Automating Inequality presents a vivid if deeply disturbing account of the digitization of social services in the United States. At its heart, this book is about the perils of automating distributional decision making in a society that has long regarded the poor with contempt. Drawing on her background in Science and Technology Studies, Eubanks argues that computer systems designed to combat poverty are actually making things worse. Scholars of poverty and inequality may find some of Eubanks’ conclusions unsurprising, particularly the idea that digitized forms of poverty management reproduce and exacerbate inequality. But Eubanks’s primary objective is not to revise social scientific understandings of poverty and inequality. Rather, writing for a popular audience, her primary goal is to temper technological optimism among policy makers. Further, by showing that digital tools are more expensive, less effective, more biased and more conducive to surveillance than people think, Eubanks endeavors to stir empathy and anxiety among the professional classes in the hope of building solidarity and overriding America’s antipathy towards the poor. In the introduction, Eubanks quotes a young mother on welfare who after describing how EBT (electronic benefit transfer) cards were being used as tracking devices pointedly tells Eubanks, “[y]ou should pay attention to what happens to us. You’re next” [9]. Later, Eubanks writes, “[t]he most important step in dismantling the digital poorhouse is changing how we think, talk, and feel about poverty” [205]. Ultimately, this book is most effective at showing that we cannot understand how government programs reproduce inequality in the 21^st century if we overlook automation.

Eubanks rests her argument on three revealing case studies that vary on many dimensions: geography, policy programs, technology, and even the intentions of the purveyors of these technologies. The first case (Chapter 2) examines a botched attempt to automate eligibility for Medicaid in Indiana, where rigid online applications substituted face-to-face interactions, and privatized call centers further cut ties between caseworkers and clients (punishing). The second case (Chapter 3) describes a system of coordinated entry—“the Match.com of homeless services”—that determines which members of LA’s unhoused population will access public housing. This system has actually improved housing access for LA’s most vulnerable homeless, but at great cost through the collection and storage of inordinate amounts of personal data, which have become accessible to law enforcement agencies without proper legal authorization (policing). The final case (Chapter 4) moves to Pennsylvania to describe the implementation of a well-intentioned but biased algorithm that uses only information about families who access public services to predict a child’s risk of future abuse or neglect. Similar data about families that do not use public services, such as information about parental substance abuse, is simply unavailable. Through this lopsided data collection, abuse and neglect becomes increasingly associated with poverty in the eyes of the algorithm (profiling).

Qualitative sociologists, who often struggle to balance the demands of verification with the need to protect the identity of research subjects, will find much of interest in Eubank’s journalistic approach: she used the real names of people she interviewed and hired a fact checker. Consequently, she gave her research subjects more of a say over which of their experiences she discussed. This book will inspire spirited conversation about the distinctions between ethnography and journalism, and gestures at the possibility of fruitful methodological cross pollination. A key strength of this book is the empirical diversity of its case studies and their evocative depictions, which foreground the experiences of low income Americans across the color line while avoiding stereotypes. With that said, these cases are not comparative: they are not designed to gain analytical leverage over a particular variable of interest. Rather they document a progression in technological sophistication and in moral complexity. In doing so, they help Eubanks to build a broad case for the idea that these technologies are all part of the same phenomenon—the digitization of poverty management.

To showcase the pernicious effects of automation in social provision, Eubanks adopts the provocative metaphor of the “digital poorhouse.” Metaphors are a framing device that social scientists often use as an instrument of clarification, or to focus attention on unobserved dynamics.^{Footnote 2} The relevant questions are 1) What work does this metaphor do? What does it help us to see and what does it obscure? Does it help to clarify the effects of automation in social provision? And 2) can this metaphor help us to establish a more general framework about the perpetuation of poverty and inequality in the 21^st century? In the paragraphs that follow, I attempt to address these questions.

One of the things that does make this metaphor effective is the way it situates automated decision making in historical context to show that the digital revolution in poverty management is a new manifestation of an old phenomenon. Chapter 1 presents a vivid history of the poorhouse, which originated in England in the 17th century and travelled to America with British colonists. This local government institution was supposed to administer relief. In doing so, however, its overseers drew moral distinctions between the deserving and undeserving poor, housing the “undeserving,” able-bodied poor in horrendous conditions, exploiting, humiliating and punishing this population in order to deter people from seeking relief. Eubanks writes, “Like the brick and mortar poorhouse the digital poorhouse diverts the poor from public resources. Like scientific charity, it investigates, classifies and criminalizes. Like the tools birthed during the backlash against welfare rights it uses integrated databases to target track and punish” [Eubanks 2018: 178]. Thus, Eubanks argues that these high-tech tools constitute a new manifestation of the recurring tendency of governments to deter, police and punish the poor.

