My Thoughts on Science (& a Bit on Religion, too)

[ericjherboso]

I always have such a tough time explaining how I think to other people. No one ever seems to be able to follow my train of thought no matter how hard I try to explain myself, and I consistently end up thinking that maybe there's some part of me that is crazy in a way that is obvious to others but is completely invisible to me

Comments

underpants

[ericjherboso]

The difference in applied math, I think, is that you still think of making 'refinements' of models, where I instead have this idea that whatever the current model is, it is far more likely that the reality is completely and utterly different from whatever you're thinking of now.

The concept is that there are an infinte # of ways to model a set of data points, and that each workable model of these infinite choices goes in all sorts of directions. So if you see something that looks like a linear relationship, you might think: "with further data, we might find that it's really a very flat asymptotic relationship instead". Whereas if I see something that looks like a linear relationship, I might think: "with further data, we might find that the relationship is completely random, or is determined by the number of underpants in politician's washing machines at midnight next friday, or any other completely different relationship".

I know that sounds completely stupid on a first read, but consider that with an infinite number of models, and only occam's razor to choose between them (see below for my comment on occam's rather blunt razor) there is NO reason to prefer a linear model to an underpants model, except in that it helps people to make predictions that historically have proved rather true to form. But my entire point is that there are an INFINITE number of models which would have been BETTER predictors than what we happened to use. In fact, there are an infinite number of underpants models which would have predicted better than the mathematical models we currently use. If good predictive skills were the decisive factor, then I'd STILL consider underpants models to be a genuine alternative to mathematical models.

I fully realize this statement makes me look like an idiot, but I swear that I'm being completely serious about all this.