Benjamin is a Princeton sociologist who writes about race, science, and technology without treating any of those as separate lanes. Race After Technology coined a phrase that stuck for a reason: the New Jim Code. The claim is that discriminatory designs get a second life as technical progress. Apps, risk scores, and “smart” systems can reproduce racial hierarchy while sounding fairer than the human processes they replace, because code wraps old exclusions in the authority of math.
What makes the book useful beyond rhetoric is how it treats design choices as political instruments. Default settings, training data, visibility regimes, and “equity” patches are all sites where race gets operationalized. Benjamin does not argue that robots have inner racist feelings. She argues that racism does not require intent when it is distributed through legal codes, medical protocols, financial practices, and machine learning pipelines.
Table of contents
- Preface
- Introduction: The New Jim Code
- Engineered Inequity: Are Robots Racist?
- Default Discrimination: Is the Glitch Systemic?
- Coded Exposure: Is Visibility a Trap?
- Technological Benevolence: Do Fixes Fix Us?
- Retooling Solidarity, Reimagining Justice
Chapter by chapter
Introduction: The New Jim Code
Jim Crow was an architecture of racial control enforced through law, custom, and violence. The New Jim Code, in Benjamin’s framing, is its digitized cousin: technologies that amplify social hierarchy while claiming objectivity, efficiency, or even anti-bias virtue. The introduction’s job is to make that continuity visible without collapsing every app into the same story.
She previews four patterns that organize the book: engineered inequity (systems that encode hierarchy by design or training), default discrimination (harms treated as glitches when they are systemic), coded exposure (the ambivalent politics of being seen by a system), and technological benevolence (fixes that leave underlying power intact). The tone is diagnostic and pedagogical. Benjamin wants readers who build, buy, or regulate tech to stop treating “neutrality” as a default property of software.
1. Engineered Inequity: Are Robots Racist?
This chapter attacks the idea that racism lives only in individual hearts. If non-living systems (laws, credit practices, medical protocols, zoning rules) can already encode racial hierarchy, then robots and classifiers can too. Racism here is institutional and technical, not merely attitudinal. A system can discriminate without a bigot in the loop.
Benjamin walks through examples where race becomes an operational variable even when race is formally omitted: proxies in data, culturally loaded training sets, design assumptions about “normal” users. Discrimination gets displaced into the pipeline, and accountability gets outsourced to “the algorithm did it.” She also notes how anxiety about robots often centers white futures (“what if we are next”) while ignoring people already living under automated control. For engineers, the chapter reframes racism as a systems property you can design for or against, not only a HR training topic.
2. Default Discrimination: Is the Glitch Systemic?
The glitch story is comforting: a racist outcome appears, someone apologizes, a patch ships, the brand recovers. Benjamin argues that many harms are features of systems trained on unequal social data and deployed into unequal institutions. Predictive policing and recidivism risk scores are central cases. Tools trained on historically biased arrest and charging data send more scrutiny to the same neighborhoods, producing more data that “confirms” the prediction.
She distinguishes obvious discrimination from insidious default settings that look race-neutral while reproducing hierarchy. “Fixing” a score without changing the institution that uses it can deepen the same pattern, because the institution still needs a ranked population to manage. The chapter also notes collateral damage: tools built to target one group can harm adjacent populations once the logic of scoring spreads. This is one of the clearest popular accounts of feedback loops in carceral tech.
3. Coded Exposure: Is Visibility a Trap?
Coded Exposure asks who benefits from being seen by a system. Facial recognition that “works better” across more faces is often sold as an equity win. Benjamin presses on the ambivalence: visibility can mean recognition and service for some people, and capture, policing, or exclusion for others. Under an unjust surveillance regime, higher accuracy can mean more efficient harm.
She engages debates around race, phenotype, and machine vision, including the shaky leap from genetic ancestry talk to appearance-based classification. The chapter’s technical-political point is that evaluation metrics (accuracy, equalized error rates) cannot answer the prior question of whether the system should exist in that context. For privacy and civil-rights work, this chapter is the antidote to “just make the model fairer” as a complete response to face surveillance.
4. Technological Benevolence: Do Fixes Fix Us?
Here Benjamin looks at well-meaning interventions: diversity dashboards, bias audits sold as absolution, apps that promise to close gaps while expanding the same institutional reach. Technological benevolence is the move where inclusion into an extractive system is marketed as justice. Representation in datasets or on design teams can matter, and still leave the purpose of the tool untouched.
The chapter is skeptical of reform that never asks who the tool serves and what power it concentrates. “Do fixes fix us?” is a real question about whether anti-bias patches become a substitute for demilitarizing or defunding the underlying practice. Readers coming from industry ethics teams may find this chapter uncomfortable. That discomfort is the point. If your remediation plan never contemplates not deploying, you are optimizing legitimacy, not justice.
5. Retooling Solidarity, Reimagining Justice
The final chapter turns from diagnosis toward abolitionist imagination. Benjamin wants design practices accountable to communities historically harmed by tech, not only audited for disparate impact after deployment. Solidarity becomes a design requirement: who is in the room before the pipeline is built, who can refuse, who shares in the value, who can shut a system down.
She is shorter on blueprints than on orientation. There is no 12-step procurement checklist. There is a demand to reimagine what counts as innovation when the metric is collective freedom rather than engagement or clearance rates. Pair this chapter with concrete campaigns (face recognition bans, worker data rights, procurement fights) and it becomes a north star rather than a floating ethic.
Where I’d push back
If you want a systems-engineering handbook for fairness metrics, this is the wrong book. Benjamin deliberately keeps the lens sociological. That is a strength for naming power, and a limit for readers who need concrete model specs, evaluation harnesses, or statutory text. Some case material has also aged into a longer literature (COMPAS debates, face recognition bans, generative-model harms) that students will want to update with later empirical work.
Read beside Weapons of Math Destruction and you see a shared target (opaque scoring systems) with different primary optics: O’Neil writes as a former quant mapping damage across markets; Benjamin writes as a race scholar mapping how “neutral” tech launders hierarchy. Next to Crawford’s Atlas of AI, Benjamin is less about mines and more about the social codes that ride on top of the stack.
For organizers, policy people, and engineers who keep hearing “the model is biased” and need a clearer account of how bias becomes infrastructure, this is still one of the cleanest entry points. The New Jim Code framing travels into today’s generative systems even though the book was written before that wave.