4 ms·

You never really know until you know you don't know? What we know now is that we think we know -- that is, until we don't. Theories are meant to be superseded

by cnp 12y ago

You never really know until you know you don't know? What we know now is that we think we know -- that is, until we don't. Theories are meant to be superseded.

roywiggins 12y ago
Extraordinary claims, extraordinary evidence, etc. The prior probability of ancient nuclear weapons is so small that there would need to be extremely good evidence to begin to shift my appraisal of how likely it is. (am I Bayesianing right?)
- tjradcliffe 12y ago
  Close on the Bayesianing: there's no threshold, so it isn't quite correct to say you'd need extremely good evidence to "begin to shift" your appraisal. But you'd need extremely good evidence to shift your threshold very much. Bayes' rule tells us the plausibility [$] of a proposition after taking into account some evidence should be the prior plausibility times the probability of the evidence occurring if the proposition is true divided by the probability of the evidence regardless of whether or not the evidence is true, so vastly improbable occurrences that would be way more probable if the proposition was true count as very good evidence, but lots of things can shift our degree of belief a tiny little bit. However, you are correct that the prior has such a minuscule prior plausibility that it'd take astonishing evidence that couldn't be explained otherwise to raise the posterior plausibility up to "worth taking very seriously" level. Beyond that, there is the practical consideration that we impose a threshold on our beliefs for purely pragmatic reasons. Below some plausibility it just isn't worth our time to even consider the issue. So in that sense, you are Bayesianing with the best of them. [$] I am a heterodox subjective Bayesian, in that I maintain a strict separation between "plausibility" and "probability". Plausibilities are degrees of belief by a knowing subject in a proposition, and as such are inherently subjective. Probabilities are measures of the likelihood of events, and by far the most common measure is the frequency. I take orthodox Bayesiansism to be in error by identifying plausibilities with probabilities, which is what creates the great subjective/objective quango. Bayes' rule on my view is Pl(H|E) = Pl(H)*Pr(E|H)/Pr(E) where Pr = probability and Pl = plausibility.

by cnp 12y ago

You never really know until you know you don't know? What we know now is that we think we know -- that is, until we don't. Theories are meant to be superseded.

roywiggins 12y ago

Extraordinary claims, extraordinary evidence, etc. The prior probability of ancient nuclear weapons is so small that there would need to be extremely good evidence to begin to shift my appraisal of how likely it is. (am I Bayesianing right?)

tjradcliffe 12y ago
Close on the Bayesianing: there's no threshold, so it isn't quite correct to say you'd need extremely good evidence to "begin to shift" your appraisal. But you'd need extremely good evidence to shift your threshold very much. Bayes' rule tells us the plausibility [$] of a proposition after taking into account some evidence should be the prior plausibility times the probability of the evidence occurring if the proposition is true divided by the probability of the evidence regardless of whether or not the evidence is true, so vastly improbable occurrences that would be way more probable if the proposition was true count as very good evidence, but lots of things can shift our degree of belief a tiny little bit. However, you are correct that the prior has such a minuscule prior plausibility that it'd take astonishing evidence that couldn't be explained otherwise to raise the posterior plausibility up to "worth taking very seriously" level. Beyond that, there is the practical consideration that we impose a threshold on our beliefs for purely pragmatic reasons. Below some plausibility it just isn't worth our time to even consider the issue. So in that sense, you are Bayesianing with the best of them. [$] I am a heterodox subjective Bayesian, in that I maintain a strict separation between "plausibility" and "probability". Plausibilities are degrees of belief by a knowing subject in a proposition, and as such are inherently subjective. Probabilities are measures of the likelihood of events, and by far the most common measure is the frequency. I take orthodox Bayesiansism to be in error by identifying plausibilities with probabilities, which is what creates the great subjective/objective quango. Bayes' rule on my view is Pl(H|E) = Pl(H)*Pr(E|H)/Pr(E) where Pr = probability and Pl = plausibility.