In September 1902, Emil Bloch applied to the US Patent Office for protection of a new method of producing a “Combined Looking and Display Glass.” He called this glass a “transparent mirror,” meaning that the glass could be both “highly opaque”—reflective—from one side and “clearly transparent” from the other. Treated with the cheap and durable chemical backing that he had discovered and then subjected to specific lighting conditions, the glass was a mirror from the brightly lit front, while a window from the darkened rear.
Bloch envisioned his transparent mirror as an advertising display device. But it soon proved useful in a wide range of surveillance and monitoring situations: watching interrogations in police stations, scrutinizing people in public places by hiding security personnel or cameras, viewing test subjects in market research and psychology experiments. Some versions have been used in industry to allow an employer to view employees’ workspace. In all these applications, the observers can see but not be seen. This allows them to either avoid affecting the people being observed or, if the presence of the mirror is known, alert the observed to the fact that they are being watched. Now generally called “one-way” or “two-way” mirrors, these devices have an inherent ambiguity that is better captured by Bloch’s use of “opaque” and “transparent” for the two different sides of the glass. The distinctive positions split viewpoints into objective observer and subjective observed and create a corresponding asymmetry of power.11xDan Graham, Two-Way Mirror Power: Selected Writings by Dan Graham on His Art, ed. Alexander Alberro (Cambridge, MA: MIT Press, 1999), 54.
The image of the transparent mirror conjures up memories of old cop shows and the famous social psychological laboratories of the 1950s, where researchers turned out disquieting studies of influence and conformity. But if the technology is dated, the surveillance and monitoring functions it served and the dynamics of its mode of observation are not. These functions and dynamics have been taken over by a mushrooming number of electronic methods, from video cameras and GPS tracking to point-of-sale systems and productivity apps. Carried out by machines, these methods are less costly and obtrusive, more efficient and comprehensive, and have, in turn, contributed to a dramatic expansion of the scope and reach of unobserved observation. Our actions are, with growing frequency, whether in real time or retrospectively, being watched.
Nowhere has this power of disembodied observation become more pervasive than in the workplace. Managerial supervision is, of course, a defining component of modern organizations, and one expected by employees. Unsurprisingly, the incentive to increase surveillance has grown as more and more employees work remotely, on a “contract” or “freelance” basis. Without the personal contact of a shared physical space, there is less opportunity to develop a relationship of trust.22xIfeoma Ajunwa, Kate Crawford, and Jason Schultz, “Limitless Worker Surveillance,” California Law Review 105, no. 3 (2017): 735–76. See 746. Over time, workers have become largely accustomed to the conventional reasons why employers monitor their work—from maintaining productivity to protecting business assets and trade secrets to addressing concerns about legal liability. And organizations themselves are under powerful institutional and regulatory pressure to increase practices of inspection, measurement, and assessment to demonstrate their efficiency and accountability.33xMichael Power, The Audit Society: Rituals of Verification (Oxford, England: Oxford University Press, 1997).
But with the proliferating means to electronically track, monitor, and capture information about people, answers to the questions of what is “under surveillance,” who is doing it, and to what purposes are changing rapidly. Especially in office environments, the workers themselves and their interactions are of growing importance. Information about their everyday actions and choices, from web browsing to Twitter feeds, or captured by company software systems and sensing devices, has become the “data” for a new type of data-based organizational science and reengineering project. Like miners drilling a mountain bearing a lode of precious metal, data scientists aim to unveil features of employees’ lives—their behaviors, personal characteristics, relationships, and networks—that have hitherto been opaque to conventional methods of observation. The patterns and trends extracted from this “dark data” can then be engineered into influence systems that will enhance productivity, increase efficiency, and predict future events.44x“Dark data,” according to the research company Gartner, is “the information assets organizations collect, process and store during regular business activities, but generally fail to use for other purposes (for example, analytics, business relationships and direct monetizing). Similar to dark matter in physics, dark data often comprises most organizations’ universe of information assets.” Gartner, Inc., “Gartner Glossary: Dark Data,” accessed March 6, 2020, https://www.gartner.com/en/information-technology/glossary/dark-data.
