But there also were non-Euclidean geometries that were discovered without any observation. It was not possible to test them until much later (and mostly because

But there also were non-Euclidean geometries that were discovered without any observation. It was not possible to test them until much later (and mostly because we knew where to look). Yet they turned out to better describe the world. I think what he is saying is that the choice of the hypothesis is not only governed by our observations, it's also governed by our brain that prefers hypotheses with specific properties. Not just "simple", more like "complete", "elegant" or "solid", without unseemly ad-hoc patches. For example, the original geocentric system with epicycles was actually much better in describing planet motions than the naive heliocentric idea of Copernicus, because Copernicus thought that planet orbits were circles and they travel with a constant speed. Yet it was accepted and worked upon (contrary to the facts!) because it was more elegant.