The dangerous irrelevance of string theory

We comment on the relation between string theory and empirical science, grounding our discussion in cosmology, a subject with increasingly precise data in which this connection operates at several levels. It is important to take into account the phenomenon of dangerous irrelevance: over long times or large field ranges, physics can become sensitive to higher scales than the input energies. This pertains in inflationary cosmology (and possibly other aspects of horizon physics). String theory also contributes to our understanding of observational constraints and search strategies at the level of low energy field theory. We illustrate this with a current example concerning a new form of non-Gaussianity generated by very massive degrees of freedom coupling to the inflaton. New constraints on such fields and couplings can be obtained from existing data, increasing our empirical knowledge of the universe. This builds in part from the development of the string landscape, which is neither random nor an abdication of science as has sometimes been suggested. {\it Invited contribution to the proceedings of the conference `Why trust a theory'.}