The new science builds on the old positivist dream of Auguste Comte for la physique sociale, a “social physics,” by which human affairs can be managed purely through scientifically discovered facts and laws of behavior.55xSee Joseph E. Davis, “Toward the Elimination of Subjectivity: From Francis Bacon to AI,” Social Research 86, no. 4 (Winter 2019): 845–69, for historical representations of this dream over the course of the modern age. Now, with the power of Big Data, human complexity can finally be captured and quantified, replacing opinion, guesswork, and gut feelings with a machinery of social behavior and skill in manipulating it. In the quantified workplace, the workers themselves will be no less active than management as tools for shaping an ever more productive and harmonious team. Supervision does not disappear. With the new machinery, it is extended outward, inward, and laterally. The managers remain, but now the workers join them. They too are drawn behind the mirror, not only to exert unobserved performance pressure on fellow workers, but to monitor their own behavior from a detached, managerial perspective and intensify their effort.
From Docile Bodies to Strategic Partners
Today, every procedure, text, post, like, chat, e-mail, call, search, file transfer, site visit, and swipe can be encoded and stored. People’s movements and proximities are easily tracked, as are their tastes, opinions, moods, and health status. In the quantified workplace, management systems, social platforms, navigation software, and many other collection tools have created a situation in which “real-time information on employee engagement, actions, sentiment, productivity, location, quality, and aspirations is everywhere, making it possible to tie talent [employees] to business outcomes in ways that were almost impossible before.”66xJean Paul Isson and Jesse S. Harriott, People Analytics in the Era of Big Data (Hoboken, NJ: Wiley, 2016), xiii. A new era of employee “SuperVision” has arrived.77xJohn Gilliom and Torin Monahan, SuperVision: An Introduction to the Surveillance Society (Chicago, IL: University of Chicago Press, 2012). Collecting and turning all this data into a new asset has rapidly become a high priority for a majority of enterprises, large and even small. There is value to be extracted, and “those who don’t seize the opportunity,” according to IBM, “risk falling behind their competitors.”88x“What Is Predictive Analytics?,” IBM, https://www.ibm.com/analytics/predictive-analytics. Some have suggested that it would also be beneficial to investors to incorporate new measures of a firm’s employees into their calculus for forecasting its earnings.
The new science being brought to bear on human resource management is a combination of various Big Data fields, going under names like “people analytics,” “predictive analytics,” and “workforce analytics.” All are forms of machine-based computational science, applying powerful analytic techniques to many types of quantified data in order to compile a fine-grained picture of individuals and collectives that can be used to further rationalize and influence behavior for societal and organizational purposes.99xDavid Lazer et al. “Computational Social Science,” Science 323, no. 5915 (2009): 721–23. For employers, the new science, which I will refer to generically as “workforce science,” promises to produce novel insights into their employees and organizational cultures, and generate a big payoff. According to its thought leaders, by embedding the science within their business processes, companies can spur motivation and innovation, enhance performance and engagement, improve team chemistry and individual accountability, attract and advance the right people, and in other ways gain a competitive edge and better business results.
As every employer knows, workplace problems are common. For one, employees are seldom engaged in their work in the way their organizations want them to be. According to the research company Gartner, which conducts quarterly workplace surveys, “Globally, employee engagement is low, and has been for at least the past two decades.”1010xJackie Wiles, “What Is Employee Engagement?” Smarter with Gartner (blog), October 15, 2019, https://www.gartner.com/smarterwithgartner/what-is-employee-engagement/. In its report for the second quarter of 2019, Gartner states that only about “15 percent of the US workforce could be described as engaged…reporting both high discretionary effort [going above and beyond the call of duty] and high intent to stay.”1111xR.J. Cheremond, “Far Fewer US Employees Are Job Hunting,” Smarter with Gartner (blog), September 26, 2019, https://www.gartner.com/smarterwithgartner/far-fewer-u-s-employees-are-job-hunting/. Compared to the global average of about 8 percent, this figure is actually high.
The polling and consulting firm Gallup, which also conducts regular workplace surveys, defines engaged more subjectively, in terms of enthusiasm and commitment to work and the workplace. Gallup reports that the majority is in the “not engaged” category in every country, including the United States, which has relatively more engaged workers (about one-third by the Gallup definition) and high numbers in professional and knowledge-based jobs. The “not engaged” people, according to Gallup, are “psychologically unattached to their work and company…putting time—but not energy or passion—into their work.”1212xState of the Global Workplace (2017), Gallup, https://www.gallup.com/workplace/238079/state-global-workplace-2017.aspx.
