{"id":"3367baec-946c-447f-9dad-cebe6e74e3a9","arxiv_id":"2503.04751","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper introduces Human-Driven AI in Higher Education, a five-phase governance framework mapping human intelligence onto the AI lifecycle, but provides no empirical validation that it achieves ethical AI outcomes.","lead":"This paper proposes a conceptual framework called HD-AIHED for keeping human oversight at the center of AI use in universities, and it claims the framework can align AI adoption with UNESCO, OECD, and other ethical standards. A generalist reader might use it as a checklist of governance components, but the paper contains no empirical test of whether the framework actually works.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The framework's central 'ensures' claim rests on circular validation: Section 7.3 says objectives were met, while Section 2.6 says empirical testing is future work, and Table 8 only relabels cases in the framework's own phases.","rationale":"The reader's weakest assumption identified essentially the same load-bearing problem: the framework's validity rests on treating a non-systematic meta-synthesis and the author's mapping of university AI tools into five pre-defined phases as sufficient evidence, while the mapping may simply impose the framework's own categories onto the cases. My review confirms and sharpens this concern with a specific textual contradiction: Section 2.6 says empirical testing is future work, while Section 7.3 claims all research objectives, including empirical validation, were met. Because the central claim is an assurance claim about future governance outcomes, the absence of any non-circular empirical support is fatal to the paper as written. The framework may be a reasonable conceptual proposal, but the manuscript's stated contributions overstate what has actually been demonstrated. I therefore see no reason to change the reader's REJECT verdict; if anything, the internal contradiction between 2.6 and 7.3 strengthens it. The proposed concrete test would settle whether Table 8 counts as validation or only as retrofitting, and it could be completed from the manuscript's own cited sources.","tokens_in":39913,"tokens_out":2436,"duration_ms":27105,"concrete_test":"Independently reconstruct the evidence trail behind Table 8: for each of the roughly fourteen institutional rows, list the specific cited source or institutional report that documents the phase-specific HD-AIHED components claimed for that institution, such as SWOC analysis, AI ethical review board approval, pilot testing, or a named feedback loop. If a majority of rows have no source actually documenting those components, and the phase labels appear only as the author's post hoc categorization, then the mapping is circular and provides no validation of the framework's effectiveness.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that the HD-AIHED framework enables ethical, inclusive, human-centered AI integration in higher education. For that claim to hold, applying the framework would have to produce the promised governance and equity outcomes. The paper provides no such evidence. Section 2.6 explicitly states that 'future empirical testing is required' and Section 2.7 concedes reliance on secondary data; yet Section 7.3 asserts 'This study successfully meets its research objectives,' including RQ3's mandate to 'develop and empirically validate' the framework. That is a direct internal contradiction. The only purported validation is the mapping in Section 2.5 and Table 8, which classifies AI applications at named universities into the framework's own phases: Adoption, Design, Deployment, Evaluation, and Exploration. Those columns are the framework's categories, so the exercise cannot fail and yields no independent confirmation. There are no outcome metrics, no baseline comparisons, no counterfactual cases, and no inter-rater protocol to prevent the author from imposing the framework's vocabulary onto each institution. Consequently, the abstract's claim that the model 'bridges AI research gaps, addresses global real-time challenges, and provides tailored, scalable, and ethical strategies' and Section 5.3's claim that it 'ensures ethical AI governance, institutional sustainability, and adaptive transformation' are unsupported assertions rather than established results. This is the load-bearing gap: the practical value of the framework is exactly the assurance it claims, and that assurance depends on validation evidence that the manuscript itself defers to future work.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes the Human-Driven AI in Higher Education (HD-AIHED) framework, a conceptual five-phase model for integrating human oversight into AI adoption, design, deployment, evaluation, and exploration in higher education. The framework includes stakeholder participation, SWOC analysis, AI ethical review boards, feedback loops, and alignment with UNESCO/OECD/GDPR standards. The paper synthesizes secondary literature and a set of institutional examples to argue that the framework can address algorithmic bias, privacy risks, governance gaps, and scalability challenges. It claims in the abstract and in Section 5.3 to ensure ethical AI governance and institutional sustainability, and Section 7.3 asserts that all research objectives, including empirical validation, have been met.","tokens_in":40098,"tokens_out":4667,"duration_ms":48343,"significance":"The