“There’s No Data Like More Data”

Abstract

This article examines the role of automatic speech recognition research in the rise of data-driven machine learning as a privileged and pervasive form of computational knowledge. It focuses on IBM’s Continuous Speech Recognition group between 1972 and 1993 as they fueled speech recognition’s “statistical turn,” uprooting the field from the simulation of human reason and language understanding and redirecting it toward the acquisition of data for large-scale pattern recognition. This shift, I argue, was instrumental in the remaking of artificial intelligence and computational modeling into radically data-centric pursuits that underpin algorithmic culture today. In doing so, this history offers a critical piece in the story of how we became data-driven, highlighting how efforts to turn language into data consequently turned data into an imperative, preparing the way for the widespread incursion of algorithmic authority across everyday life.