These observations might lead one to wonder what’s new about the digital poorhouse? We are in the midst of a digital revolution so would we not expect all the inequities that characterized older bureaucratic forms to recur in their new digital formations? Eubanks seeks to show that these tools do not just reproduce pre-existing inequities. Rather, they make them worse.

Part of what is new about the digital poorhouse is its heightened capacity for surveillance. Eubanks entreats the reader to think about the digital poorhouse as a web of surveillance that affects all Americans, but has especially perilous effects on poor people and especially poor people of color. Eubanks has stripped out most references to social theory, but traces of Weber, Latour, and Foucault are prevalent throughout the text. There is a distinctly Foucaultian quality in the described shift from a brick and mortar poorhouse to an invisible, digital poorhouse. Eubanks describes the digital poorhouse as an “invisible digital prison for poor and working people.” And yet, whereas Foucault’s^{Footnote 3} panopticon exacts psychological rather than physical punishment, Eubanks demonstrates that today’s digital poorhouse—while less visible—has very physical effects, often depriving people of essential medical resources, and heightening the risks of incarceration and family separation.

Another new quality of the digital poorhouse seems to be the diminishing of human connection and human discretion over distributional decisions. The loss of human connection is particularly evident in Chapter 2 which describes how Indiana adopted an automated eligibility system in which beneficiaries had to apply for relief online. Case workers were replaced by untrained privatized call takers, and cases were split into discrete tasks carried out by different people. We also see a diminishing of discretion in Chapter 4, which describes how budget cuts in 2012 led the Allegheny County child welfare office to search for an “automated triage system” that would help the office determine how to allocate its dwindling resources. The contract went to an academic team that proposed to design a tool that would assess a child’s risk of future abuse and neglect. The resultant algorithm does not eliminate human discretion, but it does appear to shift discretion from the street level bureaucrat to the more remote engineer of the algorithm. This might not be such a problem if the algorithm was accurate, but its reliance on data collected from people who receive public assistance introduces bias into the system. The model’s accuracy is far from perfect. The algorithm was designed to support the intake screeners, but Eubanks concludes that “in practice the algorithm seems to be training the intake workers” [142]. It is true of course that people are also biased and imperfect in their decision making. As one interviewee told Eubanks, “[t]hey come with their own opinions but sometimes you can change their opinion… You can’t fix that number” [167].

Rather than just pointing out new features on an old form, Eubanks argues that the new form actually exacerbates inequality more than the analog poorhouse did. This argument is most compelling in Chapter 2, which shows quite clearly that the state governor was motivated to slash welfare rolls, and used electronic systems to increase barriers to relief. This was most glaring in the adoption of an online application form which created difficulties for people without computers or reliable internet access. But it was also apparent in claimants’ stupefying efforts to contend with an exasperating electronic system.

In Chapter 3, Eubanks shows how digital systems exacerbate inequality by creating new ways for law enforcement agencies to access information about the homeless without proper legal oversight. The story of LA’s coordinated entry program is more complex than that of Indiana’s automated eligibility initiative, because the author describes a system that does succeed in housing some of LA’s homeless population. And yet, in what is one of the most fascinating parts of this book, the author describes how LA’s matching system creates a new classification that draws more granular distinctions between more and less deserving segments of the unhoused population. By housing some people the system legitimates itself, but the unfettered collection and storage of data about LA’s unhoused citizens makes it easier for the police to access personal information, track and surveil this population without a warrant—infringing on due process rights with a newfound facility.

Like any thought-provoking piece of scholarship, Eubanks’ account churns up new questions: What are the mechanisms by which social values (such as contempt for the poor) creep into technical systems? Is the problem the technology itself or the privatization and outsourcing of distributional decisions in ways that create greater distance between clients and decision makers? Are these technologies being used to administer public assistance in other welfare states, and to what effect? Eubanks shows that systems and algorithms are shaped by human prejudice—to what extent do these systems also shape attitudes?

In summary, the metaphor of the digital poorhouse frames digitization as the newest iteration of an old system that punishes the poor. And it seeks to draw attention to the devastating effects (including everything from cutting off access to life-saving medication because of a minor error on a complicated set of forms to removing a child from its home for the wrong reasons) these technologies can and often do have, even when adopted with good intentions. What Eubanks really wants is to change how Americans “think, talk, and feel about poverty.” She certainly has made a compelling case that scholars of the welfare state cannot overlook the digital tools of administration. Future accounts of social policy development would be remiss not to take this lesson to heart.