Algorithmic Necropolitics

Definition

Algorithmic Necropolitics: [Adapted] (Mbembe) When algorithmic systems deprioritize the survival, health, or flourishing of certain populations: credit scoring, medical triage, or disaster response.

Foundation

Algorithmic necropolitics extends Achille Mbembe’s concept of necropolitics (the sovereign power to determine who may live or die) into computational systems (Mbembe, “Necropolitics,” 2003). Where Mbembe examined how colonial and postcolonial states exercise lethal power, algorithmic necropolitics describes how automated systems systematically deprioritize the survival, health, and flourishing of certain populations.

The killing is rarely direct. These systems manage life chances computationally, through resource allocation, risk assessment, and priority ranking: who receives medical care, financial resources, disaster relief, or even platform visibility during an emergency. Existing inequalities enter these systems as data and come out amplified, laundered through seemingly neutral mathematics.

The concept captures how algorithms join the machinery Mbembe analyzed as necropower’s instruments: systems that sort populations into the protected and the disposable, determining whose survival is administered and whose is quietly discounted. Unlike human necropolitics, algorithmic versions operate at scale with reduced visibility and accountability, making their lethal logics harder to identify and resist.

Mechanism Analysis

Algorithmic necropolitics operates through several interconnected mechanisms that transform bias into systematic life-and-death consequences.

Training data mechanisms embed historical patterns of neglect and violence into predictive models. When algorithms learn from data reflecting centuries of medical racism, housing discrimination, or unequal disaster response, they systematize and accelerate those patterns. A medical algorithm trained on records in which Black patients’ pain went undertreated (a bias documented directly in clinicians; Hoffman et al., 2016) will continue the pattern with mathematical precision.

Optimization targets create perverse incentives that sacrifice vulnerable populations for aggregate metrics. When hospital systems optimize for throughput rather than equity, algorithms may systematically deprioritize complex cases that disproportionately affect marginalized communities. When disaster response systems optimize for “efficiency,” they may consistently redirect resources away from areas perceived as less valuable.

Resource allocation algorithms function as computational triage systems, making real-time decisions about who receives finite resources. These systems often optimize for factors that correlate with existing privilege (income, zip code, employment status, insurance type), effectively ensuring that those already advantaged receive preference in life-critical situations.

Feedback loop amplification occurs when algorithmic decisions create conditions that justify future discrimination. Predictive policing algorithms that increase surveillance in marginalized communities generate more arrest data, which the system then uses to justify continued over-policing (Lum and Isaac, 2016). This creates an escalating cycle where algorithmic “predictions” become self-fulfilling prophecies.

Opacity mechanisms make these deadly patterns difficult to identify or challenge. When algorithms are proprietary, complex, or constantly updating, affected communities cannot easily demonstrate how computational systems are systematically working against their survival and flourishing.

Case Studies

Healthcare algorithms demonstrate algorithmic necropolitics most clearly. Optum’s widely used algorithm for identifying patients needing additional care systematically underestimated the needs of Black patients, effectively rationing healthcare resources away from those who needed them most (Obermeyer et al., 2019). The algorithm used healthcare spending as a proxy for health needs, but because Black patients historically receive less care due to discrimination, they appeared “healthier” to the system even when experiencing severe illness. The researchers quantified the stakes: correcting the bias would have raised the share of Black patients flagged for additional care from 17.7 percent to 46.5 percent.

The undertreatment of Black patients’ pain is among the best-documented biases in American medicine: in one study, half of white medical trainees endorsed false beliefs about biological differences between Black and white patients, and those who did rated Black patients’ pain as lower and recommended less accurate treatment (Hoffman et al., 2016). Any algorithm trained on the records this bias produced inherits it. A model that learns dosing patterns from decades of discriminatory treatment reproduces the discrimination at scale, encoding medical racism into automated decision-making that hospitals then treat as objective.

Financial algorithms create necropolitical effects through credit scoring and loan approval systems. These algorithms often use zip code, shopping patterns, and social network data as proxies for creditworthiness, factors that correlate strongly with race and class (O’Neil, 2016). When these systems deny mortgages, business loans, or even basic banking services to entire communities, they perpetuate cycles of disinvestment that directly impact health outcomes, life expectancy, and survival.

Disaster response systems revealed their necropolitical logic after Hurricane Katrina. Louisiana’s Road Home program calculated rebuilding grants from appraised home values rather than actual rebuilding costs; after decades of redlining had depressed appraisals in Black neighborhoods, the formula systematically shortchanged Black homeowners facing identical repair bills. Civil rights organizations sued under the Fair Housing Act, and HUD ultimately settled for $62 million with the affected homeowners, the formula amended (NAACP Legal Defense Fund, Greater New Orleans Fair Housing Action Center v. HUD; HUD settlement announcement, 2011). The pattern persists across disasters: researchers tracking households through major disasters found that white residents of heavily damaged counties tend to gain wealth in the years that follow, while Black residents of the same counties lose it (Howell and Elliott, 2019). Recovery itself is distributed along racial lines, and the formulas doing the distributing wear the costume of actuarial neutrality.

Content moderation algorithms exhibit necropolitical patterns by suppressing information vital to marginalized communities’ survival. During the COVID-19 pandemic, algorithms designed to combat “health misinformation” sometimes flagged legitimate community health resources, harm reduction information, and organizing efforts by marginalized groups as suspicious content, effectively silencing voices trying to protect their communities (Baker et al., 2020). The pattern extends beyond the pandemic: platforms have repeatedly removed harm reduction content outright, treating information that keeps drug users alive as promotion of drug use (Harm Reduction Journal, 2024).

Systemic Context

Algorithmic necropolitics operates within broader systems of racialized capitalism that treat certain populations as surplus, disposable, or less deserving of resources and care. The inequality precedes the algorithm; the algorithm industrializes it, providing technological cover for discriminatory outcomes along the way.