Further, employee malaise is widespread. Concerns include depression, anxiety, fatigue, sleep disturbances, “anger management” issues, substance abuse, and a wide variety of other ill-defined personal troubles. These problems show up at work in testy conversations, team conflict, and “withdrawal behaviors” such as abrupt resignations, chronic absenteeism, and presenteeism (coming to work but doing little). In the United States, for instance, estimates of the economic costs generated by people who screen positive for depression rose from $83.1 billion in 2000 to $210.5 billion in 2010. Half of this total cost—the nonmedical part—derived from lost productivity in the workplace. Presenteeism was by far the most significant workforce cost (nearly $80 billion), followed by absenteeism due to reported illness, unwillingness to go to work, or short-term disability.1313xPaul E. Greenberg et al., “The Economic Burden of Adults with Major Depressive Disorder in the United States (2005 and 2010),” Journal of Clinical Psychiatry 76, no. 2 (2015): 155–62. In response, workplace wellness programs, combining mental and physical health interventions, have cropped up across the world of work.
Many of the persistent problems are concentrated among younger workers, who, compared with any older generation, have markedly different expectations of work. Members of the younger generation, the so-called millennials, change jobs more often, are less engaged in the workplace, and are more likely to be categorized as “not engaged.” According to a Gallup study, millennials “want to be free of old workplace policies and performance management standards, and they expect leaders and managers to adapt accordingly.” They “don’t want bosses—they want coaches.” Used to constant communication and feedback through social media, they “don’t want annual reviews—they want ongoing conversations” that aim to “develop their strengths,” not “fix their weaknesses.” More than a paycheck, “they want a purpose” and sense of mission, and lack tolerance “for workplaces they believe stunt their growth.” Engaging the younger generation has been a deep management challenge that the installation of Ping-Pong tables and latte machines has done nothing to address.1414xState of the Global Workplace. Also, How Millennials Want to Work and Live (2016), Gallup, https://www.gallup.com/workplace/238073/millennials-work-live.aspx.
For organizations, the promise to identify and reduce employees’ “wasted potential” is not merely a means of increasing efficiency and innovation. It is also a means of surviving in an ever tougher competitive environment. Earlier employee monitoring strategies were meant to shape workers as “docile bodies” and to strengthen supervision and control by managers. But the goal now is different. A smart company, according to the new science, must have active, fulfilled, and engaged employees, who are “strategic partners” in the enterprise, motivated to a self-mastery that adds business value.1515xWiles, “What Is Employee Engagement?” Wiles writes that “repeated studies over the past 10 years have shown that the business performance of organizations in the highest quartile of employee engagement scores outpaces that of competitors. High employee engagement correlates with higher average revenue growth, net profit margin, customer satisfaction and earnings per share.” Analytics holds the keys to producing them.
The Quantified Worker
According to workforce science, when it comes to hiring, managing, and retaining employees, companies traditionally have been flying blind. Dependent on interpersonal relationships, established practice, intuition, and gut feeling, the old approach is loose, fuzzy, business-as-usual, but not remotely “evidence-based” or “data-driven.”1616x Isson and Harriott, People Analytics, xiv. While employee reviews, surveys, and feedback channels that gauge performance and perspectives are useful, they are infrequent and require substantial resources. The information they collect is static and typically subjective. Hiring and retention are often plagued by the same lack of precise and “real-time” intelligence.
Workforce science, on the other hand, claims to revolutionize what can be seen and empirically measured. Alex Pentland, head of the MIT Human Dynamics Lab and analytics pioneer and entrepreneur, proposes the following thought experiment. Imagine “if you could see everybody in the world all the time, where they were, what they were doing, who they spent time with, then you could create an entirely different world.” (Pentland makes clear that he also thinks this would be a far better world.1717xInterview with Alex Pentland, “Reinventing Society in the Wake of Big Data,” Edge.org (blog), August 30, 2012, https://www.edge.org/conversation/alex_sandy_pentland-reinventing-society-in-the-wake-of-big-data. ) On the scale of the firm, such a “God’s-eye view” of employees is the stated ambition of workforce science, implying a transparent, uncooked representation without a priori structuring, strategy, or social dynamics. In the torrent of data that organizations are amassing or have available to them there reside, the science promises, observable indicators that, when organized and analyzed, enable a high-resolution seeing of employees and applicants. Armed with this detailed and putatively objective view of their personnel, organizations can engineer—by predicting and influencing—whole new levels of engagement, efficiency, performance, and even creativity.
