Algorithmic psychometrics and the scalable subject

— Luke Stark —

In June of 2014, researchers from Facebook and Cornell University found themselves in the midst of an unexpected crisis. The authors had published an article in the Proceedings of the National Academy of Sciences (PNAS) describing a 2012 experiment that had subtly changed the content of the News Feeds of almost 700,000 Facebook users. The study, as the PNAS paper described it, was to assess ‘whether [Facebook] posts with emotional content are more engaging’. The researchers were interested in the ‘positive’ or ‘negative’ emotional valences of a user’s Facebook posts, and how those posts might influence the emotional expressivity of users exposed to them – an effect known as ‘emotional contagion’. Yet when media outlets began to react to the published results of the PNAS study in the last week of June 2014, their framing of the work was almost uniformly negative: ‘Facebook manipulated 689,003 users’ emotions for science’, read a representative headline.

The Facebook emotional contagion study should be understood as a seminal event in the history of digital technology’s social impact: as a moment when the tacit co-development of the psychological and computational sciences became exposed to public view and – for some – a matter of concern. The site’s technical affordances, as Davis observes, ‘affect not just emotive expressions, but [also] reflect back to users that they are the kind of people who express positive emotions’. The emotional contagion study illuminated the subtler, more pervasive ways in which Facebook’s interaction design and analytic practices were already implicated in the day-to-day emotional states of the platform’s users.

Facebook’s interest in tracking, measuring and manipulating the moods of its users is indexical to a broader integration of the computational and psychological sciences with wide-reaching implications for the social use of digital media. This shared history stretches to the birth of modern electronic computing, but has been understudied by both historians and scholars in science and technology studies (STS): the Facebook “emotional contagion” study was one of the first incidents to make this psycho-computational complex a matter of twenty-first century controversy. For instance, the first, but certainly not the last, in May of 2017, a leaked document from Facebook’s Australian division suggested that the company had offered advertisers the ability to target advertisements to teenagers based on real-time extrapolation of their mood.

Facebook is by no means the only digital actor turning to mood tracking as part of a broader effort to collect behavioral and psychological data about users. Exemplified by a number of converging trends in human-computer interaction design, applied psychology, data science and advertising, the extraction, collection and analysis of data regarding human emotion is a bigger and bigger business, part of a broader ‘behavioral turn’ in digital commerce. Scholars have begun to document the emergence of what Zuboff terms ‘surveillance capitalism’ and its reliance on behavioral tracking and manipulation. Affect, emotion, and mood are critical yet curiously under-emphasized elements in this story

. A renewed focus from STS scholars on human emotion as a vector in the intersecting histories of psychology and computer sciences reveals not only how these disciplines have converged in the era of machine learning, big data and artificial intelligence, but how, as some scholars have already noted, they were never really separate in the first place.

One field at the center of the contemporary integration of psychology and computation is psychometrics, or ‘any branch of psychology concerned with psychological measurements’. In the spread of these forms of measurement via digital media platforms and systems, psychometrics is a powerful element in what Citron and Pasquale identify as ‘scored societies’, in which classificatory artifacts like credit scores, insurance profiles and social media data calculated and analyzed by machine learning algorithms and other computational tools are increasingly determinative of many aspects of our lives. In light of fracases like the Facebook emotional contagion study, where can the influence of psychological sciences be seen as components of the techniques and technologies of contemporary data science, machine learning, user experience design and massive data analysis? What are the political stakes in these relationships between computation and the psy sciences within systems of digital control?

I argue the human self as represented by digitally mediated psychometrics should be understood as a scalable subject. The ‘scalable subject’ could be understood as a refinement of the now-widespread notion of the ‘data double’ drawn from: the digitally articulated person is plastic, perpetually molded from both without and within. Informed by the history of psychology’s intersection with computer science, the subject of digital control is not only plastic but also scalable: shaped and made legible at different orders of technical analysis by those affordances of social media platforms grounded in psychological science, and thrown back to the human person as a model with which to conform or suffer. In highlighting the through-lines and contemporary stakes of this history, I show the figure of the scalable subject as foundational to the contemporary ‘computational politics’ described by Tufekci, in which progressive forces are already at a disadvantage. In our epoch of digital control, the merged norms and design tenets of computation and psychology govern the ordering of people and things. As such, we need a clear articulation of how the logics of a psycho-computational complex have developed in tandem with, and reinforced, the techniques and technologies of the human sciences and the political stakes of the scalable subjects they have made.

For broader background and references, see:

Stark L (2018) Algorithmic psychometrics and the scalable subject. Social Studies of Science 48(2): 204-231.

Luke Stark is a Postdoctoral Fellow in the Department of Sociology at Dartmouth College and a Fellow at the Berkman Klein Center for Internet & Society at Harvard University.