What Is Critical Theory, and Why Do We Need It in the Age of AI?

Artificial intelligence is often introduced as a technical problem. We are told to ask whether the model is accurate, whether the system is safe, whether the interface is useful, and whether the company has followed the right risk-management checklist. These questions matter. They are not enough.

Critical theory begins where technical description becomes socially inadequate. It asks how a technology is made, whose interests it serves, whose labor and data it depends on, what forms of power it extends, what harms it normalizes, and what futures it quietly makes harder to imagine. In the age of AI, those questions are not optional. They are part of the basic literacy required for democratic life.

This post explains what critical theory is, how it differs from ordinary criticism, and why AI makes its methods more urgent.

Critical theory in one sentence #

Critical theory is a tradition of social thought that studies how domination works and how people might become freer from it.

That definition is simple, but it needs care. Critical theory is more than a habit of saying that every new technology is bad. It is a method of connecting ideas, institutions, economic systems, culture, knowledge, and everyday life. It asks why arrangements that harm people can still appear natural, neutral, efficient, or inevitable.

The Stanford Encyclopedia of Philosophy describes Frankfurt School critical theory as a tradition that combines philosophical reflection and social-scientific analysis with an emancipatory aim. It is self-reflexive, interdisciplinary, materialist, and oriented toward freedom from domination. In Max Horkheimer's classic formulation, critical theory differs from "traditional theory" because it does not treat society as a neutral object to be described from the outside. It asks how knowledge itself is shaped by social conditions, and how theory might help transform those conditions rather than merely explain them.

That distinction matters for AI. A traditional technical account might ask whether a model produces a correct answer. A critical account also asks why this model exists, why this task is being automated, who owns the infrastructure, who supplied the training data, who labels or moderates the outputs, who is monitored by the system, who can contest its decisions, and who benefits if the system becomes normal.

The Frankfurt School: critique after catastrophe #

Critical theory is often associated with the Frankfurt School, a group of philosophers and social theorists linked to the Institute for Social Research in Frankfurt, founded in 1923. Its best-known first-generation figures include Max Horkheimer, Theodor W. Adorno, Herbert Marcuse, Walter Benjamin, Erich Fromm, Friedrich Pollock, Leo Löwenthal, Franz Neumann, and Otto Kirchheimer. Later generations include Jürgen Habermas, Axel Honneth, Seyla Benhabib, Nancy Fraser, Rainer Forst, Rahel Jaeggi, and Amy Allen.

The first generation wrote in response to crises that still sound familiar: authoritarianism, mass propaganda, economic instability, antisemitism, racism, bureaucratic control, consumer culture, and the political failures of liberal democracy. They wanted to understand why modern societies could produce scientific progress and administrative efficiency while also producing domination and fascism.

Horkheimer and Adorno's Dialectic of Enlightenment argued that reason can turn against itself when it becomes only instrumental reason: the calculation of means without reflection on ends. Marcuse's One-Dimensional Man analyzed how advanced industrial societies absorb dissent and train people to experience domination as comfort. Habermas later shifted the tradition toward communication, public reason, and the conditions of democratic deliberation. Fraser, Benhabib, Allen, and others pushed the tradition to engage more seriously with feminism, race, colonialism, capitalism, and the limits of Eurocentric narratives of progress.

The tradition is not unified. It contains disagreements about reason, normativity, social movements, capitalism, identity, colonialism, and democracy. That is a strength. Critical theory is less a fixed doctrine than a set of questions about power, knowledge, domination, and emancipation.

What critical theory actually does #

Critical theory is useful because it gives names to patterns that otherwise remain hidden. Four concepts are especially important for AI.

1. Ideology #

Ideology is broader than propaganda. In critical theory, ideology refers to forms of belief, practice, and common sense that make domination appear natural or unavoidable. A social order is most powerful when people stop seeing it as a social order at all.