The necessary high-volume (i.e., “big”) data can be captured from employees or more actively produced by direct surveillance technologies.1818xOn the distinction between surveillance and “capture,” see Philip Agre’s still relevant “Surveillance and Capture: Two Models of Privacy,” The Information Society 10 (1994): 101–27. My thanks to Samuel Lengen for bringing this article to my attention. Captured data include both the normal employee records generated over time, such as raises, performance ratings, and sick time used, and the “digital footprints” that are created by employees’ everyday interactions with technologies that have been specifically designed to capture information and activity—all those texts, e-mails, Slack messages, Microsoft Teams, and Trello to-do lists. More direct surveillance technologies are typically used to produce data about dynamic and qualitative aspects of employees and employee interactions. Automated sensors, trackers, and detection systems used or tested for the workplace can monitor facial expressions, face-to-face conversations, activity patterns, physical posture, and much else.1919xFor a brief review of applications, see Roy Gelbard et al., “Sentiment Analysis in Organizational Work: Towards an Ontology of People Analytics,” Expert Systems 35 (2018): e12289.
For workforce science, all this microactivity data is just the beginning. To provide new and actionable knowledge, it has to be translated into meaningful observations. First, the data has to be “read” by a machine. For example, consider the rapidly growing area known as “sentiment analytics.” In workplace applications, it involves efforts to gauge employee emotions by inferring their moods and feelings from the digital footprints of whatever written communications are being monitored or from any sensor data in use. Making such inferences is extremely complex and error-prone. How, based on tone of voice, cell phone use, Facebook Workplace messages, or facial expressions, does a computer accurately identify if employees are sad, content, disengaged, or good for team chemistry? Emotions have to be theoretically modeled and then incorporated into the design of algorithms and machine learning programs that scour the data to detect, decipher, classify, and tag the modeled sentiments.
Machine reading, more or less by itself, is sufficient for some surveillance applications. Generally, these are applications in which simple detection is at issue. Sentiment analysis based on the simple content of monitored communications, for instance, can be used to determine the positive or negative feelings of employees toward new initiatives, changes in benefits, new management practices, and the like. Software programs have been designed that scan e-mail accounts, computer files, or website visits and flag anything predetermined as inappropriate or offensive, whether profanity or the web address of a pornography site. Using digital footprints, managers can track patterns of communication among people in different departments, creating a picture of an employee’s network and its limits. Comparing a worker’s “constraint score” with those of other employees allows management to see whose networks are too “inbred.”2020xPaul Leonardi and Noshir Contractor, “Better People Analytics,” Harvard Business Review (November–December 2018), 80. Such forms of close observation enable direct management interventions.
But in the larger ambition of workforce science, the real action is in a second step of analysis, in which data is cleaned, organized, and analyzed to detect patterns, trends, and mathematical probabilities. Given the size of data sets and the common practice of triangulating data from multiple sources, algorithms and machine learning are needed to combine data, search for correlational patterns, and conduct statistical analysis. Using these tools on the trove of routinely collected employee data—salaries, company tenure, education level—the computer manufacturer HP Inc. (formerly Hewlett-Packard) sought to identify and quantify the factors that drive employee attrition. Comparing those who quit to those who stayed, HP data scientists singled out a combination of variables, their relative importance, and how they interact to generate a computer model of which types of employees are most loyal and which most likely to leave. From their pattern analysis (subsequently continuously updated), they created a “flight risk” score for every employee in the company (more than 300,000 in 2011).2121xEric Siegal, Predictive Analytics (Hoboken, NJ: Wiley, 2013), 45–51. As is common with these analytics, the score is not any sort of direct measure; it is a prediction. While analysis is based on the behavioral data and attributes collected from individual employees, the interpretive key is in the statistical regularities that hold true across the whole data set. From these regularities, probable characteristics and behaviors of employees—loyalty in the HP case—can then be forecast.
