Paper short abstract:

The “Wizard of Oz technique” used in Natual Language Processing research aims to anticipate technological innovation in IA by simulating a program with a human substitute. Analysing the epistemic specificities of this protocol sheds a new light on ritualisation and non-human interaction in science.

Paper long abstract:

Research in Artificial Intelligence is based on epistemic specificities, which the great debates that have accompanied its development since the 1950s have expressed without, however, circumscribing them with precision. By focusing on scientific practices themselves and, in particular, on a form of experience that is specific to the field of natural language processing, this paper aims to demonstrate how the tension represented by bringing the machine closer to the human being has produced a form of ritualization usually incompatible with science. The “Wizard of Oz technique” is based on the interaction of a human with a simulated machine. Its purpose is to anticipate the progress of computer science and its consequences. Following through scientific publications the uses of this protocol since its apparition in AI labs during the 1970s, I analyse how, by playing the role of the program, the researcher materializes an unrealized innovation that refers to a non-human - the AI - whose future existence is projected and personified.

This research received the support of the Human at home project (Université de Montpellier, FEDER, Occitanie)