REVIEW 3 major objections 5 minor 2 references
AI Literacy: An Exercise in Power-Knowledge
T0 review · 3 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash
Pith's one-line read This paper claims that dominant AI literacy frameworks train competent consumers of AI rather than epistemic agents, and it proposes a three-part framework—contextual use, critical interrogation, participatory governance—to change that.
desk verdict A coherent normative case for power-aware AI literacy, but the empirical critique of existing frameworks rests on a small sample and quotes that partly undercut it. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The key mechanism is the distinction between two models of literacy. The 'competency model' frames the AI-literate person as a consumer who evaluates outputs and uses tools responsibly. Against it, the paper sets a model of epistemic agency built on three dimensions: Contextual Use (using AI reflectively for one's own epistemic purposes, including generative prompting), Critical Interrogation (reading AI systems as knowledge apparatuses through genealogical analysis, bias recognition, epistemic mapping, and counter-prompting), and Participatory Governance (understanding and joining the democratic governance of AI as a sociotechnical system). The three dimensions are mutually reinforcing, not
What would settle it
Take the current set of national, state, and industry AI literacy standards and code them for whether any learning outcome requires students to explain who shapes training data, whose knowledge traditions are underrepresented, or how citizens can participate in AI regulation. If such outcomes are common rather than rare, the paper's claim that existing frameworks systematically omit power would be contradicted.
Extended reading notes
Core claim
The paper's central claim is that AI literacy frameworks organized around technical competency and responsible use systematically omit power as a construct, and this omission is not incidental but structural. Generative AI systems are argued to be 'power-knowledge apparatuses': they produce probabilistically structured accounts of what is knowable, encoding the institutional and cultural hierarchies of their training data. A literate user, therefore, is not simply an efficient and responsible consumer of outputs but an agent who can interrogate the assumptions built into the system, recognize whose knowledge traditions are centered and whose are marginalized, and participate in governing the
Load-bearing premise
The claim rests on the assumption that the four frameworks the paper examines—a U.S. labor department framework, a university framework, a UN agency framework, and a joint economic-commission framework—fairly represent existing AI literacy frameworks and that none of them substantively engages training-data politics, governance participation, or the structural conditions of AI knowledge production.
Editorial extensions
If this is right
- AI literacy curricula that teach only prompt skills, hallucination detection, and responsible use are incomplete; they need to include training-data politics, the political economy of AI, and governance participation.
- Assessments of AI literacy should measure not just output evaluation but the capacity to interrogate a system's framing and to seek alternative framings.
- Adopting the framework would add genealogical analysis, structural bias recognition, epistemic mapping, and counter-prompting to the standard competency curriculum.
- The agent/consumer distinction implies that a technically sophisticated user who only uses AI as a productivity tool is not fully AI-literate in the paper's sense, while a modest-skills user who interrogates assumptions can be.
- Participatory governance mechanisms—public consultations, community data sovereignty, civic advocacy—belong inside literacy education, not outside it.
Reading between the lines
- An implication the paper leaves implicit: if this critique generalizes, competency-based literacy programs may function as a form of adaptation to AI systems, reproducing the 'banking model' the paper criticizes; this could be tested by analyzing a broader sample of AI literacy curricula for any treatment of power.
- The framework could be operationalized: a curriculum that adds counter-prompting exercises and governance participation could be compared against a competency-only curriculum on measures of students' ability to identify absent perspectives and to articulate governance preferences.
- The argument implies that the democratizing potential of AI depends on literacy in the paper's sense; without it, AI access may widen rather than close epistemic gaps—a claim testable with longitudinal data on how differently educated users use AI tools.
- The framework connects naturally to neighboring work on critical data studies and decolonial approaches, but the paper leaves that connection implicit.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper advances a normative argument that existing AI literacy frameworks are dominated by a technical-competency model that positions learners as consumers of AI outputs, and that this orientation is insufficient because it omits questions of power. Drawing on Foucault's power-knowledge, Freire's critical pedagogy, and critical digital literacy scholarship, the authors propose a reconceptualization of AI literacy as epistemic agency, organized around three dimensions: Contextual Use, Critical Interrogation, and Participatory Governance. The paper further argues that unequal access to AI tools reproduces epistemic hierarchy and that the omission of power analysis from literacy curricula reflects institutional interests rather than oversight.
