A reservationist epistemology

If I have a meta-ideology, it’s that the world needs a lot more ideologies. Unfortunately, it doesn’t have them. So I thought I’d fill in by making up some of my own.

If you’re a formalist, you believe that any stable and predictable allocation of contested resources is an antidote to friction. If you’re a neocameralist, you want to formalize the modern corporatist state, converting it into a joint-stock company whose beneficiaries are precisely defined, and erasing the last vestiges of the rotary system.

But UR would be a boring blog indeed if it did nothing but harp on these old ideas. So perhaps it’s time for a new ideology, reservationism. A reservationist is anyone who reserves the right to think for himself. (Or herself.)

Of course, I’d like to think that anyone who thinks for herself (or himself) will arrive at the same conclusions as me, and thus be a formalist and a neocameralist. But since I have no way of guaranteeing this result, the monicker is not redundant.

The central dogma of reservationism is that reason is irreducible and untranscendable. Reason is no more and no less than common sense. It is not possible to construct a useful definition of common sense, nor is it possible to construct a system of thought that improves on common sense. Any system that purports to do so is either (a) bogus, or (b) justifiable via common sense, and thus a special case of it.

For example, mathematics is a special case of reason. Mathematics proves theorems by reducing complex formal propositions to a series of obvious steps. Since there is no mathematical definition of obviousness, there is no mathematical definition of proof.

I dislike the word “science” and wish people would stop using it, but inasmuch as the results of “science” are trustworthy, this trust can be explained by reason. For example, one reason why falsifiable “science” is trustworthy is that “science” is a social system which lionizes those who present falsifying evidence, and puts those whose theories are falsified in small metal cages, where passersby can jeer and poke them with sticks. When this social incentive structure breaks down, false theories which are logically falsifiable can be socially validated as “science.”

Furthermore, the formal methods of mathematics and “science” are effective only for a very small set of problems. These problems do not include most of the things that most people disagree about. A reservationist has no objection to formal deductive and inductive reasoning, but he or she stoutly resists the proposition—all too common in the 20th century—that they can or should replace all other forms of thought.

The great enemy of the reservationist is the automatist. An automatist is a small, grubby person who believes he can reduce or transcend reason. In the last two centuries, enormous armies of automatists have proposed all kinds of replacements for common sense. The fact that these replacements often travel under the name of “reason” itself is best explained adaptively.

Automatists tend to fall into four camps. The stupidest are literalists, who believe that instead of thinking, we should accept the literal text of some holy book or other. The most dangerous are officialists, who believe that truth is whatever the government says it is. The most annoying are popularists, who believe that the most fashionable thoughts, as of right now, are the most likely to be true. And the most pernicious are algorithmists, who believe they have some universal algorithm which is a drop-in replacement for any and all cogitation.

Automatists are automatists not because they are evil, but because they are too familiar with special cases of reason in which their flavor of automatism is indeed reasonable.

For example, if you’re a chemist, you might well come to believe that the CRC Handbook is the literal word of God. Certainly this belief is unlikely to serve you wrong in your chemical career. If Congress enacted the CRC Handbook as part of the US Code, the result would be a fascinating state of affairs in which Federal and natural law concurred, thus enabling a reasonable chemist to be both literalist and officialist. But this would not go one angstrom toward validating literalism or officialism as replacements for reason.

Automatists, in general, are fascinated by the elucidation of natural law. They have devised many effective techniques for reverse-engineering the structure of nature. I have no quarrel at all with these algorithmist methods. Where they work, they work because they are reasonable. There is no need to go as far as Feyerabend in rejecting formal methods.

Reservationists, in general, are fascinated by the interpretation of human affairs. In human history, politics, and economics, we observe patterns which appear to be patterns of cause and effect. If we can understand these patterns, we can predict the effects of our actions, and since most people are well-intentioned, we can wipe out war, poverty, bad television and tooth decay. If we misunderstand these patterns, our well-intentioned actions may indeed cause war, poverty, bad television and tooth decay.

One way to look at the reservationist problem is to imagine that a superintelligent alien, an agent of some galactic supercivilization of unimaginable wisdom and antiquity, is orbiting the earth in an invisible spaceship. The alien’s name is unpronounceable, so we’ll call her Beatrice. Beatrice has a mind-boggling array of cameras, invisible drones, and other monitoring devices by which she achieves effective omniscience. However, she is strictly prohibited from affecting the world in any way.

Beatrice’s job is to explain what’s going down on Planet Three. Every year, she sends a hypercable back to the Large Magellanic Cloud which summarizes political, economic and intellectual developments on Earth. So that human concepts only need to be translated once, Beatrice’s annual cable is written in English. Let’s say it’s no more than 20,000 words.

Wouldn’t you like to read Beatrice’s reports? I certainly would. I’d like to understand Beatrice’s view of Planet Three, at least to the limited extent that my pathetic monkey brain can even begin to follow her vast and oceanic wisdom. In fact, I’d like to think that if Beatrice’s rules of engagement were relaxed and she was permitted to start a blog, it would look very much like UR, although I’m sure she would be funnier, less neurotic, and more punctual in answering email and moderating comments.

What I don’t think is that any literalist, officialist, popularist, or algorithmist methods can even begin to help us in emulating Beatrice’s worldview. In fact, I think these automatist methods are tremendously distracting and destructive, which is why I spend so much time trying to find and resurrect forgotten writers whose thought strikes me as unbiblical, unofficial, unpopular and certainly unalgorithmic. (My new favorite: Albert J. Beveridge.)