In AI discourse, ideology appears when prediction is treated as objectivity, automation as progress, scale as public benefit, and efficiency as an unquestioned good. A hiring algorithm may be described as neutral because it uses data. A welfare algorithm may be described as efficient because it reduces administrative cost. A generative AI system may be described as creative because it produces fluent text or images. Critical theory asks what those descriptions hide.

2. Reification #

Reification means treating social relations as things. People, histories, judgments, and conflicts are made to appear as objects, scores, rankings, categories, or data points.

AI systems intensify this problem because they work by translating human life into machine-readable form. Students become risk scores. Workers become productivity traces. Patients become prediction profiles. Writers, artists, and users become training data. The issue is not only that categories can be wrong. It is that the act of categorization can change how institutions see people and how people are allowed to appear before institutions.

3. Instrumental reason #

Instrumental reason asks how to optimize means. It does not ask whether the goal should be pursued. A system can be technically effective and socially destructive.

Critical theory is useful here because much AI governance is tempted to focus on technical mitigation: reduce bias, improve explainability, add a human in the loop, publish a model card. These steps can help. But they do not answer the prior question: should this system exist in this setting at all?

UNESCO's Recommendation on the Ethics of Artificial Intelligence makes a similar point in policy language. It says AI methods should be appropriate and proportional to a legitimate aim, should not violate human rights, and should be assessed across the AI life cycle. It also states that AI should not be used for social scoring or mass surveillance purposes. Critical theory gives a philosophical vocabulary for why proportionality and legitimacy cannot be reduced to engineering performance.

4. Emancipation #

Critical theory is not only diagnosis. Its central question is how people can become less dominated by systems they themselves sustain. Emancipation does not mean that every social problem has a simple solution. It means that people should be able to understand, contest, and reshape the institutions that govern their lives.

This is the point at which critical theory becomes democratic. AI systems are increasingly used in education, employment, policing, welfare, finance, media, health care, and government administration. If people cannot understand or contest those systems, then decisions that affect public life move away from public reason and into proprietary infrastructure.

Why AI makes critical theory urgent #

AI did not create capitalism, racism, patriarchy, coloniality, surveillance, or bureaucratic domination. It can, however, automate and extend them. That is why critical theory matters now.

AI turns social judgment into infrastructure #

The EU AI Act classifies AI systems by risk. It bans certain unacceptable-risk systems and imposes obligations on high-risk systems. The European Commission lists high-risk areas such as critical infrastructure, education, employment, essential services, law enforcement, migration, justice, and democratic processes. These are not marginal use cases. They are domains where people receive rights, opportunities, care, punishment, mobility, credit, and political voice.

When AI enters these domains, it does more than assist decision-making. It can reshape what counts as evidence, which explanations institutions accept, and how people are sorted before they are heard. Critical theory helps us ask whether procedural safeguards are sufficient, or whether some uses turn public judgment into administrative automation.

AI makes power look like prediction #

Prediction carries authority. A predicted risk can become a reason to deny bail, flag a student, reject an applicant, discipline a worker, prioritize a patient, or target a neighborhood. The danger is not only inaccurate prediction. The danger is that prediction can become a way to govern people through probabilities they cannot inspect or contest.

Virginia Eubanks's Automating Inequality examined automated decision systems in social services and welfare administration. Safiya Umoja Noble's Algorithms of Oppression showed how search systems can reproduce racist and sexist hierarchies while presenting results as neutral information retrieval. Ruha Benjamin's Race After Technology argues that discriminatory systems can appear objective or benevolent because they are coded into technical form. These works extend the critical-theory question: how does domination persist when it no longer announces itself as domination?

AI depends on extraction #

AI systems require data, computation, energy, minerals, labor, infrastructure, and capital. Kate Crawford's Atlas of AI frames AI through its material and planetary costs: natural resources, labor, privacy, equality, and freedom. Nick Couldry and Ulises Mejias describe "data colonialism" as a process through which human life is appropriated as data for capitalism. Shoshana Zuboff's theory of surveillance capitalism describes how human experience can become raw material for prediction products.

