According to the World Economic Forum, machine translation should outperform human translation by 2024… Quality translation based on machine learning has recently (2016-2017) been made available online and for free, via programmes like DeepL or Google Translate. Among the issues raised by these new technologies is the shifting geography and sociology of research, and the power plays behind: who has been developing these tools, who has been financing them? I have conducted an exhaustive survey of the 50 most relevant research articles on neural machine translation published in the year 2017 and available on Google Scholar. Not only does it reveal a geographical shift towards Asia but also a disciplinary shift from Humanities to Engineering. This redefines the imaginary of translation, from the scholarly figure of St Jerome to an AI assistant. It also impacts the epistemology of translation and more generally the attitude towards language, moving from a grammar-based approach to a neural/mathematical approach, as evidenced in the corpus. Such reframing of an essential mode of transmitting and exchanging meaning will have economic and political implications. Because neural machine translation requires powerful IT capacities and huge training corpuses, one can expect a further concentration of its major providers, leading to an oligopoly. And since machine translation is destined to be embedded in most apps and connected devices, issues of reliability, accountability and privacy are bound to surface.