{"id":"13306979-a868-4e74-9209-79af1c0f94f7","arxiv_id":"2411.17840","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"DoD grant solicitations from 2007 to 2023 use a recurring 'one small problem' rhetorical pattern that turns acknowledged limits of AI into justifications for more military research funding, entangling academic AI researchers in war-fighting agendas.","lead":"This paper examines 16 years of U.S. Department of Defense grant solicitations and argues that even calls framed as 'basic research' push academic AI researchers toward military goals. It offers scholars and policymakers a critical vocabulary for seeing how funding language recruits researchers into war-fighting agendas.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The enlistment claim rests on a one-directional reading of solicitation texts: the paper shows that DoD calls frame research militarily, but it does not establish that this framing recruits researchers or shapes their agendas, which is the central causal claim.","rationale":"The reader's weakest_assumption accurately identifies the load-bearing causal mechanism: the performative, recruiting force of solicitation language. My reading of the full text confirms that the paper provides rich textual evidence of military framing in DoD solicitations, including specific examples from VBFF, MURI, Minerva, and DARPA calls, and the 'one small problem' rhetorical analysis is well supported by the quoted passages. However, the leap from 'solicitations articulate military agendas' to 'researchers are enlisted, and the basic-research trope shelters them from moral questions' is not empirically secured. The paper explicitly says its analysis is 'indicative' and does not claim a comprehensive survey, so the absence of researcher-side evidence is a genuine gap rather than a fatal flaw. Because the verdict is already CONDITIONAL with a stated weakness matching mine, and because the paper would still make a meaningful contribution as a textual/critical analysis even if the causal mechanism were weakened, I do not change the verdict. I would, however, recommend that the condition for acceptance include either the reframing of the central claim as an articulation/imagination claim or additional evidence of researcher uptake. The paper deserves credit for its corpus work (7187 PDFs from grants.gov, DBSCAN clustering, LDA topic modeling) and for clearly flagging the 'one small problem' rhetorical device with concrete quoted examples. The 'mutual' aspect of 'mutual enlistment' is the thinnest part: the paper shows solicitations tracking academic trends (e.g., GANs appearing in solicitations shortly after academic prominence), but it does not show academic researchers shaping DoD problem formulations, which is the other direction of the claimed mutuality. This is a matter of evidence weight, not internal inconsistency, so it is a correctness-risk concern rather than a circularity or soundness problem.","tokens_in":16516,"tokens_out":1999,"duration_ms":16821,"concrete_test":"Analyze a sample of funded proposals or researcher-authored publications under the cited programs (e.g., MURI, VBFF, Minerva) and code whether researchers adopt the solicitations' military framing in their problem statements, or alternatively conduct 10-20 semi-structured interviews with principal investigators who received these awards, asking how they interpreted 'basic research' and whether the solicitation language shaped their project framing or their moral assessment of the work. If researchers systematically adopt the framing, the enlistment claim is supported; if they treat solicitation boilerplate as irrelevant to their actual research agendas, the central causal claim weakens.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract and conclusion claim that grant solicitations 'work as a vehicle for the mutual enlistment' of DoD agencies and academic AI researchers, and that the basic-research trope 'offers shelter from significant moral questions.' What is actually demonstrated is that many DoD solicitations, including those labeled 'basic research,' contain explicit military framings and problem formulations. That is a solid textual finding. What is not demonstrated is the performative, recruiting effect on researchers: that reading or responding to these texts enrolls researchers into military problem formulations or provides them with moral wiggle room. The paper's own evidence is textual plus one cited study of NLP funding culture (Gururaja et al. 2023), which concerns DoD benchmark culture's effect on the NLP field, not solicitation-language enlistment. No researcher interviews, no analysis of awarded proposals, no comparison of the language researchers actually use before versus after funding, and no evidence that researchers who receive these grants adopt the DoD's framing. The closest in-text mechanism claim is in the section 'How basic research is directed to military ends' ('It enrolls them into thinking like the DoD...'), but this is asserted, not shown. The 'mutual' part of 'mutual enlistment' is especially thin: the paper documents that solicitations respond to academic trends (e.g., GANs appearing shortly after prominence in academic discourse), but this is timing correlation across a few examples, not an account of how academic researchers shape DoD agendas. The central claim would still be plausible if reframed as 'solicitations articulate a military imaginary and seek to enroll researchers,' but the stronger claim of actual enlistment and moral shelter goes beyond the evidence.