These arguments differ, but together they challenge the idea that AI is immaterial. AI is not a cloud. It is an industrial system. It draws on mines, data centers, underpaid labeling work, copyrighted and uncopyrighted cultural production, user traces, and concentrated corporate control. Critical theory helps connect the polished interface to the social and material chain that makes it possible.

AI changes the public sphere #

Habermas's work on the public sphere remains useful because democracy depends on more than voting. It depends on shared conditions for public reasoning: access to information, the ability to contest claims, and institutions that can be held accountable.

Generative AI complicates those conditions. It can produce persuasive text at scale, generate synthetic images and audio, automate spam and propaganda, personalize persuasion, and flood information channels with low-cost content. These capacities do not automatically destroy democracy. But they put pressure on the institutions and habits that public reason requires.

UNESCO's AI recommendation explicitly links AI governance to communication and information. It calls for media and information literacy, transparency in automated content generation and moderation, access to diverse viewpoints, and mechanisms for redress. This is critical theory in policy form: the health of a democracy depends on the conditions under which people can know, speak, disagree, and organize.

AI governance needs critique, not only compliance #

NIST's AI Risk Management Framework is a serious governance tool. It is voluntary, use-case agnostic, and designed to help organizations manage risks to individuals, organizations, and society. Its core functions are commonly summarized as govern, map, measure, and manage. NIST has also released a generative AI profile to help organizations identify risks specific to generative AI.

The EU AI Act is more legal and binding. It uses a risk-based approach, including prohibited practices, high-risk obligations, transparency duties, and rules for general-purpose AI models.

These frameworks are important. Critical theory does not replace them. It asks what they may miss.

A compliance framework can ask whether a high-risk system has documentation, logging, human oversight, and risk mitigation. Critical theory asks whether the institution should be using that system, whether affected communities had power in the decision, whether the system shifts responsibility away from accountable humans, whether it deepens existing inequality, and whether its benefits depend on harms displaced elsewhere.

Good governance asks, "Is this system safe and compliant?" Critical theory adds, "Safe for whom, compliant with whose interests, and necessary for what kind of society?"

Critical AI studies: beyond one-off takedowns #

A newer field, critical AI studies, is trying to build methods adequate to contemporary AI. One recent methodological intervention, Fabian Offert and Ranjodh Singh Dhaliwal's "The Method of Critical AI Studies, A Propaedeutic" (2024), argues that criticism of generative AI should not rely too heavily on single examples of bad outputs. Probabilistic systems require methods that can handle variation, interfaces, pipelines, data histories, deployment contexts, and institutional uses.

This is an important correction. It is easy to critique AI by showing a hallucination, a biased image, or an absurd chatbot answer. Those examples can be revealing, but they do not always prove a general claim. A stronger critique asks how the system behaves across cases, how it changes over time, what pipeline produced it, what institution deploys it, and what social function it performs.

That is also the spirit of the academic humanizer discipline used in this post: do not overclaim, do not let rhetoric outrun evidence, and do not replace analysis with slogans. Critical theory is strongest when it is specific.

What critical theory can teach AI builders and users #

Critical theory is often treated as external opposition to technology. That is a mistake. It can improve how AI is designed, governed, adopted, and resisted.

1. Ask the prior question #

Before asking how to build the system, ask whether the task should be automated. Some tasks require care, judgment, trust, responsibility, or democratic legitimacy. Automation may support those tasks, but it can also hollow them out.

2. Follow the whole life cycle #

UNESCO emphasizes the AI system life cycle: research, design, development, deployment, use, maintenance, monitoring, and termination. Critical theory adds that we should also follow labor, data, energy, ownership, and institutional incentives.

3. Treat neutrality claims as hypotheses #

When a system is called neutral, objective, or data-driven, treat that as a claim to be tested. What data was used? Which categories were chosen? Which histories are encoded? Which groups are overexposed to error? Who can appeal?

