Re:Ask HN: Anyone interested in building tools for showing b...
by Chris2048 on Thursday November 17, 2016 @11:11AM
(#12978665)
I'm not sure this is True, as you can use probabilistic arguments.
Re:Ask HN: Anyone interested in building tools for showing b...
by pizza on Thursday November 17, 2016 @07:00PM
(#12983054)
Probabilistic models are what I was referring to; all probabilistic models should produce a prediction, and also a measure of confidence. That measure of confidence can never be 100%. Similarly, you can never have an incontrovertible belief that is not a mathematically/logically-constructed fact 'outside' of the messy real world in some abstract chamber walled off from anything that is like a real life observation, no matter how certain. Basically, you can't have P(A) = 0 or 1 as your prior probability, because then you break Bayes' rule in the sense that new evidence is meaningless - https://en.wikipedia.org/wiki/Cromwell%27s_rule
Mathematics relies upon proofs, but there is no such thing as a scientific proof.. You can find sufficient conditions for a phenomenon and have a wholly water-tight argument as to why it occurs, but you can't use use observation to find necessary condition.
This is because observations are evidence, and if a future observation shows problems with your theory, you should not discard your evidence (unless of course your secondary theory is that your evidence exhibits a model for some reason why it doesn't explain what it seems to explain on the surface, e.g. experimental error), yada yada..