conceptual contribution is potentially useful: the five-phase structure is coherent, the stakeholder taxonomy is comprehensive, and the emphasis on human oversight, dynamic feedback, and alignment with global ethical frameworks addresses recognized gaps in AI governance in higher education. The paper also provides a concrete future research agenda, including a structural equation modeling design. However, these strengths are undermined by a direct internal contradiction concerning validation: the manuscript simultaneously states that empirical testing is future work and that empirical validation has been successfully completed. The only evidence offered for validation is a mapping exercise that uses the framework's own categories, which cannot provide independent confirmation. If the overclaims are removed and the framework is repositioned as an untested conceptual proposal, the paper could make a modest but useful contribution. As it stands, the central claims of empirical support and guaranteed ethical outcomes are unsupported.","major_comments":[{"comment":"There is a direct internal contradiction regarding empirical validation. Section 2.6 states that 'future empirical testing is required to evaluate the HD-AIHED model’s applicability,' and Section 2.7 concedes reliance on secondary data and the need for longitudinal and experimental studies. Yet Section 7.3 asserts 'This study successfully meets its research objectives,' which include RQ3's mandate to 'develop and empirically validate' the framework. No empirical data are reported anywhere in the manuscript. This contradiction matters because the abstract's claim that the model 'bridges AI research gaps, addresses global real-time challenges, and provides tailored, scalable, and ethical strategies' depends on the validation claim.","section":"§2.6, §2.7, §7.3"},{"comment":"The purported validation is circular. Section 2.5 lists 'Testing Real-World AI Applications' and 'Analyzing Case Studies of Real-World Challenges' as validation activities, but the only concrete artifact is Table 8, which maps AI applications at named universities into the framework's own five phase columns (Adoption, Design, Deployment, Evaluation, Exploration). Because the columns are the framework's categories, the exercise cannot fail: every application can be assigned to a phase. There are no outcome metrics, no baseline comparisons, no counterfactual cases, and no inter-rater protocol. Section 6.2 therefore overstates the evidence when it says the mapping 'offers a validated framework.'","section":"§2.5, §6.2, Table 8"},{"comment":"Table 8 presents institution-specific AI applications (e.g., 'University of Cambridge – AI-Powered Campus Energy Optimization,' 'Oxford University – AI-Powered Emotional Intelligence Analysis') without any citations or sources. These entries are presented as empirical cases, but the reader cannot verify that the named institutions deployed these specific tools or that the described inputs and phases correspond to reality. Without verifiable sources, the mapping cannot serve as evidence for the framework's applicability or effectiveness.","section":"§6.2, Table 8"}],"minor_comments":[{"comment":"There are two tables numbered 'Table 8' in the manuscript: one in Section 6.2 and one in Section 7.2.1. Renumber the second table.","section":"§6.2, §7.2.1"},{"comment":"Section 7.1 refers to 'The AIED-HDMFB model,' an acronym that appears nowhere else in the paper. This is presumably a typo for HD-AIHED and should be corrected.","section":"§7.1"},{"comment":"The text contains repeated passages: Section 3.4.1 has three near-identical paragraphs starting 'Global platforms like Coursera and edX...', and Section 5.2.5 repeats the same sentence about Institutional AI Ethical Review Boards twice. These should be consolidated.","section":"§3.4.1, §5.2.5"},{"comment":"Citation style is inconsistent: some citations use bracketed numbers, others use author–year (e.g., 'Singh, 2024', '(Bond et al., 2024)'), and some are phrased awkwardly like 'as [35] notes.' Standardize the citation format.","section":"Throughout"},{"comment":"Figure callouts are confusing: Section 5.2 refers to 'the accompanying Figure 2' when describing the integrated system, but Figure 2 is titled 'Structural Components of Framework'; Section 5.2.7 refers to Figure 4 in a way that seems to describe Figure 5. Check all figure cross-references.","section":"§5.2, §5.2.7"}],"recommendation":"major_revision","confidential_remarks":"The paper's central contradiction—claiming empirical validation while explicitly deferring it to future work—is serious and would normally justify rejection. I recommend major revision rather than rejection because the conceptual framework itself is coherent and the overclaims are fixable: the authors could remove the 'empirically validated' language, reframe Section 7.3 as describing the conceptual development rather than empirical success, and reposition Table 8 as an illustrative mapping rather than validation. If the authors instead insist on the validation claim, the paper would need substantial new empirical data that are not present. The journal should consider whether a purely conceptual framework with no empirical support is within its scope; if so, the revision can succeed within the manuscript's current scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The short version: this paper proposes a five-phase \"HD-AIHED\" governance framework for AI in higher education. The framework itself is coherent, and the literature review is thorough. But the paper's central claim that the framework \"ensures\" ethical outcomes is unsupported, and the validation section is circular. The abstract and Section 5.3 overstate what a conceptual model can do.