Prediction from Big Data is based on the underlying assumptions that what we do is determined by our social environment and that high-volume data always contains necessary regularities. A particularly influential formulation of these assumptions, reflecting the growing ambition to apply physics to human social interaction, is the social physics approach pioneered by Pentland, the MIT data scientist and entrepreneur mentioned above.2222xA number of books employing social physics have been published in recent years. These works, all by physicists, include Mark Buchanan, The Social Atom: Why the Rich Get Richer, Cheaters Get Caught, and Your Neighbor Usually Looks Like You (New York, NY: Bloomsbury, 2007); Philip Ball, Why Society Is a Complex Matter: Meeting Twenty-First Century Challenges with a New Kind of Science (Berlin/Heidelberg, Germany: Springer, 2012); Serge Galam, Sociophysics: A Physicist’s Modeling of Psycho-Political Phenomena (New York, NY: Springer, 2012); and Albert-László Barabási, Linked: How Everything Is Connected to Everything Else and What It Means for Business, Science, and Everyday Life (New York, NY: Basic Books, 2014). Human behavior, according to Pentland, is largely automatic, shaped by implicit social learning (copying the actions and strategies of others) and directed to shared habits of action. Individual thoughts and reasons—“internal cognitive processes”—which are unobservable, matter little.2323xAlex Pentland, “The Death of Individuality,” New Scientist (April 5, 2014): 30–31; and Social Physics: How Social Networks Can Make Us Smarter (New York, NY: Penguin Books, 2015), 16, 190. Pentland is fond of drawing analogies from other social primates, which also “exhibit behavioral cultures” with “idea flows” and “repertoires of habits,” Social Physics, 45. Really understanding why people behave as they do is not a question of their personal beliefs or intentions; it is a question of social dynamics and influence, of who is communicating what, with whom, and to what effect on changes in behavior. Like the motion of particles in physics, these dynamics—“one person’s state impact[ing] other people’s states and vice versa,” in Pentland’s words—can be defined and modeled in a formal, mathematical way. This computational approach is possible, Pentland claims, because social physics has discovered statistical regularities (“social behavioral laws”) in large data sets that govern the “mechanisms of social interactions” over time within any network structure, from the level of companies to a whole society.2424x Pentland, Social Physics, 5–6, 81.
The key to prediction, Pentland argues, is information about the exposure of people to the behavior of others in a social network or group. “You can tell all sorts of things about a person,” he observes in an interview, “even though it’s not explicitly in the data, because people are so enmeshed in the surrounding social fabric that it determines the sorts of things that they think are normal, and what behaviors they will learn from each other.” The critical task of Big Data analytics, he continues, is in “finding connections” among people and between their behavior and particular outcomes.2525xPentland, “Reinventing Society in the Wake of Big Data.” Knowing these connections and their mathematical properties makes predictions (inferences) possible. They allow a calculation of behavior (at individual and collective levels) by providing a statistical estimate of the social influences on behavior.
By itself, such sophisticated data analysis may be of little new organizational value. It has to be “leveraged effectively,” according to a book on people analytics, “to drive and optimize business action that is coordinated at all levels of the organization.”2626xIsson and Harriott, People Analytics, 332. New abilities to see the network and group interactions that shape employee, team, and organizational behavior don’t automatically generate the power to manage and promote efficiency and innovation. “Many network analyses companies do,” according to a Harvard Business Review article, “are little more than pretty pictures of nodes and edges.”2727xLeonardi and Contractor, “Better People Analytics,” 80. The further challenge for workforce applications is how to use the dynamics of social influence being uncovered by analytics to compel people to action and foster deeper engagement and aggregate productivity gains. Although some data scientists speak of analytics as another tool for the hands-on managers, the larger goal is to engineer systems, to find ways that organizations can design and wield centralized and automatic forms of social influence.
Influence Technology
Talk of worker surveillance and centralized control has a distinctly dystopian ring to it. Even in the enthusiastic writings of data scientists like Pentland, references to George Orwell, 1984, and “Big Brother” abound. These writers do not shy away from issues of privacy, data ownership, and organizational transparency. To the contrary, they admonish employers to “get ahead” of privacy concerns and typically recommend that they “tread lightly” with their employees, telling them what information is being collected about them and keeping some analytics at a team or higher level that does not identify specific individuals. The goal is to foster engagement not alienation and resistance.
