The diffusion of innovation and new ideas is fundamental for economic growth, cultural shifts, and institutional change. How are new ideas created, and how do they diffuse? Specifically, how do spatial connections affect the generation and diffusion of ideas? Intuitively, a denser
network should increase the number of novel ideas and augment their diffusion. I study knowledge production in Germany in the 19th century, relying on several novel large-scale datasets, including the universe of bibliographic records, covering all published texts in all fields of knowledge, and detailed and comprehensive railway statistics. New ideas – both novel concepts and new combinations of existing ideas – are measured thanks to recent advances
in machine learning and topology applied to data analysis. I show that the railroad network increased the creation of new ideas, but decreased their diffusion. This was a by-product of specialization: with the railroad, groups of scholars could focus on narrower topics and co-locate with similar professionals; they learnt more from similar groups, but became disconnected from dissimilar ones. These patterns are those that paved the way for modern knowledge production, and they had important consequences for state institutions, on top of scientific discoveries and growth. Public officials and legislators were trained with narrower but deeper expertise. The bureaucracy became more specialized, and legislation more exhaustive.