\n\nWhat's actually new: the specific configuration of five phases (adoption, design, deployment, evaluation, exploration) mapped onto human intelligence layers, decision-making checkpoints, and twelve feedback loops. That specific synthesis does not appear in the cited literature. The paper also does a decent job cataloging global ethical frameworks (UNESCO, OECD, EU AI Act) and identifying practical gaps like the absence of institutional AI ethics boards. The SWOC-in-adoption idea is reasonable.\n\nThe soft spots are real. Section 2.6 says future empirical testing is required, and Section 2.7 concedes reliance on secondary data. But Section 7.3 claims all research objectives were met, including RQ3's mandate to \"develop and empirically validate\" the framework. That is a direct internal contradiction. The only purported validation is Table 8, which maps AI applications at named universities into the framework's own five phases. That exercise cannot fail: the categories are the framework's own, so the mapping provides no independent evidence. There are no outcome metrics, no baseline comparisons, no inter-rater protocol. The abstract's \"ensures\" language is therefore not justified.\n\nThe paper also has minor issues: duplicated text, numbering errors (two Table 8s), and some sloppy formatting. But the core problem is the gap between what is claimed and what is demonstrated.\n\nWho is this for? Someone looking for a comprehensive checklist of human oversight mechanisms for AI in higher education might find the framework heuristically useful. But it should not be sold as a validated solution. The paper would benefit from being rewritten as a purely conceptual proposal, with all \"ensures\" claims removed and validation explicitly deferred.\n\nMy recommendation: send it to peer review, but the referee should insist on removing the overclaims and reframing the mapping exercise as an illustration, not validation. If the author does that, the framework could be a useful contribution to the governance literature. As it stands, the internal contradiction makes the current form unpublishable.","headline":"A coherent five-phase governance framework undermined by a circular validation and a direct contradiction between deferred testing and claimed empirical success.","tokens_in":40706,"tokens_out":2066,"would_cite":false,"duration_ms":21335,"reading_group":"maybe","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 ethical AI in higher education requires human intelligence embedded in every phase of the AI lifecycle, and proposes the HD-AIHED framework to do that.","keywords":["AI ethics","higher education","human-centered AI","AI lifecycle","AI governance","participatory design","SWOC analysis","ethical AI frameworks"],"falsifier":"An empirical falsifier would be a comparative pilot in which institutions using the framework show no reduction in documented algorithmic bias incidents, privacy complaints, or stakeholder trust gaps relative to matched institutions using standard AI procurement over a two-to-three-year period. A simpler check: if the mapping of university AI applications into the five phases can be reproduced by independent coders only when they are told the phase names, the framework may be imposing its categories rather than explaining the cases.","tokens_in":39628,"feed_emoji":"🎓","tokens_out":4166,"duration_ms":44187,"temperature":0.7,"pith_summary":"This paper argues that the right answer to whether AI should drive higher education is neither full automation nor avoidance: human intelligence should steer AI at every step. It proposes the Human-Driven AI in Higher Education (HD-AIHED) framework, a five-phase model covering adoption, design, deployment, evaluation, and exploration, with human decision points and feedback loops built into each phase. The claim is that this structure, combined with SWOC readiness analysis and university AI ethics review boards, prevents the bias, privacy, and governance failures that have accompanied AI tools in universities. If correct, institutions get a concrete governance blueprint that aligns AI with global ethics guidance while keeping the technology adaptable to local capacity.","feed_headline":"Five-phase AI framework keeps human judgment in charge at universities","feed_subtitle":"Adding ethics boards, SWOC reviews, and feedback loops at every phase lets campus AI augment rather than replace people.","key_machinery":"The central object is the HD-AIHED (Human-Driven AI in Higher Education) framework. It is a governance workflow in which five Human Intelligence layers (HI1–HI5) correspond to the AI lifecycle phases of adoption, design, deployment, evaluation, and exploration; decision-making points (DM) set strategic direction, and feedback loops (FB1–FB12) carry continuous corrections between layers. The framework also embeds a SWOC (Strengths, Weaknesses, Opportunities, Challenges) analysis at the adoption and exploration phases and assigns the AI Ethical Review Board, students, faculty, administrators, and external stakeholders specific governance roles.","core_discovery":"The paper's central claim is that ethical AI in higher education is not achieved by choosing better