{"id":"7d68a27d-dab9-44fb-a095-d24d26f1c2bf","arxiv_id":"2607.27547","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"AI literacy, reframed through Foucault and Freire, is an exercise in power-knowledge that should cultivate epistemic agents rather than competent consumers.","lead":"A philosophy-of-education essay argues that current AI literacy frameworks train people to consume AI output rather than understand the power structures inside the systems, and proposes a literacy of contextual use, critical interrogation, and participatory governance. It matters because governments and universities are building AI literacy curricula now, and those curricula decide who gets to question AI systems.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Blanket claim that existing AI-literacy frameworks omit power rests on a four-framework sample whose own brief descriptions partially contradict it.","rationale":"The reader's CONDITIONAL verdict is well-calibrated. This is a conceptual essay in critical information studies, and the normative proposal—Contextual Use, Critical Interrogation, Participatory Governance as complements to competency-based literacy—is coherent, honestly scoped, and grounded in the cited critical literature. The paper is not making a formal or machine-checkable claim, and its own conclusion correctly disclaims being a complete curriculum. The weakest point, as the reader noted, is the empirical/interpretive generalization from four selected frameworks to 'existing frameworks,' especially the assertion that all four give 'no substantive treatment' to power-related topics. That premise is load-bearing because the insufficiency claim is the paper's motivating conclusion; if one of the four frameworks already substantively engages governance or structural critique, the central dichotomy between competency-based 'consumers' and critical 'agents' becomes an ideal type rather than a description of the existing field. My proposed test—verbatim coding of the four frameworks against the paper's own categories—is a concrete way to settle this. If the coding shows such content exists, the paper should move to a graduated claim: some frameworks partially address power, but none treats it with sufficient depth or centrality. That is a weaker but still defensible thesis and would not overturn the paper's positive proposal. I therefore agree with the reader's weakest-assumption identification and would keep the verdict CONDITIONAL rather than REJECT, because the concern is addressable and the theoretical contribution stands even if the historical generalization needs qualification.","tokens_in":10479,"tokens_out":3953,"duration_ms":48520,"concrete_test":"Retrieve the verbatim texts of the four cited frameworks (U.S. DOL 2026, Penn State 2026, UNESCO 2024, EC/OECD via Milberg 2025). Using the paper's own categories—training-data politics, participatory governance, structural conditions of AI knowledge production—code every competency/outcome statement in each framework for explicit or implicit coverage, with two independent coders. If any one framework contains substantive coverage of any of those three categories, the blanket 'no substantive treatment' claim is empirically false; the paper must either restrict its scope to 'most frameworks' or reframe its thesis as 'current frameworks treat these dimensions insufficiently rather than not at all.' A secondary check: repeat the selection by a stated criterion (e.g., first ten hits for 'AI literacy framework' in a standard search) and re-run the coding.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that 'existing frameworks' are dominated by a competency/consumer model and that, 'across all four,' they have 'no substantive treatment of training data politics, governance participation, or the structural conditions of AI knowledge production' (section 'AI Literacy as Empowerment, Not Competency')—depends on the representativeness of four hand-picked frameworks and on an all-or-nothing characterization of them. The paper provides no sampling criteria, and its own descriptions undercut the blanket verdict: UNESCO is said to frame students as 'co-creators,' Penn State is said to attend to 'equity, labor, and democracy,' and the EC/OECD framework calls learners 'not just consumers... ethical stewards.' These features are then dismissed as still 'converging' on a careful user, but that dismissal is interpretive, not demonstrated by quotations from the frameworks. If UNESCO's or Penn State's actual competency lists contain substantive governance or critical-interrogation content, then the claim that existing frameworks systematically omit power is false; at most the frameworks could be argued insufficiently deep or unevenly implemented. The same unproven leap underlies the assertion that omission 'reflects the interests of the institutions,' which is asserted without evidence. The theoretical contribution survives as a normative proposal, but the empirical premise that motivates it is not established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","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.","tokens_in":10667,"tokens_out":4212,"duration_ms":47665,"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":[{"comment":"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","section":"AI Literacy as Empowerment, Not Competency"},{"comment":"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.","section":"AI Literacy as Empowerment, Not Competency"},{"comment":"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.","section":"AI Literacy as Empowerment, Not Competency / Competency Creates Consumers; Literacy Creates Agents"}],"minor_comments":[{"comment":"Grammatical issue: 'competent consumer of AI-generated information' should be 'competent consumers' to agree with the plural subject.","section":"Abstract"},{"comment":"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.","section":"Figure 1"},{"comment":"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.","section":"A Three-Part Framework"},{"comment":"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.","section":"AI Literacy as Empowerment, Not Competency"},{"comment":"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.","section":"Conclusion"}],"recommendation":"major_revision","confidential_remarks":"The normative core of the paper is sound and worth publishing after revision. The main concern is the empirical under-support for the blanket claim about existing frameworks; this can be addressed by a more transparent sample selection, direct textual evidence, or a softened claim. The 'institutional interests' assertion should be reframed as a hypothesis or supported. The paper also overlaps with existing critical AI literacy scholarship (e.g., Stamboliev 2023; Velander et al. 2024) and should position itself more clearly relative to that work. No concerns about author integrity; the issues are evidentiary and presentational."