3 ms·

This bothers me too. Theories with too many adjustable parameters aren't really testable. And so, following Popper, I don't see how they can actually be scienti

This bothers me too. Theories with too many adjustable parameters aren't really testable. And so, following Popper, I don't see how they can actually be scientific.

nonbel 8y ago
Check out Imre Lakatos. "Falsifiability" is too strict since "every model is wrong". Lakatos calls that Popper0, which is not what Popper ever advocated. In a nutshell: The real thing going on is you are supposed to use Bayes' rule, where you normalize how well your explanation explains the observations ( p(H[0])p(H[0]|D) ) to the sum of all known explanations ( p(H[0:n])p(H[0:n]|D) ). Thus if you have a vague explanation that can fit anything, it won't be much better than a million other possibilities so who cares. We only care when someone comes up with an accurate prediction that would be otherwise surprising. There are many more interesting ideas too (eg, "degenerating research programmes"). Edit: I should also emphasize that my main point above is that there is a misuse of statistics going on though.
- mirimir 8y ago
  Yeah, I know Lakatos. > We only care when someone comes up with an accurate prediction that would be otherwise surprising. Sure. But the problem is when predictions don't pan out, parameters just get tweaked. Or new perturbations are introduced. Theories just keep getting more and more complicated.
  - nonbel 8y ago
    Yes, that is called a "degenerating research programme". The theory lags the observations.
- nonbel 8y ago
  Typo: ( p(H[0])p(D|H[0]) ) to the sum of all known explanations ( p(H[0:n])p(D|H[0:n]) ).
XorNot 8y ago
No they're just potentially not particularly useful models. To be a theory they have to be making some direct claim we can use to predict future results. To be rejected we have to find evidence where one adjustment would invalidate previous evidence (bring the hypothesis into contradiction with itself). Being a good scientist means being positivist about your models - if the model works, it works, and it's irrelevant in some respects what it says something actually is if the name is not part of an additional claim of how things should behave.

This bothers me too. Theories with too many adjustable parameters aren't really testable. And so, following Popper, I don't see how they can actually be scientific.

nonbel 8y ago

Check out Imre Lakatos. "Falsifiability" is too strict since "every model is wrong". Lakatos calls that Popper0, which is not what Popper ever advocated. In a nutshell: The real thing going on is you are supposed to use Bayes' rule, where you normalize how well your explanation explains the observations ( p(H[0])p(H[0]|D) ) to the sum of all known explanations ( p(H[0:n])p(H[0:n]|D) ). Thus if you have a vague explanation that can fit anything, it won't be much better than a million other possibilities so who cares. We only care when someone comes up with an accurate prediction that would be otherwise surprising. There are many more interesting ideas too (eg, "degenerating research programmes"). Edit: I should also emphasize that my main point above is that there is a misuse of statistics going on though.

mirimir 8y ago
Yeah, I know Lakatos. > We only care when someone comes up with an accurate prediction that would be otherwise surprising. Sure. But the problem is when predictions don't pan out, parameters just get tweaked. Or new perturbations are introduced. Theories just keep getting more and more complicated.
- nonbel 8y ago
  Yes, that is called a "degenerating research programme". The theory lags the observations.
nonbel 8y ago
Typo: ( p(H[0])p(D|H[0]) ) to the sum of all known explanations ( p(H[0:n])p(D|H[0:n]) ).

XorNot 8y ago

No they're just potentially not particularly useful models. To be a theory they have to be making some direct claim we can use to predict future results. To be rejected we have to find evidence where one adjustment would invalidate previous evidence (bring the hypothesis into contradiction with itself). Being a good scientist means being positivist about your models - if the model works, it works, and it's irrelevant in some respects what it says something actually is if the name is not part of an additional claim of how things should behave.