The cycle that began with [YOU] on AI insists that the moral character of a technology is revealed not where it is celebrated but where it is endured. Eubanks is the thinker who supplies that principle with documentary proof, going where the industry never looks—to the welfare applicant whose benefits vanished, the family flagged by a score it cannot see, the dying woman whose Medicaid was revoked because the automated system could not distinguish illness from noncompliance. Her lens is the necessary counterweight to any account of AI that begins only with capability. The cycle asks what these systems amplify; Eubanks supplies the evidence from below, the documented record of what amplification feels like when you are the one being amplified against.
Her concept of the digital poorhouse reframes the cycle’s central image. If [YOU] on AI describes these systems as mirrors that reflect and amplify intention, Eubanks specifies which intention is most likely to be reflected when the subject is poor: the intention to manage, to sort, and to contain, routed through historical administrative data that is the sedimented record of every past judgment the institution made. The mirror is honest, which is the problem. It shows us what we have actually done, encoded as what the system will now do at scale. And it does so behind a screen of mathematical objectivity that, as Eubanks documents, converts contestable political decisions into apparently incontestable facts of nature, shutting down the very argument that might change them.
She stands in the cycle’s gallery alongside Kate Crawford, who maps the resource extraction behind AI’s infrastructure, as the thinker who maps its human extraction—the poor and working-class communities who are made to serve as the experimental population for every new system of surveillance and automated judgment. Where Crawford asks what the machines are made of, Eubanks asks what they are aimed at, and her answer—that the most invasive and punitive tools are systematically tested on those least able to refuse them—is the view from the other end of the screen that the industry most needs to hear and most consistently avoids.
Her practical demand is equally precise. The cycle’s vision of alignment between machine behavior and human flourishing requires, in Eubanks’s framework, a political choice that technology alone cannot make: a choice about what we owe one another, and in particular what we owe the poor. Until that choice is made honestly, the most aligned algorithm will still faithfully execute the old answer, which is that we owe them management. Her sentence about eradicating poverty is, in the end, the deepest alignment problem the cycle can name—and the one that no technical fix will reach.
Eubanks arrived at the study of technology through the welfare office rather than the lab, and the direction explains the instrument. She came up through welfare-rights organizing, helping found a grassroots group affiliated with the Poor People’s Economic Human Rights Campaign, and later co-founding the Our Data Bodies Project, a community-based research initiative that works with low-income communities to understand how data is collected and used against them. Her academic appointment at the University at Albany, SUNY, where she joined the political science faculty in 2004, was built on a method the technology industry cannot replicate: extended, trust-built fieldwork inside the offices, apartments, and hearing rooms where poor and working-class Americans collide with the administrative state.
Her 2011 book Digital Dead End examined how poor communities in Troy, New York, used and experienced technology, establishing the empirical and methodological foundation for what would follow. The 2018 book Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor brought three years of reporting across Indiana, Los Angeles, and Allegheny County into a single argument about structural continuity—that the digital systems now administering public assistance descend, through traceable institutional lineage, from the county poorhouse, scientific charity, and the eugenic registry. The book won the Lillian Smith Book Award and the McGannon Center Book Prize and was widely adopted as a teaching text in schools of public policy, law, and computer science.
Her analytic move, which gives her work its unusual force, is to refuse the industry’s preferred framing of novelty. Every account of the AI moment begins from the premise that something unprecedented has happened. Eubanks does not dispute the capability; she disputes the innocence. The capability is new; the purpose it has been bent toward—sorting the deserving from the undeserving poor, managing indigence so that it does not require political address—is old, and unless we examine the old purpose, the new capability will simply execute it faster. This historical grounding is what makes her critique structurally different from most algorithmic-bias discourse, which treats bias as a technical defect a better engineer could fix. Eubanks locates the defect in the decision to build the tool at all, and in the choice of whom to build it against.
The digital poorhouse. The county poorhouse was a nineteenth-century institution of confinement and moral sorting whose function—to draw a hard line between the deserving and undeserving poor, and to make receiving help an experience of shame—was never abolished. It dematerialized, Eubanks argues, into a distributed system of databases, eligibility algorithms, matching engines, and predictive models that performs the same functions at greater speed and scale, and behind a screen of mathematical objectivity the old poorhouse never claimed. A predictive model trained on historical administrative data reconstructs the statistical shadow cast by every past discriminatory decision and calls the reconstruction a prediction. The digital poorhouse is not a metaphor; it is a claim about software architecture.
Neutrality as strategy. Eubanks treats the claim of algorithmic neutrality not as a description of the technology but as a strategy for evading accountability. Every automated decision system encodes value-laden choices—what problem it will solve, whose data it trains on, what counts as a successful outcome, who bears the cost of a false positive. These choices are saturated with the interests of those who make them and hidden once the system runs, which is exactly what produces the appearance of objectivity. The danger is epistemic: when a human caseworker denies your benefits, you can argue; when an algorithm denies them, the denial arrives wrapped in the authority of computation, which forecloses argument by dressing a contestable political judgment as an incontestable fact of arithmetic. Eubanks demands that every automated system be understood as a political artifact that could have been built otherwise.
Surveillance as the price of help. In the domain of poverty, Eubanks documents, consent as ordinarily understood is fictional. The unhoused person who refuses the vulnerability assessment does not preserve their privacy; they forfeit their chance at housing. The benefits applicant who declines to share their data loses the benefits. Help and surveillance arrive as a single package, and there is no option to take the first without the second. The data extracted in these transactions—mental health history, substance use, household composition, contact with public agencies—is then stored in shared databases accessible to law enforcement and other arms of the state, becoming a permanent, cross-referenced record that can be turned against the subjects in contexts they never imagined. The poor live under a regime of total information awareness that the wealthy have largely escaped, not because they have done anything wrong but simply because they have had to ask for help. The training data that powers AI systems is disproportionately drawn from exactly this population, who had no power to withhold it.
Managing the poor instead of ending poverty. The deepest of Eubanks’s claims is that automated social services are not failed attempts to end poverty but successful instruments for managing it—for making poverty administrable, surveillable, and tolerable to a society that has decided not to abolish it. This reframing changes the meaning of every efficiency gain the technology promises: a faster poorhouse is still a poorhouse, and a smarter one is more effective at being what it is. Optimization supercharges purpose without redirecting it. Technology, she observes, became a way of smuggling politics into the system without an actual political conversation—converting a question about justice and resources into a question about scoring and allocation, and allowing the underlying injustice to continue unexamined behind a dashboard showing that the system is running well.
Design justice and the reciprocity test. Eubanks poses a design diagnostic that cuts through technical complexity: if this system were aimed at anyone other than poor and working people, would you be able to build it at all? The vulnerability assessments that condition housing on the disclosure of intimate mental health and behavioral data, the predictive scores that govern whether families keep their children, the denial machinery that reads illness as noncompliance—none of these would survive application to populations with the power to refuse them. Her demand is not merely oversight but the right of refusal: the right of an affected community to decide that a system should not exist at all. This challenges the technological determinism that treats AI deployment as a force of nature and insists instead that technologies are choices, made by people, which can be contested and reversed.