Ruha Benjamin

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI argues that these tools will reshape every human institution and that the urgent work is to become worthy of being amplified. Benjamin adds the prior question: amplified for whom, and at whose cost? She is the cycle’s sharpest lens on the fact that the systems are not blank amplifiers but systems with histories embedded in them—histories of which voices were amplified and which were silenced, which knowledge was archived and which was destroyed, which bodies were studied and which were experimented upon. To train a model on the internet is to train it on a monument to inequality.

Her framework reframes every question about fairness and bias in these systems. The dominant instinct of the technology industry is to answer social problems with technical fixes: bias in hiring becomes a hiring algorithm; bias in lending becomes an automated underwriting model. Benjamin’s hardest insight is that the technical fix does not escape the social problem—it buries it deeper, now harder to see and harder to challenge because it wears the authority of code. The discrimination of the algorithmic era hides inside a model that no one fully understands and that everyone is encouraged to trust. Her concept of technological benevolence names the most insidious variant: systems marketed as fixes for inequality that reproduce or deepen it, under the protective cover of good intentions.

The cycle’s central observation—that AI collapses the distance between what a person can imagine and what they can build—takes on a harder meaning through her lens. If the new tools make imagination cheaper to realize, then the question of whose imagination gets amplified becomes more urgent, not less. She presses the observation one step further: because those who monopolize resources monopolize imagination, a technology that empowers everyone to build faster is not the same as a technology that distributes power. Which one we get depends on choices about access, ownership, and control that are being made right now, mostly by the people who already hold both.

Origin

Born in 1978 and raised across South Central Los Angeles, rural South Carolina, the Marshall Islands, and southern Africa, Benjamin grew up watching how science, medicine, and law landed differently on different bodies in different places. She studied sociology at Spelman College and took her doctorate at Berkeley, eventually settling at Princeton as a professor of African American Studies and founding a laboratory devoted to data and justice. In 2024 she was named a MacArthur Fellow. The credentials matter less than the angle of vision they protected. She was trained to ask who benefits, a question the engineering disciplines are not built to ask and the marketing departments are paid to suppress.

Her 2019 book Race After Technology: Abolitionist Tools for the New Jim Code introduced the framework that organized her subsequent work. The title’s deliberate echo of Michelle Alexander’s The New Jim Crow—the argument that mass incarceration reproduced racial caste through ostensibly colorblind means—was exact: Benjamin argued that automated systems do the same thing in technological form. Her 2022 book Viral Justice: How We Grow the World We Want turned from diagnosis toward construction, developing the theory that small deliberate acts, spread virally, can tip large systems toward justice.

The evidence for her central claims arrived in concentrated form during the period of her writing. Facial recognition systems routinely misidentified darker-skinned faces at rates many times higher than lighter ones. Predictive policing systems directed police to communities already over-policed, producing the arrests that “confirmed” the prediction. Medical algorithms trained on historical data systematically underestimated pain in Black patients. Hiring tools trained on past hiring decisions learned the exclusion those decisions encoded. Benjamin’s framework had named the mechanism; the cases provided the instances.

Key Ideas

The New Jim Code. Technologies reflect and reproduce existing inequities while being promoted as more objective or progressive than what they replace. Benjamin organizes the concept into four dimensions: engineered inequity (explicit amplification of hierarchy), default discrimination (harm from neglect of anyone outside the designer’s default imagination), coded exposure (being intensely surveilled but never genuinely seen), and technological benevolence (products marketed as fixes for inequality that deepen it under the cover of good intentions). The framework does not require bad actors. It requires only the ordinary well-intentioned process of building technology for the people in the room, which systematically fails the people who were never invited in.

Race as technology. Benjamin’s most disorienting move is to invert the ordinary relationship between race and technology. She proposes that race is itself a technology—a means to sort, organize, and design a social structure, an invention engineered to allocate power. If race is a sorting technology and algorithmic governance is a sorting technology, they are not two different things that occasionally interact; they are kin, sharing a function. Racialized zip codes, she observes, are the output of Jim Crow policies and the input of New Jim Code practices: the discrimination of a previous era becomes the raw material of the new one, encoded now in a variable the algorithm cannot see and cannot question.

Default discrimination. Harm emerges not from malice but from the accumulated weight of design processes that fail to imagine certain people. The engineer imagines a default user who is almost always someone like the engineer, and everyone who deviates from that default becomes an edge case handled later. The system works fine for the people checking it, because they are the people it was built for. The work of justice is to check it against the people it was not built for, and to count their experience as the real measure. Benjamin’s corrective is to design for the margins: a system robust enough to work for the most vulnerable will be robust for everyone; a system optimized for the default will fail the margins by definition.

Crime production, not crime prediction. Predictive policing systems are trained on historical arrest data, which is not a record of where crime occurs but a record of where police have made arrests. The algorithm concludes that heavily policed communities are high-risk and directs more police to them, generating more arrests, which generate more data confirming the prediction. The feedback loop is self-fulfilling. Benjamin names it a crime production algorithm: it does not discover that a neighborhood is dangerous; it enforces the belief that it is dangerous and collects the evidence that the enforcement produces. The same structure pervades risk assessment tools in bail, sentencing, and parole, where training on prior criminal justice contact amplifies a system’s own historical bias.

Viral justice. The same dynamics that spread injustice—replication, accumulation, compounding of small acts into large patterns—can be enrolled in the service of justice. Benjamin gathers what she calls everyday insurrections and beautiful experiments: local, modest, achievable rearrangements of a community’s relationship to technology. These are not consolation prizes. They are the mechanism by which large change actually moves: the seeds from which larger transformations grow and the proof that a different way of doing things is possible. The metaphor carries a warning: virality is morally neutral, and injustice has historically spread more readily. Justice requires deliberateness and sustained effort to outcompete the default.