John Wilkins - Frequentism vs Bayesianism

frequentists think that probability has to do with observed frequencies of things in a sample right so I say what's the probability that the the next ball that I will draw from the lottery you know that thing that bounces all the balls around and in the lottery draw is going to be even right now you might say our priori it's one in two but then you don't know whether the balls are in fact a fair sample of 1 to 100 or whatever you don't know that so a frequent assayer's count them and then you'll know a Bayesian works on the idea that you have some prior commitments about the way the world is in this case for example you might say well it's very unlikely that the government would permit a biased sample of balls in the lottery so therefore it's got to be what all the numbers from 1 to 100 and therefore it's going to be a 50-cent percent chance and that's using your prior assumptions to estimate the probability of this particular case that's what Bayesian ISM is about as I understand it now the thing is that that works well in the case of the lottery it's a lot harder when you're dealing with the probability that there will be a certain disease in a population for example and your prior assumptions set the likelihood that you're going to find something and then you find something and you say right this is unlikely therefore it's a very good result that we found it or it's very likely that it would happen by chance so therefore it doesn't mean very much and it gets very very messy but most people working in that field as I understand it think that if we aren't actually Bayesian zin practice we're pretty close to it and Bayesian probability is something that is it good enough way to model the way that scientists infer things in complex cases I don't really have a view on the matter I don't really think it's resolved I think if we are Beijing's we're unconscious patients and I'm not sure that Bayes Bayesian probability is all that accurate at describing what we do because we're not perfectly rational we don't make our inferences using only the information that we've got there's also all kinds of cognitive biases that humans are likely to encounter for example we find agency where there is none right we talked about deities and and intentions of other living organisms where there may be none and so on so human beings have got a number of biases particularly with probabilistic reasoning we're not good at reasoning in a probabilistic fashion which was why statistics was invented to try and overcome that and there's a big field on the philosophy of statistics and the philosophy of probability well it sort of depends on whether you think Bayesian probability is a description of what we do or a prescription for what we should do if you think it's a description of what we do then I don't think it's all that great if you think it's a prescription of what we should do then it may actually be very useful indeed because it then becomes a methodology for eliminating some of the biases and for pointing out where our priors might be wrong and that we can go and repair that as it were the other thing is that with new observations you tend to refine your priors over time so a Bayesian NOAA is able to improve and refine their their accuracy over time so if the scientific researcher uses a Bayesian methodology or abrasion approach to analyzing their data and the probabilities that they're encountering it may actually eliminate quite a lot of error that I think is an interesting and open question at the moment you

Channel: Science, Technology & the Future

Views: 21,588

Keywords: science, technology, singularity, transhumanism, Adam Ford, Future, humanity+, future by design, Science Technology & the Future, Social Issues, scifu, scifuture, Frequentist Probability, Bayesian Probability, Bayes, Frequentist

Id: tsuJM_bHSgA

Length: 5min 23sec (323 seconds)

Published: Thu Aug 14 2014