On the Praxeology of Perceptrons: Rebuilding “Mark I,” Respecifying Machine Learning

• In Event: Lab Studies Reloaded? Machine Learning, Ethnography, and Critical STS

Abstract

In the late 1950s, a group of researchers lead by psychologist Frank Rosenblatt built a prototype “perceptron” – that is, a first “pattern learning and recognition device” (CAL 1960:1). The device was inspired by a neural network model of brain information processing and, in its technical realization, filled an entire room of the Cornell Aeronautical Laboratory (CAL) in Buffalo, New York. Today, a “Perceptron [may be cast] in just a few Lines of Python Code” (Mavicc 2017). This contribution returns to the initial “Mark I Perceptron Operators’ Manual” (CAL 1960), intended as “a guide to the setting up and operation of the machine” (p. 1), to rebuild a mock-up version of “Mark I.” Insofar as the manual does “not require extensive familiarity with the theory of perceptron systems” (ibid.), it provides an apt opportunity to reflect upon “Mark I”’s operating principles and technical realization from a praxeological perspective – that is, a perspective which makes explicit (some of) the “vulgar enabling practices” (Button & Sharrock 1995) that those principles and this realization tacitly rely upon. In so doing, the contribution articulates experimental history of science (Fors et al. 2016) with ethnomethodology’s program of technical self-instruction (Garfinkel 2002) to respecify current STS of machine learning, mostly based on discursive analysis and controversy studies (e.g., Cardon et al. 2018; MacKenzie 2017; Olazaran 1996).

Authors

• Philippe Sormani, University of Lausanne

• Hunter Longe, California Art School