Title: Noisy Hypotheses in the Age of Discovery Science

Abstract: We draw attention to one specific issue raised by Ioannidis (2005), that of very many hypotheses being tested in a given field of investigation. To better isolate the problem that arises in this (massive) multiple testing scenario, we consider a utopian setting where the hypotheses are tested with no additional bias. We show that, as the number of hypotheses being tested becomes much larger than the discoveries to be made, it becomes impossible to reliably identify true discoveries. This phenomenon, well-known to statisticians working in the field of multiple testing, puts in jeopardy any naive pursuit in (pure) discovery science.