algorithms alone but by embedding human oversight into the entire AI lifecycle. The HD-AIHED framework maps five 'phased human intelligences' onto the five AI lifecycle phases—critical evaluation at adoption, resource selection at design, barrier removal at deployment, outcome evaluation, and future-scope planning—and connects them with feedback loops and decision-making checkpoints. The author maintains that this arrangement ensures accountability, inclusivity, transparency, and ethical compliance, and that it can be adapted to both well-resourced and under-resourced institutions.","pith_inferences":["The framework's five phases are general enough that they could be applied outside higher education, for example to AI adoption in schools or public-sector agencies, though the paper does not make that claim.","A natural test of the model would compare two otherwise similar universities, one using HD-AIHED and one using standard AI procurement, across several semesters; the paper does not report such data.","The model's heavy reliance on stakeholder participation may be easier to implement in institutions with strong administrative capacity than in underfunded ones, so the scalability claim may depend on institutional readiness rather than on the framework itself.","The quantitative SEM validation proposed in the paper could serve as the empirical check: if the hypothesized paths between the five phases and ethical outcomes do not hold in survey data, the framework's causal logic would need revision."],"forward_implications":["A university adopting HD-AIHED would run a structured readiness and SWOC review before buying or building any AI tool, rather than piloting tools first.","Bias and privacy risks would be checked at each phase, with designated humans—not the AI system—holding final approval through decision-making checkpoints.","Feedback loops would make AI systems continuously revised by student and faculty input, so the framework is designed to adapt as institutional needs change.","The same five-phase structure could be used by institutions in different regions, with each phase's criteria adjusted to local infrastructure and regulatory context.","Establishing AI ethical review boards would become a standard governance requirement, giving students a formal role in AI oversight."],"supporting_citations":[{"why":"Supplies the global ethics standard (UNESCO) that the framework aligns with and uses as a compliance benchmark.","marker":"[31]"},{"why":"Supplies the OECD AI principles that ground the framework's human-centered values and transparency requirements.","marker":"[50]"},{"why":"Documents the UK algorithmic grading failures, an example of the ethical harm the framework is designed to prevent.","marker":"[26]"},{"why":"Documents fragmented AI adoption and digital divide issues, motivating the framework's regionally adaptable governance.","marker":"[10]"},{"why":"Provides the empirical basis for bias audits and fairness reviews embedded in the framework's evaluation mechanisms.","marker":"[20]"},{"why":"Argues that AI tools are not neutral, which underlies the paper's insistence on continuous human oversight.","marker":"[36]"},{"why":"Supports the idea of human-centered adaptive systems, which the paper extends into phased human intelligence across the AI lifecycle.","marker":"[85]"}],"fun_headline_variants":["Five human checkpoints keep AI ethical on campus","Human oversight at every stage keeps campus AI ethical","Five checkpoints put humans in control of AI in higher ed","New framework puts human judgment at every AI decision point"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The framework's validity rests on the assumption that a non-systematic synthesis of English-language secondary sources, plus the author's mapping of university AI tools into the five phases, is enough evidence that the model works across diverse institutional contexts.","fun_headline_variants_meta":{"raw":{"variants":["Five human checkpoints keep AI ethical on campus","Human oversight at every stage keeps campus AI ethical","Five checkpoints put humans in control of AI in higher ed","New framework puts human judgment at every AI decision point"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000821,"raw_usage":{"total_tokens":3576,"prompt_tokens":908,"completion_tokens":2668,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":524,"completion_tokens_details":{"reasoning_tokens":2605}},"tokens_in":524,"tokens_out":2668,"duration_ms":17878,"temperature":1.0,"reasoning_tokens":2605,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T21:15:11.627391+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"An empirical falsifier would be a comparative pilot in which institutions using the framework show no reduction in documented algorithmic bias incidents, privacy complaints, or stakeholder trust gaps relative to matched institutions using standard AI procurement over a two-to-three-year period. A simpler check: if the mapping of university AI applications into the five phases can be reproduced by independent coders only when they are told the phase names, the framework may be imposing its categories rather than explaining the cases.","supporting_citations":[{"cited_title":"More than tools? Making sense of the ongoing digitizations of higher education","cited_arxiv_id":null,"evidence_quote":"Argues that AI tools are not neutral, which underlies the paper's insistence on continuous human oversight."}],"review_version":1}