have serious justification. We don’t give Harvard professorships to crackpots “disproving” the Pythagorean Theorem, even if the cost (~$200K) is tiny compared to Harvard’s budget. There are better things capital can be spent on.
2.4 CT isn’t complete yet — X will be in a later version.
A theory full of holes doesn’t justify grand claims about having solved psychology. At best, it means that one will be able to solve psychology at some point in the indefinite future, and such claims are usually ignored for a reason. If a hole hasn’t been filled in, then by definition, we don’t know if it can be. The odds of being able to fill in every hole go down exponentially with the number of holes.
2.5 But if CT works, even at tiny odds, won’t the payoff be huge?
This is an argument called Pascal’s Mugging. Essentially, using naive expected utility, one’s utility function is always dominated by tiny odds of vast payoffs. However, this reasoning is a known fallacy and should be avoided.
3. Methodology 3.1 Even if CT is wrong, isn’t it worth taking time to understand and research it?
No, because of prior probabilities. No one spends all day refuting physics theories from chatbots, because a chatbot isn’t the sort of thing you’d expect to produce valid physics. The prior probability of Chatbot_Physics, before you even look at details, is extremely low. Similarly, the process that produced CT isn’t a process you’d expect to make sensible psychology. It’s never going to work, any more than throwing random bits of metal will make a working vacuum cleaner. (The rest of the FAQ is for people who don’t buy this section.)
3.2 How are valid theories created?
First, a theory starts with a problem it’s trying to solve. This gives you a metric to measure progress. In physics, one might observe that some things stay still, and others move very fast. In economics, one sees some countries growing and others falling into poverty. In chemistry, some things don’t react and other things explode. Why? What causes it? Every explanation starts with the data it’s trying to explain.
Very often, there’s an existing theory explaining some of the data. The challenge is to invent a better theory, explaining it more precisely. Here, too, one starts off with data not fitting the existing hypothesis. Newton’s laws are great, sure, but what about friction? What about light? What about the precession of Mercury? First the question, then the answer.
3.3 What data does CT try to explain?
None. There are no psychological experiments, no surprising new data, that CT was created to explain. There are also no mismatches, no holes in existing theory, that CT purports to fill. Rather, CT was created first, and “explanations” of existing data were invented afterwards. If one starts with the answer, instead of the question, you can’t even know if you’re moving forward rather than backward. There’s no specific mystery you’re trying to solve, no criterion of success.
3.4 What previous work does CT build on?
It doesn’t. This is not an exaggeration — none of the (quite lengthy) Leverage documents describing CT even have bibliographies. There are also no predecessors by others in the field, no beta versions, no proposed alternatives, no updates in response to new experiments… CT appears to spring fully formed from the head of Zeus.
Anyone even slightly familiar with the history of science knows this isn’t how discoveries work. There are certainly revolutions in science, like Newtonian mechanics or special relativity. But these build on previous results in the field. In, e.g., Newton’s case, he already had precise, verified, mathematical descriptions of the behavior of falling bodies (from Galileo) and the planets (from Kepler). There’s a reason Newton was English, and not Greek or Roman or Egyptian.
3.5 How was CT discovered?
We don’t know. At least, it doesn’t seem to be published anywhere. This is a huge problem — if you pick a theory out of a hat, the odds of it being correct are astronomical, because the space of possible theories is huge. To even be worth considering, there should be a strong explanation for why we’re thinking about a theory in the first place. What makes it a likely suspect, if you will, compared to some other one with a few bits shifted around. This explanation doesn’t seem to exist.
3.6 What is Leverage’s methodology?
“If there’s anything methodological we tend to agree on, it’s a process. Writing drafts, getting feedback, paying close attention to detail, being systematic, putting in many, many hours of effort.” — Geoff Anders As far as I know, Leverage has no explanation for why their group and their methods should do better than the millions of experts already studying psychology, but they are still very confident they will. It appears to resemble the 90% of incoming students who think they’ll be in the top 10% of the class. Insofar as Leverage describes their methodology (above), it’s sufficiently vague to also describe any investment bank, organizations not noted for their academic output.
4. Biology 4.1 How does CT account for known features of neuroanatomy?
It doesn’t. The biology of the brain, how neurons work, which lobes perform which functions, and so on are never mentioned or discussed. The words “brain”, “neuron”, and “lobe” don’t appear in the eighty-page document Connection Theory: Theory and Practice .
To be fair, many successful psychology theories don’t reference specific neurology either. However, this is much more troublesome for a theory purporting to explain everything about how the brain works, rather than just how one specific class of people behave in one situation.
4.2 How does CT apply to human predecessors, like chimps and bonobos?
It doesn’t. As far as I know, no effort has ever been made to do so. Again, most psychology theories don’t do so either. However, for a theory which claims a complete explanation of the brain, applicability to chimps is an obvious test. Chimps and humans share 95% DNA, all their basic neuroanatomy, and almost all their brain structures. If one looks at how chimp muscles work, or chimp joints work, or chimp livers work, it’ll be very similar to humans. The same should hold for the brain, which is also a human organ.
4.3 How did CT beat competitors in the adaptive environment?
This is never explained, or even hinted at. Evolutionary psychology can be somewhat voodoo-ish, but most established psychology can be adapted for an evolutionary framework. We retaliate against enemies because Tit-for-Tat wins the Iterated Prisoner’s Dilemma. We like sex and food because it helped our ancestors survive. Biases like scope insensitivity were much less of a problem in an adaptive context, and so on. A complete explanation of the brain should take evolution into account. For example, CT postulates humans have a single set of fixed, long-term goals. Why would this serve ancestral humans? The goal of evolution is very general (have more surviving offspring), and one would intuitively be better off changing one’s goals to match the situation, e.g. seeking another mate if yours is killed or betrays you.
In particular, a theory should be game-theoretically stable (a Nash equilibrium) under ancestral conditions. A big reason socialism doesn’t work is that New Socialist Man isn’t adaptive; humans who sacrifice all for the good of the tribe get outcompeted.
5. Evidence 5.1 Didn’t CT make successful predictions?
Not in any meaningful sense. The Leverage document Evidence for CT describes “tests” that were conducted. However, these “tests” examined the subjective assessments by Leverage employees, of the behavior of Leverage employees. Such “tests” could be made as successful or unsuccessful as desired, simply by testers changing their own behavior to whatever behavior they had earlier predicted for themselves . Yet, Leverage claims these “tests” provide “very strong evidence” for the correctness of CT.
5.2 How are these predictions compared to other theories?
They aren’t. Leverage simply asserts, without evidence or explanation, that “No other existing psychological theory has anything close to this degree of predictive power.” No competing theories are described, or even mentioned. The accuracy of “predictions” aren’t compared to those made by competitors to CT, to those made by naive bystanders using folk psychology, or even to random chance.
5.3 Isn’t psychotherapy using CT effective?
Perhaps. However, we also know almost any form of psychotherapy is effective, including the “naive psychotherapy” of discussing the problem with a friend for a few hours. Hence, to be meaningful, therapy using CT must be compared to at least one other form of therapy.