In fact, employees’ participation in their own surveillance is critical to the influence methods of the new workforce science. Companies have long recognized that allowing employees to exercise some decision making and control over their work leads to increased cognitive engagement and flexibility. Excessive monitoring and hierarchical supervision, by contrast, leads to distress and disconnection. New models that took hold in the 1980s, such as “total quality management,” which reorganized work on the basis of teams and devolved responsibility, came with promises of employee empowerment, discretion, flexibility, and fulfillment.2828xGraham Sewell, “The Discipline of Teams: The Control of Team-Based Industrial Work through Electronic Surveillance,” Administrative Science Quarterly 43 (1998): 397–428. See esp. 400–401. Unlike older models, they were compatible with increasing employee skill levels. Organizations sought to build a culture that fostered trust and commitment, encouraging peer influence and inculcating “excellence” and the “habits of highly effective people.”2929xThe quotes are from the titles of two best-selling books from the 1980s: Thomas J. Peters and Robert H. Waterman Jr., In Search of Excellence: Lessons from America’s Best-Run Companies (New York, NY: Harper & Row, 1982); and Stephen R. Covey, The Seven Habits of Highly Effective People: Powerful Lessons in Personal Change (New York, NY: Simon and Schuster, 1989).
Workforce science casts the new surveillance not only as a windfall for business efficiency and productivity but as a powerful tool for employee emancipation and self-improvement. Drawing on many of the same technologies that generate Big Data, including apps and mobile phones, engineers are using analytics to develop automatic, instrumented systems to influence employee engagement and performance. The systems work by flexibly leveraging workers’ own intrinsic motivations—their preexisting relationships and their desire for social acceptance and recognition. They especially target the high performers and key influencers, spurring them to use their ingenuity, knowledge, and connections for company purposes and systematizing their contributions and setting them up as standard-bearers for others. Since the dynamic is a social one, employee immersion is essential.
This view of human motivation is neither novel nor revolutionary in the world of work. In his 1956 book Work and Authority in Industry, the sociologist Reinhard Bendix observed that for Elton Mayo, the key theorist of the human relations school of management that emerged in the 1930s, productivity hinged on positive social relations with fellow employees. The key principle of worker motivation, according to Mayo, was “the desire to stand well with one’s fellows, the so-called human instinct of association,” which “easily outweighs the merely individual interest” that so much of management practice was then focusing upon.3030xQuoted in Reinhard Bendix, Work and Authority in Industry: Ideologies in Management in the Course of Industrialization (Berkeley, CA: University of California Press, 1974), 313. First published 1956. People on the job, in short, are not just pursuing wages; they prize informal group life and their subjective relations with others. For workforce science, mobilizing and steering this “eager human desire for cooperative activity” toward greater employee self-mastery is just the challenge toward which its analytics and human-machine designs are directed.3131x Ibid., 317.
Among the new technological applications is “gamification,” the integration into surveillance systems of the data-driven techniques that smart-game designers use to engage players. The developer Betterworks, for example, offers “Continuous Performance Management” software programs that also use the dynamics of smart-game technology, such as rapid feedback, competition, targets, and leveling up. These programs, according to the company, replace “outdated annual review processes.” They are a form of “quantified work,” which, like the fitness tracking of the Quantified Self movement, provide a “way to capture progress and gain feedback on a frequent basis using enterprise systems intended for everyday use.”3232x“So What Exactly Is Quantified Work?” Betterworks (blog), February 4, 2014, https://blog.betterworks.com/so-what-exactly-is-quantified-work/. To spur productivity through social reinforcement, employees’ progress toward target goals is quantified and displayed on a digital dashboard that other employees can see. Progress is represented, to quote a media story, by “a digital tree that grows with accomplishments and shrivels with poor productivity.” The visibility exposes workers to peer scrutiny, and “cheers” or “nudges” impel them toward higher performance.3333xConor Dougherty and Quentin Hardy, “Managers Turn to Computer Games, Aiming for More Efficient Employees,” New York Times, March 16, 2015, https://www.nytimes.com/2015/03/16/technology/managers-turn-to-computer-games-aiming-for-more-efficient-employees.html.
