If all you've seen is a shadow, all you've seen is a shadow. How do you know it's not someone's twin? Or some sort of puppet facsimile?
Now, that's a contrived example. More realistically, I doubt we'll ever be able to get enough detailed measurements to really understand how genetics, nutrition, exercise, and environment all play together to affect human health. Clearly each is complex enough on its own and to combine them together quickly magnifies the observation problem to be entirely intractable. We can imperfectly generalize and work with populations to come up with various guidelines that seem to work well with populations. We can break down the problems, measuring particular cells or organs. But to recompose all the experimentation and observation into a full system view is much more complex than solving chess or go. It's much more complex than predicting the earnings forecast of a single company. The complexity of the problem space is just ridiculous.
So we try to push the boundaries of the complexity we can deal with, but there are diminishing returns over time. And clearly there are outer bounds we will never be able to approach.
True enough, but I don't see how any of that is specific to empiricism? This doesn't sound like a case where empiricism overcomplicates and a non-empirical approach could do better; rather it seems like reality is just that complex, and our only options are to deal with it or to simplify and accept a coarser picture (which is perfectly compatible with empiricism).
Now, that's a contrived example. More realistically, I doubt we'll ever be able to get enough detailed measurements to really understand how genetics, nutrition, exercise, and environment all play together to affect human health. Clearly each is complex enough on its own and to combine them together quickly magnifies the observation problem to be entirely intractable. We can imperfectly generalize and work with populations to come up with various guidelines that seem to work well with populations. We can break down the problems, measuring particular cells or organs. But to recompose all the experimentation and observation into a full system view is much more complex than solving chess or go. It's much more complex than predicting the earnings forecast of a single company. The complexity of the problem space is just ridiculous.
So we try to push the boundaries of the complexity we can deal with, but there are diminishing returns over time. And clearly there are outer bounds we will never be able to approach.