Log in

No account? Create an account

Forms of Doubt

Over the last many years, I've operated from a framework of probability requiring me to never believe anything with certainty; I still do this, seeing it as part of radical empiricism; I see reality as the only kind of truth, and any other things we generally call truth to only be so by analogy with a weaker claim to the term. I consider everything I call true to have a slight hedge to it; no proofs, no absolute solidity. If it is meaningful to judge it as truth, it must be falsifiable, and there is always some way, however unlikely it seems, by which I might be forced to give it up; the chances of this are practically (but not theoretically) identical to how likely I think it to be true.

I've been working to construct a doubting part of myself with a different perspective to challenge me with on this point; essentially a stronger form of the position that would fudge nearly-certain things to certainty. Perhaps without the ability or belief in reasoning with infinitely-grained probability statistic. If someone were to say "I am certain in $belief", and if pressed on it they said "I would revise if presented with contrary evidence, but it seems really really unlikely to me", I'm not sure how I would criticise that perspective. It's not Bayesian, but then I don't accept much of the philosophical trappings that normally go with Bayesian thinking (I consider frequentist reasoning to be generally stronger-when-available, although this is a complex topic that I think I covered in a previous entry). My gut instinct is to say "that's not how probability works", but pitting my definition against the one I'm talking about here doesn't feel like a clear win either.

Maybe I should say that it's a viable perspective, but weaker in predictive power (in the rare areas where it differs, to the extent of the size of the fudging it does) than my framework? I'm almost nervous about arguing against it because it easily falls into the messy realm of the role of judgement in statistical reasoning, and while we can capture *some* of that kind of thing through nicer mechanical reasoning methods (SVDs and similar), so much of the areas where we want predictive power (even if weak) is hard to quantify; the conversation (inner though it is) might easily fall into that area.