The beginning of systematic MT programs took place in the USA, where the first experiments started in 1949 in the context of the Cold War, and ended in 1966 with the publication of the ALPAC report (Automatic Language Processing Advisory Committee).The interest of reconsidering this short initial period of about fifteen years is firstly because it was rich in a profusion of ideas of all kinds which were implemented for MT with an astonishing and unprecedented enthusiasm and optimism. Secondly, it allows us to examine how the theoretical and social expectations of the actors had to be confronted with the constraints posed by the limitations of early computers and of linguistic research Successive stances can be observed. Each of them had to cope with the issues of quality, speed and cost. Three key personalities presided over the advent of early MT, each representing a particular angle of attack and stance, Warren Weaver, Andrew D. Booth and Yehoshua Bar-Hillel. In my paper, I will examine to what extent these stances were able to overcome the diverse constraints, to what extent they were opposed or, on the contrary, found some sort of agreement. Finally, one may wonder to what extent they led to the ultimate failure of MT or whether, conversely, they opened the way to new scientific fields.