One new, and very popular (among the smart set) algorithmist automatism is Bayesianism. Bayesians are followers of Bayes’ theorem, a result in probability theory. The Bayesians tend to congregate at the group blog LessWrong (formerly Overcoming Bias), where they get together and figure out how many blue balls are in the white urn.

Here is a good intuitive explanation of Bayes’ theorem by one prominent Bayesian. Please take my word for it: this level of hubris is not at all atypical. When they say things like “in cognitive science, Bayesian reasoner is the technically precise codeword that we use to mean rational mind,” they really do mean it. Move over, Aristotle!

Of course, in Catholicism, Catholic is the technically precise codeword that they use to mean rational mind. I am not a Catholic or even a Christian, but frankly, I think that if I had to vote for a dictator of the world and the only information I had was whether the candidate was an orthodox Bayesian or an orthodox Catholic, I’d go with the latter.

Let’s take a slightly closer look at Bayes’ theorem, and see why these people are on crack.

Bayes’ theorem is a pure product of mathematics. It is extremely true and extremely reasonable. If \( A \) and \( B \) are stochastic events, \( P(A \vert B) \) really does equal \( P(A) P(B \vert A) / P(B) \).

The only problem is that this little formula is not a complete, drop-in replacement for your brain. If a reservationist is skeptical of anything on God’s green earth, it’s people who want to replace his (or her) brain with a formula.

We can see this by looking for cases of cogitation for which Bayes’ theorem is about as relevant as tits on a boar hog. Believe it or not, there turn out to be one or two such cases.

First, what Bayes’ theorem gives us is a way of constructing one value from three others. We know: \( X = WY/Z \). Therefore, if we know \( W \), \( Y \), and \( Z \), we can know \( X \). Or if we know \( X \), \( W \) and \( Z \), we can know \( Y \). And so on. Algebra! Do it yourself at home!

Now, there are certainly plenty of cases in which it is actually useful to calculate \( P(A \vert B) \) from \( P(A) \), \( P(B) \) and \( P(B \vert A) \). Spam filtering is one. \( P(A_m) \) is the probability that message \( M \) is spam, \( P(B_s) \) is the probability that it contains some string \( S \), \( P(B_s \vert A_m) \) is the probability that if it’s spam it contains S, and \( P(A_m \vert B_s) \) is the probability that if it contains \( S \) it’s spam. If we keep a database of past messages, we can estimate \( P(B_s \vert A_m) \) and \( P(B_s) \) by assuming that spam messages are similar to other spam messages, and likewise for non-spam. Then we can construct \( P(A_m) \) iteratively by starting with the percentage of all messages that are spam, and reapplying this algorithm for various \( S \).

Note how interesting and special a case this is. It is precisely a case in which we have good estimates for \( Y \) and \( Z \), and a crappy estimate for \( W \) which can be improved by iterating with a large set of \( Y \)’s and \( Z \)’s. Thus it makes sense to use Bayes’ theorem.

But fundamentally, we are calculating one variable from three. The Rev. Bayes was a great man, no doubt, but his theorem does not contradict the Garbage Theorem. If there is garbage in \( W \), \( Y \) or \( Z \), there will be garbage in \( X \). If we can iterate the computation for a wide variety of reliable \( Y \) and \( Z \), we may be able to dilute the garbage in \( W \) to oblivion, but without reliable \( Y \) and \( Z \), what we have is a magic box that turns garbage into fresh, tasty food.

To make this more concrete, let’s look at how fragile Bayesian inference is in the presence of an attacker who’s filtering our event stream. By throwing off \( P(B) \), any undetected pattern of correlation can completely foul the whole system. If the attacker, whenever he pulls a red ball out of the urn, puts it back and keeps pulling until he gets a blue ball, the Bayesian “rational mind” will conclude that the urn is entirely full of blue balls. And Bayesian inference certainly does not offer any suggestion that you should look at who’s pulling balls out of the urn and see what he has up his sleeves.

Once again, the problem is not that Bayesianism is untrue. The problem is that the human brain has a very limited capacity for analytic reasoning to begin with. When you persuade it to obsess over one nifty, but basically unimportant, result in probability theory, you are doing your best to convert it into the Bayesian equivalent of this French civil servant.

A Bayesian will not exactly come out and tell you that you shouldn’t think deductively or intuitively. He will just drop little hints that all other ways of thinking are special cases of Bayes’ theorem, which in a sense is true, because they can be used to compute \( Y \) and \( Z \). As the Turks say, all you need is three horseshoes and a horse. And this is why rationalism is the enemy of reason: you start with a fully shod horse, and end up with a horseshoe. Hey, where’d my brain go?

A reservationist is perfectly comfortable in applying Bayesian inference to Bayesian problems. Similarly, the Bible is a pretty good source of Biblical history, Centcom’s statements about the war in Iraq tend to be quite reliable, I agree with most people that cold beer is refreshing and warm beer is nasty, and Pythagoras was right that the angles in a triangle sum to \( 180^\circ \). This does not make me a fundamentalist Christian, a Bush supporter, a demotist or a Pythagorean.

Meanwhile, if you are actually interested in “overcoming bias,” there’s no better way to start than reading the Lincoln biographies of Beveridge (1928) and Masters (1931). Clear out some of them stinky old personality cults. After Lincoln, try Flynn (1944) on FDR. If that don’t update your priors, I dunno what will.