Originally Posted by Dave Rogers
I never said ONE "corrrect" hypothesis. The assumption is fully warranted and commonly observed. For any set of data we can create an infinite number of differing hypotheses that match these (and all other) observations. (Some of) these "currently accurate" hypotheses will make differing predictions under other conditions. So clearly some of the classes of 'currently accurate' hypotheses will later be verified and others falsified if/when we test.
Yes - obviously some hypotheses that match the current observations contradict each other in other domains and ranges of conditions. So selecting any one hypothesis without additional work will be certain to lead to errors. Therefore Occam's Razor often sleects a hypothesis that is later falsified. Occam's razor is intended to be a 'neutral' selection rule, however the definition of "fewer presuppositions' or "simpler" doesn't bear close scrutiny. We should at least have a schema that doesn't reject potentially accurate hypotheses, as Occam's Razor clearly does.
Wrong question. It's which of millions of hypotheses that "accurately predict all known data" shall we accept ? Occam'sR requires the one with the fewest presuppositions, but that means we will be choosing hypotheses that will later fail, forcing revision. Perhaps we should consider the set of ALL supported hypotheses.
Yes - and this second 'better' hypothesis invalidates the first forcing revisions. The first hypothesis was wrong all along, and it was undoubtedly be used to make erroneous or at last imperfect extrapolations and predictions. It could be a costly blunder to accept the 'simplest' hypothesis rather than considering all supported hypotheses.
[QUOTE}So what, precisely, are the properties required of your "correct" hypothesis other than correctly matching all the data?
You miss the point. I am NOT creating any new requirement, and I am not suggesting that I can predict the single "correct" hypothesis. Never did, so don't strawman. I am pointing out that Occam's razor selects a single hypothesis as a tentative basis and in many cases the hypothesis is later falsified. It's entirely possible that there is a better basis for scientific development than this.
For example In Newtownian mechanics kinetic energy KE = (1/2)*m*v^2
In relativistic mechanics KE = (1/2)*m*v^2 * [ 1 + [(3/4)(v/c)^2 + (5/8)*(v/c)^4) + ....]]
which are terms in the maclaurin expansion of the Lorentz factor.
Obviously work in the early 20th century falsified Newton's eqn and supported Lorentz.
So all data available to Newton would have equally supported both Newtown hypothesis and the Lorentz form and a million others. The only distinguishing feature is that Newton had no argument in favor of the "extra terms". Or obversely - Newton had no basis for rejecting the extra Lorentz terms. Why one is considered a greater presupposition than the other isn't well founded. What is clear is the Occam'sR rejected the (later) supported hypothesis in favor of the (later) falsified hypothesis. That's not a good feature.
Perhaps we should consider a new scientific method that accepts (tentatively) ALL supported hypotheses, and only eliminates the falsified ones. This would leave the open questions open, rather than fill the gaps with simplistic presuppositions.