I study a major transformation that occurred over last three decades. Now, every day, as we open our computers and phones we interact with and through machine learning. I ask why machine learning has spread since the mid 1990s across retail, finance, love, security, music, healthcare, manufacturing, and other institutions. Machine learning is a category of algorithms that “learn by themselves.” Programmers need relatively little a priori knowledge about the substantive domain. Instead, programmers decide over which model to use and how to adjust the model to the data based on the model’s predictive accuracy.
The pervasive character of machine learning makes it important as well as theoretically interesting. Mauss (1990 [1925]:3) pointed to the richness of studying “total social phenomena,” or those in which “[e]verything intermingles in them.” In machine learning, following Mauss, “all kinds of institutions are given expressions at one at the same time.” The total character of machine learning provides an empirical window into why technologies of surveillance have come to “touch daily life at every point,” giving rise to “surveillance societies” (Lyon 2003:5).
I propose a novel explanation for why machine learning became a total social phenomenon. My emerging argument focuses on the interaction between the commodification of machine learning and a discourse favorable to automation. Technology corporations and consultants have made machine learning easier to buy and more interchangeable through cloud computing and thanks to easy-to-use software. Organizational decision-makers desire machine learning because they draw on a discourse favorable to automation. They use this discourse to frame organizational problems, imagine an automated future, and narrate a path from present problems to the future.
The taken-for-granted explanation for why machine learning is so pervasive is that “it works.” But that is not so clear. Consultants selling machine learning acknowledge that measuring the return on investment of machine learning is not easy. In 1996, a national partner of what is now PricewaterhouseCoopers encouraged corporations to “take on faith” that projects “will affect the bottom line.”[1] Similarly, in 2018 two Deloitte-affiliated consultants argued that “[a]lthough the early successes are relatively modest, we anticipate that these technologies will eventually transform work.”[2] Empirically, this evidence forces us to consider a new explanation. Theoretically, we may want to place instrumental rationality in a temporal context.
Emilio Lehoucq is a Ph.D. candidate in sociology at Northwestern University. He is interested in science and technology, social movements, culture, law, religion, statistical learning, and comparative-historical methods. His dissertation examines why machine learning became pervasive across the private and public sectors since the mid 1990s. His work has been published in Law & Social Inquiry and Mobilization.
References
Lyon, David. 2003. Surveillance After September 11. Cambridge: UK: Polity Press.
Mauss, Marcel. 1990 [1925]. The Gift: The Form and Reason for Exchange in Archaic Societies. London: W.W. Norton.
[1] Jim Knowles, Explore Data Warehousing, Datamation, Vol. 42, No. 18, 1 December 1996.
[2] Thomas H. Davenport and Rajeev Ronanki, Artificial Intelligence for the Real World: Don’t Start with the Moon Shots, Harvard Business Review, January-February 2018, p. 116.
I propose a novel explanation for why machine learning became a total social phenomenon. My emerging argument focuses on the interaction between the commodification of machine learning and a discourse favorable to automation. Technology corporations and consultants have made machine learning easier to buy and more interchangeable through cloud computing and thanks to easy-to-use software. Organizational decision-makers desire machine learning because they draw on a discourse favorable to automation. They use this discourse to frame organizational problems, imagine an automated future, and narrate a path from present problems to the future.
The taken-for-granted explanation for why machine learning is so pervasive is that “it works.” But that is not so clear. Consultants selling machine learning acknowledge that measuring the return on investment of machine learning is not easy. In 1996, a national partner of what is now PricewaterhouseCoopers encouraged corporations to “take on faith” that projects “will affect the bottom line.”[1] Similarly, in 2018 two Deloitte-affiliated consultants argued that “[a]lthough the early successes are relatively modest, we anticipate that these technologies will eventually transform work.”[2] Empirically, this evidence forces us to consider a new explanation. Theoretically, we may want to place instrumental rationality in a temporal context.
Emilio Lehoucq is a Ph.D. candidate in sociology at Northwestern University. He is interested in science and technology, social movements, culture, law, religion, statistical learning, and comparative-historical methods. His dissertation examines why machine learning became pervasive across the private and public sectors since the mid 1990s. His work has been published in Law & Social Inquiry and Mobilization.
References
Lyon, David. 2003. Surveillance After September 11. Cambridge: UK: Polity Press.
Mauss, Marcel. 1990 [1925]. The Gift: The Form and Reason for Exchange in Archaic Societies. London: W.W. Norton.
[1] Jim Knowles, Explore Data Warehousing, Datamation, Vol. 42, No. 18, 1 December 1996.
[2] Thomas H. Davenport and Rajeev Ronanki, Artificial Intelligence for the Real World: Don’t Start with the Moon Shots, Harvard Business Review, January-February 2018, p. 116.
RSS Feed
