[YeGoblynQueenne]

Thanks, I didn't kow about Einstein and quantum physics. I'll have to read a bit about that, it sounds interesing.

My original comment is grounded in an assumption that predictive power is not enough to identify a theory as correct, and neither is simplicity. There's nothing to stop any number of theories to have the same predictive power and the same kind of complexity. Sometimes, it's just very difficult to choose one, above the others.

Did I come across as confusing predictive power with complexity?

EDIT: it's interesting you bring Occam's razor up. It's part of what I'm studying, in the context of identifying relevant information in (machine) learning. There are mathematical results (in the framework of PAC-learning) that say that, basically, the more complex your training data, the more likely you are to overfit to irrelevant details. At that point, you have a model that explains observations perfectly well, but is useless to explain unseen observations (the really unseen ones- not those pretending to be unseen for the puprose of cross-validation).

...iiish. The result is that large hypothesis spaces tend to produce higher error. But, the size of the hypothesis space in statistical machine learning depends on the complexity of the data, as in the number of features. Anyway, I'm fudging it some. I'm still reading up on that stuff.