Social pressure and reinforcement by peers are also at the heart of Pentland’s social physics approach. Platforms from companies like Betterworks combine Big Data analytics with gamification and an influence model drawn primarily from behavioral economics. Pentland, who has a background in psychology, also relies heavily on aspects of social learning theory and cites the influential work of Albert Bandura, whose pivotal idea was that people learn by observation: imitating and identifying with the attitudes and behaviors, successes and failures, they observe in the people they model themselves on.3434x Albert Bandura, Social Learning Theory (Englewood Cliffs, NJ: Prentice Hall, 1977). Unlike B.F. Skinner, whose behaviorism completely rejected the role of mental processes, Bandura held that the move from learning to change is not an automatic or mechanistic process. He suggested that mental factors, like attention and motivation, mediate between environmental stimuli and human action. Pentland, too, sees mediating factors, particularly personality. What the Big Data analytics reveal, it is claimed, is how, via the mechanism of social learning, ideas and information flow in communication networks, and how, in turn, this flow translates into changes in behavior (new “habits of action”) and shapes the norms, productivity, and creative output of organizations.3535xI am only focusing on work here, but Pentland’s ambitions extend to larger social entities, including cities and whole societies.
Changing behavior in the desired direction, toward innovation and productivity, requires employers to understand and actively shape—in Pentland’s usage, “tune”—three network variables: (1) the relevant social network structure, (2) the strength of social influence among people in the network, and (3) their individual susceptibility to new ideas. Variable 1, the social network structure, is crucial because it needs to have enough reach and diversity to explore and “harvest” new beliefs and turn them into habits. Groupthink is bad and contrarians are important, so the network may need to be tuned to generate new patterns of interaction. On the other hand, because fads are bad and cohesiveness is important, the tuning may be directed to enhancing and enforcing uniformity and cooperation.
Variable 2, the strength of social influence among people in the network, affects whether new modeled behaviors will lead to changes in habits and behavioral norms. Tuning directed to enhancing repeated interaction and engagement, both through digital interaction and face to face, will make the rewards of being copied and the social pressure to do the copying and adopt new behaviors more effective and enduring. If employees are not interacting, Pentland advises, “get them talking.”3636xPentland, Social Physics, 77.
Variable 3, individual susceptibility to new ideas, while related to the strength of the relationships that can be leveraged, is also a matter of personality. Psychological variables can mediate if and how people react to “persuasive stimuli” and what sort of interventions might make them more amenable to those stimuli. Sentiment analytics and other techniques can help identify these variables. Personality traits, Pentland and colleagues argue, such as the so-called Big Five—whether someone is more or less agreeable, conscientious, extraverted, emotionally stable, and open to experience—can be automatically inferred from Big Data on personal social networks. This information can then be used to design and tune personality-based “change-inducing systems.”3737xJacopo Staiano et al., “Friends Don’t Lie,” in Ubicomp ’12: Proceedings of the 2012 ACM Conference on Ubiquitous Computing (New York, NY: Association for Computing Machinery, 2012): 321–330.
The actual tuning and restructuring involves the manipulation of network “incentives.” What drives the adoption of new behaviors, Pentland argues, is the examples of peers’ behavior. It is the people who have the most invested in a relationship, those who interact and cooperate, who can exert the most social pressure on each other.3838xPentland, Social Physics, 129–30. The incentives, therefore, will often be aimed not at the individuals who are the targets of change but at the individuals in relationship to the targets. Another social pressure strategy is to let peers see how each other is doing—through means such as computer-aided competition, techniques of visualization, and wearable sensors—creating feedback that is often relative to the benchmarks of prodigiously productive employees and “charismatic” networkers. In either case, whether “peer reward” or “peer see,” engineering change involves leveraging preexisting personal ties and enforcing greater social exposure and engagement.
Since the world is dynamic, these manipulations are never one-off interventions. Organizations need to become “living laboratories,” Pentland insists, for continuous data collection, experimentation, and refinement. As workers are increasingly oriented to the data capture and influence machinery, the machinery can be increasingly refined (“trained”) to “see” the on-going and changing activity. With social physics, the more data points, the more precisely tailored the interventions.
