Researchers and policy makersagree that new social norms could help solve large-scale problems, from climatechange to antibiotic resistance. However, our understanding of how norms changehas been limited so far by the lack of suitable data. In this talk, I willdiscuss two recent studies that shed light on this process. In the first [PNAS115, 8260 (2018)], we examined linguistic norm shifts in English and Spanish.We identified three main drivers of norm change that leave markedly differentsignatures in the data, namely (i) authority, (ii) informal institutions and(iii) a bottom-up process triggered by a small number of committed users (akinto a 'critical mass' phenomenon). We proposed a simple model that reproducesthe empirical observations. In the second study [Science 360, 1116 (2018)], wefocused on critical mass theory and tested it experimentally in artificialsocial networks. We let a group of individuals evolve their own socialconvention. Then, once the agreement was reached, we introduced fewconfederates pushing for a different norm. As their number crossed a tippingpoint - roughly 25% of the group size – the whole population would follow themand adopt the new norm. This is the first empirical evidence for the widelyadopted theory of critical mass. These results will help better understand bothhow norms change spontaneously in our societies and how to design effectivepolicies to foster collective behavioral change.