Oppressive A.I.: Feminist Categories to Understand its Political Effects

Oppressive A.I.: Feminist Categories to Understand its Political Effects is an essay by Joana Varon and Paz Peña that proposes a seven-part feminist analytical framework for diagnosing harm in publicly-deployed AI systems. The essay was first published at feministai.net on 26 April 2021 and republished on Coding Rights's Not My A.I. project site on 3 May 2021 (with a 16 November 2021 update), under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 licence. It is the first and load-bearing analytical artefact of the Not My A.I. project — the Coding Rights "feminist toolkit to question A.I. systems" — and the corpus's principal feminist-AI publication-side anchor on the Latin American slate.

Argument and central framing

The essay's central argument is that the dominant debate on automated decision-making — concentrated on technical fairness, bias mitigation, and "ethical AI" principles — leaves the prior political question untouched: whether a given AI system should be deployed at all, on whose request, in whose interest, and at whose cost. Varon and Peña reposition the diagnostic question from how to make AI fair to whether the AI in question automates oppression, and propose that the answer is best read through feminist, decolonial, and antiracist analytical categories grounded in Latin American empirical case material. The framework is offered as a set of empirical feminist categories for naming the power dynamics behind automated decision-making, drawn from case-based analysis of AI projects in the region that pose harm along gender, race, class, sexuality, and territorial axes — moving the framing question prior to deployment rather than after.

The seven categories

The essay enumerates seven analytical categories for reading a deployed or proposed AI system, each named with a short prose definition in the essay text:

1. Surveillance of the poor — "turning poverty and vulnerability into machine-readable data" through state systems that subject low-income populations to data-collection and risk-scoring regimes the rest of society does not face.
2. Embedded racism — AI systems that structurally produce and maintain racial and ethnic exclusion, not as incidental side-effect but as deployment pattern.
3. Patriarchal by Design — systems that reinforce sexism, compulsory heteronormativity, and gender binarism through their classifications, training data, and design choices.
4. Colonial extractivism of data bodies and territories — extraction of personal data and territorial information that, the essay argues, "naturalises the colonial appropriation of life" in extending older extractive patterns into the digital domain.
5. Automation of neoliberal policies — implementation of market-logic welfare reduction and conditionality through automated exclusion mechanisms, mechanising the policy preference for budget contraction.
6. Precarious Labour — the invisible and underpaid manual labour that powers AI systems (data annotation, content moderation, micro-task work) without the labour protections the visible side of the technology industry enjoys.
7. Lack of Transparency — insufficient disclosure of algorithmic decision-making to the affected publics and to oversight bodies, undermining accountability and external review of state-deployed systems.

The framework's substantive move is that the seven categories are not bias subtypes to be technically mitigated within a still-acceptable AI deployment, but political diagnostics that the authors propose surface the prior question of whether a system should exist in its current form at all.

Case material and empirical grounding

The essay is anchored in Latin American case material from the Not My A.I. project's running case-study tracker — twenty-four mapped projects across Argentina, Brazil, Chile, Colombia, and Uruguay as of April 2021, with sustained analytical attention to two: Chile's Sistema Alerta Niñez (the Ministry of Social Development and Family's child-protection predictive-risk algorithm, also the load-bearing case in the parallel Derechos Digitales comparative report) and Argentina and Brazil's Plataforma Tecnológica de Intervención Social (social-intervention predictive system). The shared case material with the Derechos Digitales comparative report is a structural feature of the Latin American feminist-and-digital-rights research field at this moment — the same documented systems carry different analytical readings in the two registers, with Varon and Peña's contribution being the political-power and decolonial-feminist framing rather than the five-dimension human-rights audit methodology Derechos Digitales installs.

Authorship and circulation

Joana Varon co-authors as Founder Executive Directress of Coding Rights, the Rio de Janeiro-headquartered Brazilian feminist think-and-do tank within whose Not My A.I. project the essay is hosted; her 2020-21 Technology and Human Rights Fellowship at Harvard Kennedy School's Carr Center routes the framework into Global North academic-policy circuits through the Carr Center discussion-paper restatement. Paz Peña is a Santiago-based Chilean consultant and researcher on the governance and societal impacts of digital technologies (including a parallel consultancy with the Chilean Cybersecurity Coordination Unit) who anchors the Chilean-case empirical layer. The framework's continuing institutional uptake includes an Internet Policy Review companion article by the same authors on AI and consent (volume 10 issue 4, 7 December 2021, DOI 10.14763/2021.4.1602), the FIRN / APC Not My A.I. final research report, and US-academic teaching circulation through venues such as UC Berkeley's D-Lab.

Posture within the corpus

Within the corpus, Oppressive A.I. is the publication-side anchor of the feminist-AI / decolonial-tech publication sub-type that the corpus's existing Latin American publication anchor — Decisiones automatizadas en la función pública en América Latina, Derechos Digitales's comparative regional report under Jamila Venturini's co-stewardship — does not occupy. The two publications form the corpus's regional pair on Latin American algorithmic-accountability research: the Derechos Digitales report installs a comparative empirical five-dimension human-rights audit methodology under a Spanish-language regional-advocacy register; the Oppressive A.I. essay installs a feminist-decolonial political-power categorisation framework under a transfeminist, decolonial, antiracist register grounded in the same Latin American case base. Both publications anchor on shared case material (Sistema Alerta Niñez carries both); both anchor an ongoing programme line rather than a one-off intervention; and both refuse the framing that Latin America is a region absorbing Global North critique rather than producing its own analytical vocabulary on AI deployment.

Oppressive A.I. is also the publication-side anchor of Coding Rights on the corpus's publications slate — the organisation's Not My A.I. project, the Chupadados investigative-journalism project, the Oracle for Transfeminist Technologies, and the From Bias to Feminist AI peer-reviewed venue are all named in the org body, but this essay is the first Publication entry in the corpus that anchors the Coding Rights publication line. It sits structurally alongside On the Dangers of Stochastic Parrots and Gender Shades as the third of the corpus's foundational academic-or-grey-literature artefacts whose analytical vocabulary the grassroots side of the make-AI-good movement has carried into public-policy organising — with Stochastic Parrots anchoring the LLM-critique vocabulary, Gender Shades anchoring the algorithmic-audit vocabulary, and Oppressive A.I. anchoring the political-power-and-decolonial-feminist vocabulary on the same axis.