Governing and Governable Subjects
The strategy behind all of these influence technologies is to make visible and exploit the power of the social. They partake of the two related aspects of what the legal scholar Julie Cohen calls the “surveillance-innovation complex.” First, the surveillance is cast in “an unambiguously progressive light” and repositioned as a “modality of economic growth.” This innovation rhetoric is presented in the workplace context as a single good with a double meaning. It refers both to enhanced productivity that will flow to the enterprise and personal benefits that will accrue to employees and their teams. Especially in light of the larger context of business competition and worker nonengagement to which it is, in part, a response, this rhetoric has a powerful resonance for companies.3939xCohen adds additional ideological functions to the rhetoric of innovation and participation, including positioning surveillance beyond the reach of legal and social control. Julie E. Cohen, “The Surveillance-Innovation Complex: The Irony of the Participatory Turn,” in The Participatory Condition in the Digital Age, ed. Darin Barney et al., (Minneapolis, MN: University of Minnesota Press, 2016), 207–27. The chance to be a vital part of an innovative, top-performing, and “cool” company, one that promotes camaraderie and values and a larger purpose, certainly has its appeal for workers as well.
Second, the surveillance is characterized as participatory, working with employees’ own desires and motivations to draw them into active engagement. This language too has a double meaning. Participatory refers to the deeper role that workers are called upon to play in the exercise of peer surveillance and peer pressure, and to their own role with the technologies through which they deal with incentives and influence and increase their own self-monitoring. These peer and self-scrutiny dynamics are already present in teamwork, but analytics and automation dramatically extend their range and systematize their impact. All mediated communications are in principle observable, and so every social relation that an employee has on the job, however informal, can be brought into the surveillance and influence nexus. This even includes changes geared to enhancing face-to-face cooperation and interaction, such as shared breaks and the arrangement of physical space. And it includes moods and emotions and the psychological relations of the worker to the group. The imperative to participate in competition and performance tracking creates exposure to continuous peer influence and unobserved manipulation opportunities. One is hard pressed to identify any natural limits, any protected spaces, that are necessarily beyond their reach.
The participatory dynamic is dependent on its seemingly natural fit with features of contemporary life. We are already accustomed to thinking of ourselves and our relations in terms of networks and to living in an everyday habitus that has been shaped in the very ways that the new science exploits. Social networking, for instance, trains us in profiling ourselves according to specified formats, maintaining our relations through techniques of self-reflexive and performative information management, and tracking and providing continuous feedback on others’ online actions—and vice versa. The practices of smart games, self-monitoring, fitness tracking, feedback-based rewards, and data sharing are all familiar. We find much of this pleasurable, and workers themselves may expect and demand that work be organized along these lines. “Our tendency to desire some types of surveillance,” according to John Gilliom and Torin Monahan the authors of SuperVision: An Introduction to the Surveillance Society, “is a fascinating dimension of the surveillance society.”4040xGilliom and Monahan, SuperVision, 9.
This familiarity obscures the nature of the transformation. Incorporating workers into the process of instrumented control creates new values that reorient their relations to coworkers and to their own subjectivity. Participation and incentivization train workers to see others with an ever more rationalized logic of commodification, optimization, and detachment. Unlike face-to-face or even direct written communication, machine-based employee exercises of peer evaluation and pressure are not typically known to the one being subjected to these actions, nor linked back to a fellow employee as the source. Further, since employee actions with regard to peers are seen by managers or team members, there is pressure to align these actions with the organization’s expectations of performance and efficiency, whatever employees’ private feelings or allegiances. The whole point, after all, is leveraging worker solidarities—their trust, loyalty, and vulnerability—for company benefit.
At the same time, participation in surveillance is also directed to reorienting how workers perceive themselves. Seeing and being seen by people whose approval they care about drives them to introject themselves into the relentless feedback-driven processes of self-optimization. Participatory surveillance, cloaked in the language of self-fulfillment and autonomy, induces employees to internalize a transparent mirror, to direct the scientific, managerial gaze back upon themselves, to self-incorporate the external standards, and to propel their own continuous conformity and improvement. To become both governing and governable, and thereby to reduce any distance between the workers’ goals and those of the organization—that is the totalizing promise of social physics in the workplace.