Thursday, December 11, 2008

Why Faith is Poor Epistemology

On John Wilkins blog, a commenter posted the following quote:

"If falsehood, like truth, had but one face, we would be more on equal terms. For we would consider the contrary of what the liar said to be certain. But the opposite of truth has a hundred thousand faces and an infinite field. --Michel Eyquem de Montaigne"

This is a notion I have seen stated in many different ways. This is what Tolstoy was getting at as well, albeit with a more limited scope, when he uttered his famous opening to Anna Karenina: "All happy families resemble one another, each unhappy family is unhappy in its own way".

I argue that this leads inexorably to the conclusion that faith is poor epistemology. Faith, belief without evidence, amounts to choosing, at random, from all possible beliefs. Yet those beliefs are overwhelmingly false. Some would say the false are infinitely larger than the true, in which case the probability of a faith belief being true approaches zero.

For example, take the true statement "My car is blue". Now randomly substitute all the other grammatically appropriate words that can be put in that sentence. "My car is red", "My car is green", "My cat is blue", "My teacher is blue", etc. The ratio (true statements) / (false statements) is clearly approaching zero. I realize this is not stated in a mathematically rigorous way, but I have no doubt it could be, although I'm ready to give up the whole idea if a mathematician can show me why I should. Likewise, I understand there can be difficulties in the semantics. How wrong is wrong? When exactly does "blue" become "green". However, these difficulties seem biased to my favor, since semantic difficulties erode the confidence in any statement, which is the thrust of my general argument. If I say you shouldn't be so confident because a disease like measles exists, it hardly helps your argument to challenge my definition of disease as being too specific. That's the sort of argument creationist trolls make.

So mathematically, logically, and pragmatically, faith is a loser.


Anonymous said...

I don't think faith is epistemology, or a form of it.

It is not epistemology when I trust my mother to bring me candy.

No, it is trust/love/hope. I guess there's implicit assumptions about the world, but in the mother case they are accurate.

So your objection can't be to faith, but to incorrect faith. But that's just an objection to the truth or falsity of the claim, just like anything else. No intrinsic problem with faith itself.

Plus, you have faith in reason. And surely reason is okay for epistemology.


Anonymous said...

Are you sure that an atheist has no faith at all? Perhaps instead he has a faith -- in nothing?

ScienceAvenger said...

Of course faith is epistemology. It is a way of gaining knowledge about the world. It's just a bad one, which is why I have a problem with it per se. Sure, it's right once in a while, just like a stopped clock. You are lucky that your faith in your mother is justified. Many children can't say that.

I don't have faith in reason. It's a good, but imperfect method of determining truth. We are prone to error, even when our motives are pure and our abilities impressive. That's why science is so successful. It assumes from the outset that we are flawed and must prove ourselves.

Atheists are not necessarily without faith. They can have faith in astrology, or their mom, or the Dallas Cowboys. But what atheists mostly do not have is a faith with regard to the existence of gods. It doesn't take faith to reject that without evidence.

Unknown said...

It's an interesting take on the problems of faith, and thanks for the Asimov link, which is a gem.

At the end of the day, however, isn't what we are left with the usual Bayesian interpretation: degree of belief induced by evidence? Russel's Teapot sounds very much like your and de Montaigne's argument: Instead of saying that there are an infinity of lies opposite truth, the Teapot represents the infinity of beliefs for which there is no evidence.

Your point that the ratio (true statements)/(false statements) can approach zero is analogous to the idea that the probability of the statement "The pencil is 12 cm long" tends to zero as your ability to measure the pencil's length improves. What matters is not your ability to predict the pencil's length arbitrarily exactly (you never will be able to outside idealised mathematical situations), but rather that your prediction be better than the previous one. This is analogous to Asimov's point about theories being less and less wrong.

Although the probability Pr("pencil length = 12cm") tends to zero, the ratio of Pr("12cm")/Pr("13cm") diverges as the measured pencil length gets closer to 12cm, and tends to zero if it is closer to 13cm. These quantities are well defined, and can be used to compare theories. E.g. Pr("flat earth")/Pr("spherical earth") tends to a very very small number, even though both theories are 'wrong'.

ScienceAvenger said...

Yes, relative error is an important concept, and sadly one few people understand. I'm amazed when I see statistics cited in raw figures, with no idea what percentage it represented.