Significance. If accepted, the paper offers a coherent and politically engaged corrective to mainstream AI literacy discourse, synthesizing established critical theory into a usable three-part framework. The writing is clear, the normative proposal is internally consistent, and the authors fairly acknowledge AI's democratizing potential. The framework's dimensions—especially Critical Interrogation and Participatory Governance—could inform future curriculum design. However, the paper's motivating diagnosis depends on a blanket empirical claim about existing frameworks that is not adequately supported. The theoretical contribution is valuable, but the empirical premise needs strengthening before the central claim can be considered established.
major comments (3)
- [AI Literacy as Empowerment, Not Competency] The central empirical premise—that 'across all four' frameworks there is 'no substantive treatment of training data politics, governance participation, or the structural conditions of AI knowledge production'—is not supported by the evidence presented. The paper's own descriptions undercut this claim: UNESCO frames students as 'co-creators,' Penn State incorporates 'explicit attention to equity, labor, and democracy,' and the EC/OECD framework calls for learners to be 'ethical stewards.' These features are dismissed as converging on a 'careful and responsible user,' but no direct quotations or content analysis are provided to show that power and governance are substantively absent. Without such evidence, the blanket characterization is not established; at most the paper can claim a tendency or an insufficiency of depth. This is load-bearing because the entire motivation for the proposed
- [AI Literacy as Empowerment, Not Competency] The assertion that the omission of power analysis 'reflects the interests of the institutions that produce AI literacy curricula' and is 'not accidental' is an explanatory claim without supporting evidence. It is presented as if self-evident, but it is a strong causal attribution about institutional design. The paper should either provide evidence (e.g., analysis of institutional mandates, funding, or governance structures) or explicitly frame this as a hypothesis to be tested. As written, this assertion supports the concluding charge of complicity and needs a more careful epistemic status.
- [AI Literacy as Empowerment, Not Competency / Competency Creates Consumers; Literacy Creates Agents] The dichotomy between 'competency' and 'empowerment' is drawn too sharply. The paper cites Long and Magerko (2020) as a competency framework, yet that framework explicitly includes critical thinking among its competencies. The paper does not show why such critical-thinking competencies are insufficient for epistemic agency, nor why 'output evaluation' cannot be a form of critical interrogation. The claim that competency frameworks are 'not designed to ask' governance questions is an assertion about designer intention. To justify the reconceptualization, the paper needs to specify what concrete epistemic practices are excluded by existing competency frameworks, rather than relying on a binary that many frameworks may already partially cross.
minor comments (5)
- [Abstract] Grammatical issue: 'competent consumer of AI-generated information' should be 'competent consumers' to agree with the plural subject.
- [Figure 1] Figure 1 is referenced in the text but not shown in the provided manuscript; ensure the figure is included and legible, and that its relationship to the three dimensions is explicitly explained.
- [A Three-Part Framework] The paper would benefit from a more explicit comparison with prior critical AI literacy frameworks (e.g., Stamboliev 2023; Velander et al. 2024; DiPaola et al. 2024) to clarify what the proposed three dimensions add beyond existing 'civic literacy' or 'postcritical' approaches. Currently the novelty is implicit rather than demonstrated.
- [AI Literacy as Empowerment, Not Competency] The phrase 'the frameworks differ considerably in their ambition' is followed immediately by 'all four frameworks converge.' Clarify that the difference is in rhetorical ambition, not in fundamental orientation, and support this distinction with evidence from the frameworks.
- [Conclusion] The claim that AI literacy discourse that 'evades politics' is 'not neutral but complicit' is a strong normative conclusion. It would benefit from a brief qualification acknowledging that non-critical frameworks may still have value in certain contexts, which would make the argument more persuasive.