The profit motive incentivizes these patterns by rewarding efficiency over equity. Healthcare systems that optimize for profit rather than care will naturally develop algorithms that favor profitable patients over expensive ones. Insurance companies that maximize shareholder value will create algorithms that identify and exclude high-cost populations. Platform companies that optimize for engagement will design systems that amplify profitable content while suppressing voices that threaten advertiser comfort.

Legal frameworks enable algorithmic necropolitics by treating discriminatory algorithms as neutral tools rather than discriminatory practices. Current anti-discrimination law struggles to address algorithmic bias, especially when algorithms don’t explicitly use protected characteristics but achieve discriminatory outcomes through proxy variables. This legal lag creates space for necropolitical algorithms to operate with minimal oversight or accountability.

The infrastructure of algorithmic necropolitics is supported by the broader digital economy’s extraction of value from marginalized communities. Data harvesting, behavioral surplus extraction, and platform capitalism all depend on treating certain populations as sources of data and profit rather than stakeholders deserving protection and care.

Resistance & Mitigation

Community organizing represents the most powerful resistance to algorithmic necropolitics. Groups like the Algorithmic Justice League, Data for Black Lives, and local community organizations have successfully challenged discriminatory algorithms through research, advocacy, and direct action. These efforts work by making visible the hidden operations of necropolitical systems and demanding accountability from the institutions that deploy them.

Technical interventions include algorithmic auditing, bias testing, and the development of alternative algorithms designed with equity as a primary goal. However, these approaches often fail to address the underlying structural issues that create necropolitical outcomes. Technical fixes without systemic change tend to optimize discrimination rather than eliminate it.

Regulatory approaches show promise when they focus on outcomes rather than intentions. Laws that require algorithmic impact assessments, mandate equitable outcomes, or create private rights of action for algorithmic discrimination can create meaningful accountability. However, regulation often lags behind technological deployment, leaving communities vulnerable during critical periods.

Policy interventions must address both algorithmic systems and the broader structures they serve. Auditing a healthcare algorithm accomplishes little if the healthcare system it serves still prioritizes profit over care; fixing credit scoring accomplishes little while the structures of economic inequality it measures remain standing.

The Optum case shows what resistance can win: researchers found the bias, named it publicly, and demonstrated that reformulating the algorithm’s target eliminated it (Obermeyer et al., 2019). Every algorithm that quietly distributes life chances was built by people making choices, and choices can be contested. The work of resistance is forcing each system to show whose lives it weighs, and at what discount.

Annotated Bibliography

Baker, Stephanie Alice, Matthew Wade, and Michael James Walsh. “The challenges of responding to misinformation during a pandemic: content moderation and the limitations of the concept of harm.” Media International Australia (2020).
Analysis of pandemic-era content moderation and the brittleness of harm-based removal at scale. Context for how “misinformation” enforcement catches legitimate health content.

Hoffman, Kelly M., Sophie Trawalter, Jordan R. Axt, and M. Norman Oliver. “Racial bias in pain assessment and treatment recommendations, and false beliefs about biological differences between blacks and whites.” PNAS 113, no. 16 (2016). https://www.pnas.org/doi/abs/10.1073/pnas.1516047113
Documents the clinical pain-treatment bias that medical algorithms inherit through their training data: half of white medical trainees endorsed false biological beliefs, and those who did rated Black patients’ pain as lower.

HUD. “HUD and Louisiana Announce Settlement Agreement to End Legal Challenge to Road Home Program” (July 2011). https://archives.hud.gov/news/2011/pr11-138.cfm
The Road Home record: the appraised-value grant formula challenged under the Fair Housing Act and settled for $62 million, with the formula amended. The adjudicated case behind this entry’s disaster-response claim.

Howell, Junia, and James R. Elliott. “Damages Done: The Longitudinal Impacts of Natural Hazards on Wealth Inequality in the United States.” Social Problems (2019).
Longitudinal evidence that white residents of disaster-damaged counties tend to gain wealth after disasters while Black residents lose it. Recovery as racialized redistribution.

Lum, Kristian, and William Isaac. “To predict and serve?” Significance 13, no. 5 (2016). https://rss.onlinelibrary.wiley.com/doi/full/10.1111/j.1740-9713.2016.00960.x
The empirical demonstration of predictive policing feedback loops: biased arrest data sends police back to the same neighborhoods, which generates more arrest data, which tightens the loop.

Mbembe, Achille. “Necropolitics.” Public Culture 15, no. 1 (2003); expanded as Necropolitics (Duke University Press, 2019).
The source concept this entry adapts: sovereignty as the power to dictate who may live and who must die. Essential for understanding what is being claimed when that power migrates into computational systems.

Obermeyer, Ziad, Brian Powers, Christine Vogeli, and Sendhil Mullainathan. “Dissecting racial bias in an algorithm used to manage the health of populations.” Science 366 (2019): 447-453. https://www.science.org/doi/10.1126/science.aax2342
The Optum case. Health costs used as a proxy for health needs systematically underestimated Black patients’ illness; correcting the bias would have raised the share of Black patients flagged for additional care from 17.7 to 46.5 percent.

O’Neil, Cathy. Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy (2016).
Catalog of proxy-variable discrimination across credit, insurance, education, and employment scoring. Source for how zip codes and behavioral data smuggle race and class into “neutral” models.

“Problematizing content moderation by social media platforms and its impact on digital harm reduction.” Harm Reduction Journal (2024). https://pmc.ncbi.nlm.nih.gov/articles/PMC11549828/
Documents platforms’ repeated removal of harm reduction content, treated as promotion of drug use. Source for the suppression of survival information.

Dictionary of Digital Oppression, version 0.2.