4. Keep humans accountable #

Human oversight should not become a ritual phrase. A person who rubber-stamps algorithmic output is not meaningful oversight. Accountability requires authority, time, information, contestability, and consequences.

5. Protect the public sphere #

AI literacy cannot mean prompt tips alone. It should include knowledge of data extraction, model limits, synthetic media, recommender systems, labor impacts, privacy, bias, and the political economy of platforms.

6. Include affected communities #

People affected by AI systems should not appear only as data subjects or end users. They should have a role in deciding whether systems are adopted, how harms are measured, and what remedies are available.

Why we need it now #

We need critical theory now because AI is becoming a general layer of social mediation. It is entering search, writing, education, health care, work, government, policing, warfare, culture, and intimate life. It changes not only what institutions can do, but what they can imagine doing.

Without critical theory, AI will be judged mainly by convenience, productivity, market value, and technical performance. Those measures are partial. A society can become more efficient and less free. It can become more automated and less accountable. It can become more personalized and less public. It can generate more content and less understanding.

Critical theory does not tell us to reject every AI system. It tells us to refuse the false choice between technological enthusiasm and technological panic. The real question is political: under what conditions, governed by whom, for whose benefit, with what rights, at what cost, and toward what forms of life?

That question is difficult because AI is not one thing. It is a family of technologies, business models, infrastructures, narratives, and institutional practices. Some uses may help people. Some may be trivial. Some may be dangerous. Some may be unacceptable even if they work.

Critical theory gives us a way to make those distinctions. It asks us to look past the demo and examine the system. It asks us to hear the people who are usually treated as edge cases. It asks us to connect bias to history, automation to labor, prediction to power, and innovation to ownership. It asks whether we are building tools that expand democratic agency or systems that make domination harder to see.

That is why we need critical theory in the age of AI. Not because it has all the answers, but because it keeps the right questions alive.

Sources #

• Stanford Encyclopedia of Philosophy, "Critical Theory (Frankfurt School)," Fall 2025 archive: https://plato.stanford.edu/archives/fall2025/entries/critical-theory/
• UNESCO, "Recommendation on the Ethics of Artificial Intelligence" (2021): https://www.unesco.org/en/legal-affairs/recommendation-ethics-artificial-intelligence?hub=1063
• NIST, "AI Risk Management Framework": https://www.nist.gov/itl/ai-risk-management-framework
• European Commission, "AI Act": https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
• Safiya Umoja Noble, Algorithms of Oppression: How Search Engines Reinforce Racism (NYU Press, 2018): https://nyupress.org/9781479837243/algorithms-of-oppression/
• Ruha Benjamin, Race After Technology (2019): https://www.ruhabenjamin.com/race-after-technology
• Kate Crawford, Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence (Yale University Press, 2021): https://yalebooks.yale.edu/book/9780300252392/atlas-of-ai/
• Virginia Eubanks, Automating Inequality (2018): https://academic.macmillan.com/academictrade/9781250215789/automatinginequality/
• Nick Couldry and Ulises A. Mejias, The Costs of Connection: How Data Is Colonizing Human Life and Appropriating It for Capitalism (Stanford University Press, 2019): https://www.sup.org/books/title/?id=28816
• Shoshana Zuboff, The Age of Surveillance Capitalism (PublicAffairs, 2019): https://www.hachettebookgroup.com/titles/shoshana-zuboff/the-age-of-surveillance-capitalism/9781610395700/?lens=publicaffairs
• Fabian Offert and Ranjodh Singh Dhaliwal, "The Method of Critical AI Studies, A Propaedeutic" (arXiv, 2024): https://arxiv.org/html/2411.18833v1
• Stephen Cave and Kanta Dihal, "The Whiteness of AI" (Philosophy & Technology, 2020): https://link.springer.com/article/10.1007/s13347-020-00415-6