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe paper is worth reading for its framework. The three dimensions — contextual use, critical interrogation, participatory governance — are a genuinely useful re-grouping, and the agent/consumer contrast is helpful for teaching. The authors anchor it in the right critical literature (Stamboliev, Coeckelbergh, Velander) and are honest that this is not a complete curriculum. That is real credit.\n\nThe soft spot is exactly where the stress-test puts it. The claim that existing frameworks 'across all four' have no substantive treatment of training-data politics, governance participation, or structural conditions is the load-bearing empirical premise, and it is not established. Four frameworks are chosen without stated criteria, and the paper's own descriptions undercut the blanket verdict: UNESCO frames students as 'co-creators,' Penn State attends to equity, labor, and democracy, and the EC/OECD calls learners 'ethical stewards.' The dismissal of these as still converging on a 'careful user' is interpretive, not demonstrated by quotation. The institutional-interest explanation is asserted, not shown. This is a fixable weakness, not a fatal one.\n\nEverything else holds together. The theory is applied coherently; the contextual-use discussion via Bates' berrypicking is a nice touch, and the concrete examples of participatory governance (EU AI Act consultation, Detroit Digital Justice) keep it grounded. The widening-gap claim is figurative, but it is presented as a structural argument, not a measurement.\n\nWho gets value: anyone working on AI literacy curricula, library/information science, or critical pedagogy. It deserves a serious referee; the empirical section needs rewriting, but the conceptual contribution is solid.\n\nRecommendation: send it to review, with the expectation that the authors be asked to state their framework-selection criteria and engage the counter-examples directly.","headline":"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.","tokens_in":11236,"tokens_out":1643,"would_cite":true,"duration_ms":18525,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"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.","keywords":["AI literacy","power-knowledge","epistemic agency","critical pedagogy","competency model","participatory governance","generative AI","critical interrogation"],"falsifier":"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.","tokens_in":10221,"feed_emoji":"🧠","tokens_out":5142,"duration_ms":48911,"temperature":0.7,"pith_summary":"This paper argues that current AI literacy frameworks, built around technical competency and responsible-use principles, train people to be competent consumers of AI-generated knowledge rather than epistemic agents. Drawing on the power-knowledge concept—the idea that knowledge systems and power are constitutively entangled—and on critical pedagogy's contrast between 'banking' education and critical consciousness, the paper reconceives AI literacy as a critical practice. It proposes a three-part framework—contextual use, critical interrogation, and participatory governance—that would equip people to evaluate AI systems, resist their structuring assumptions, and participate in their governance. If the paper is right, any AI literacy curriculum that omits training-data politics and governance participation is at best incomplete and at worst an instrument of adaptation, reproducing the epistemic hierarchy it claims to address.","feed_headline":"AI literacy should teach power, not just responsible use","feed_subtitle":"Without training-data politics and governance skills, AI literacy may train consumers rather than agents.","key_machinery":"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","core_discovery":"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","pith_inferences":["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."],"forward_implications":["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."],"fun_headline_variants":["From consumer to agent: reframing AI literacy","AI literacy should teach power, not just use","Rethinking AI literacy: power, critique, governance","Why AI literacy needs epistemic agency, not just skills"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"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.","fun_headline_variants_meta":{"raw":{"variants":["From consumer to agent: reframing AI literacy","AI literacy should teach power, not just use","Rethinking AI literacy: power, critique, governance","Why AI literacy needs epistemic agency, not just skills"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000119,"raw_usage":{"total_tokens":895,"prompt_tokens":686,"completion_tokens":209,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":430,"completion_tokens_details":{"reasoning_tokens":147}},"tokens_in":430,"tokens_out":209,"duration_ms":2926,"temperature":1.0,"reasoning_tokens":147,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T05:54:32.239892+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"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.","supporting_citations":[],"review_version":1}