Circularity Check
No substantive circularity: the paper is a normative/conceptual argument, not a derived prediction. A minor self-citation to Lund et al. (2023) is not load-bearing.
full rationale
This is a conceptual paper with no equations, fitted parameters, or quantitative predictions, so the derivational circularity failure modes largely do not apply. The central claim—that existing AI-literacy frameworks emphasize technical competency and responsible use and omit a robust account of power—is an interpretive assessment of four selected frameworks, introduced explicitly as illustrative: 'We illustrate this pattern through four publicly available AI literacy frameworks...' The audit criteria overlap with the paper's own proposed dimensions (training-data politics, governance participation, structural conditions), but the paper does not present the audit as a formal derivation; it offers a normative reconceptualization grounded in Foucault and Freire and explicitly disclaims completeness ('It is not a complete curriculum or a policy prescription'). The only self-referential support is two citations to Lund et al. 2023 in the 'Democratization and Its Contradictions' section, used to support the claim that AI amplifies existing capabilities. These citations are not load-bearing: the central argument rests on the Foucault/Freire framework and the illustrative framework audit, not on those citations. The paper's weakest points are evidentiary rather than circular: the representativeness of the four-framework sample is asserted, and the claim that omission 'reflects the interests of the institutions' is under-supported. Those are correctness/evidence concerns, not self-referential derivation.
Assumptions & free parameters
assumptions (5)
- domain assumption Foucault's power-knowledge: knowledge systems are constitutively entangled with power and shape what can be said, by whom, and with what authority.
- domain assumption Freire's banking model of education transfers to AI literacy: skills-based teaching reproduces passivity and adaptation.
- domain assumption All knowledge systems are selective, and selectivity is always a form of power.
- ad hoc to paper The four selected frameworks (DOL, Penn State, UNESCO, EC/OECD) are representative of the population of AI literacy frameworks.
- ad hoc to paper Institutional interests of curriculum-producing bodies explain the omission of power-analysis from AI literacy frameworks.
invented entities (3)
-
Epistemic agent (vs. competent consumer)
-
Generative prompting
-
Counter-prompting
Cite this review
Pith. "Pith review of AI Literacy: An Exercise in Power-Knowledge." pith.science (2026). https://pith.science/paper/HEJKRLA6
@misc{pith2026260727547,
author = {Pith},
title = {Pith review of: AI Literacy: An Exercise in Power-Knowledge},
year = {2026},
howpublished = {\url{https://pith.science/paper/HEJKRLA6}},
note = {Machine review of arXiv:2607.27547}
}
read the original abstract
As generative artificial intelligence becomes one of the most significant systems of knowledge production in our society today, questions relating to who can access and shape that production grow increasingly important in our discourse. This paper argues that the existing frameworks for AI literacy, which are dominated by technical competency and responsible-use principles, are insufficient because they enforce a "consumer" orientation toward AI rather than fostering genuine epistemic agency. Based upon Foucault's concept of power-knowledge, Freire's pedagogy of critical consciousness, and scholarship of digital literacy, this paper proposes a reconceptualization of AI literacy as a critical practice that equips individuals not just to use AI systems, but to critically evaluate them, resist their structuring assumptions, and participate in their governance. The paper further argues that unequal access to AI tools in society recapitulates longstanding epistemic injustices, and that a literacy framework oriented toward empowerment must account for these structural inequities. A three-part framework of AI literacy based on the notions of contextual use, critical interrogation, and participatory governance frames this literacy as a cultivation of epistemic "agents" rather than the training of competent consumers of AI-generated information.
Figures
Reference graph
Works this paper leans on
- [2]
-
[2023]
AI Governance: Themes, Knowledge Gaps and Future Agendas. Internet Research 33(7): 133–167. Chuan, C. H.; Sun, R.; Tian, S.; and Tsai, W. H. S. 2024. Explain- able Artificial Intelligence (XAI) for Facilitating Recognition of Algorithmic Bias: An Experiment from Imposed Users' Perspec- tives. Telematics and Informatics 91: 102135. Coeckelbergh, M. 2023. D...
arXiv 2024
Reviewed August 1, 2026 · model on record in the stance chip above.
Discussion (0). Sign in to comment.