Feb 25, 2025
Good models of reality should:
I think it’s sometimes useful to think of science and philosophy of trying to do the same thing, in that they’re looking for good models of reality. But they differ in which part they optimize for: you do philosophy when you’re mostly thinking about which explanation is simplest, and you do science when you’re mostly thinking about which explanation predicts reality best.
So for example, arguing about why there is something rather than nothing is philosophy, because all the explanations basically make the same predictions (“there is something”). Similar for, say, deciding on the best interpretation of quantum mechanics. An example of science would be something like trying to predict the weather. But I think it’s rare to do “pure” science in this sense – science usually involves some attention to model simplicity, because if you don’t care about model simplicity at all you can make a very complicated theory which matches past data well but fails to generalize. And I suppose “pure” philosophy is also rare, since the things we care to philosophize about tend to have some bearing to reality. A philosophy of “why there is nothing rather than something” when clearly there is something would be weird.
Some things can be seen as a mix of both. Consider doing a meta-analysis to determine if fluoride is harmful. On one hand you’re clearly in the business of predicting reality, because you want to know if using fluoride toothpaste will shorten your lifespan. On the other hand, I think that the key to effectively making sense of diverse research
Of course, these are definitions I propose because I found them nice, they are somewhat arbitrary. One could phrase things another way: science is about predicting reality as accurately as possible. This requires us to care about model simplicity, because simpler models generalize better because of Occam’s razor / Vapnik–Chervonenkis dimension / induction / whatever (this is its own philosophical problem I suppose). When one is holding predictions ~constant and trying to reduce model complexity, one is doing philosophy. In the case of the fluoride meta-analysis, reducing complexity while trying to still make sense of the existing research could help us find explanations which generalize and predict the future well. In the case of contemplating why there’s something rather than nothing, probably not.
Is it true that philosophy doesn’t care about predictions? I think it basically is. Insofar as it has predictions, the predictions are about what beliefs you should have. For example: maybe if you believe that God exists you should believe that there is no evil in the world. But there is evil in the world, so you have to add complexity to your theory to account for that. So the predictions are more like “implications about complications in your model” and we can still understand this process as one of trying to simplify our model.