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper examines U.S. Department of Defense grant solicitations from 2007 to 2023, focusing on calls addressed to artificial intelligence researchers. It draws on a corpus of DoD solicitations and uses search, clustering, and topic modeling to support close readings of three recurring programs (MURI, Vannevar Bush Faculty Fellowship, Minerva) and a subset of DARPA calls. The authors argue that calls framed as 'basic research' are nonetheless saturated with military problem formulations; that a recurring 'one small problem' caveat justifies continued investment; that AI emerges as a durable frame for DoD research; and, most broadly, that grant solicitations are a vehicle for the 'mutual enlistment' of DoD agencies and academic researchers, with the basic/applied trope offering moral shelter to researchers. The manuscript is explicit that its analyses are indicative rather than comprehensive.","tokens_in":16729,"tokens_out":6348,"duration_ms":62300,"significance":"If its central claims are accepted in suitably qualified form, the paper is a useful contribution to critical AI studies, STS, and science policy. It grounds its close readings in a documented corpus of primary sources, cites specific solicitations and page numbers, and identifies rhetorical devices—the 'one small problem' caveat and the 'durable frame'—that are likely to be of analytic value to other scholars. The authors are transparent about the exploratory scope of the computational analysis and acknowledge that a fuller study would require examining awarded contracts. The main weakness is that the paper's most categorical claims about 'enlistment' and 'moral shelter' are performative/causal claims that outrun the textual evidence presented; the manuscript is strongest when read as an interpretive analysis of solicitation discourse rather than as a demonstration of effects on researcher behavior.","major_comments":[{"comment":"The load-bearing claim that solicitation language 'enrolls [researchers] into thinking like the DoD in their conceptualization of research problems' and provides 'moral wiggle room' is asserted rather than demonstrated. The evidence shows that DoD calls frame research in military terms and that successful proposals must be positioned against DoD priorities, but it does not show that researchers actually adopt these problem formulations or experience moral shelter. The one empirical study cited in this context, Gururaja et al. (2023), documents effects of DoD benchmark funding on NLP field culture, not the perlocutionary effect of solicitation texts on applicants. I recommend either recasting the claim as 'solicitations work to enlist' or 'attempt to enroll,' or supporting the stronger causal version with evidence from awarded proposals, researcher discourse, or interviews.","section":"How basic research is directed to military ends"},{"comment":"The 'mutual' component of 'mutual enlistment' is supported mainly by temporal adjacency: solicitations mention techniques such as GANs 'shortly after their prominence in academic discourse,' and 'belief formation' becomes 'belief propagation.' Timing alone does not establish that academic trends shape DoD agenda-setting; it is equally consistent with selective uptake by the DoD or with co-production through shared networks. To support the mutual-agenda-setting claim, the paper would need direct evidence of academic influence on solicitation formation (for example, via advisory bodies or program manager statements) or a more modest formulation such as 'selective uptake' rather than 'mutual enlistment.'","section":"One small problem / AI as a durable frame"},{"comment":"The manuscript explicitly says that its analyses 'are intended to be indicative, opening lines of investigation rather than offering a comprehensive survey,' but the abstract and conclusion draw categorical conclusions, including that 'grant solicitations work as a vehicle for the mutual enlistment' and that the basic-research trope 'offers shelter from significant moral questions.' The mismatch matters because the diachronic claims ('AI increasingly becomes entangled,' 'durable frame') are supported by selected examples rather than prevalence counts or negative cases. I would ask the authors to either soften the categorical conclusions so that they match the explicitly indicative design or add minimal quantitative grounding, such as counts of AI-related topics per year in the three program series, for the temporal-generalization claims.","section":"Interpretive methods with models"}],"minor_comments":[{"comment":"The computational pipeline would be easier to assess with more detail: report the DBSCAN parameters (epsilon and minPts), the LDA hyperparameters, the precise procedure used to select the DARPA 'AI' subset, and the number of documents analyzed in each of the three program series.","section":"Interpretive methods with models"},{"comment":"The statement that deep learning was 'wholly unsuitable' for military problems until about 2020 is based on a small number of selected solicitations; given that other calls in the same period discuss neural interfaces and AI-related neuroscience, 'wholly' appears too strong and should be qualified.","section":"One small problem"},{"comment":"The citation 'Hastie, Tibishrani, and Friedman (2009)' contains a typo; the correct name is Tibshirani.","section":"Footnote 8"},{"comment":"Figure 1 is referenced in the text but is not visible in the manuscript version provided; if the figure is part of the paper, ensure that it is included and captioned in the final submission.","section":"Mind's Eye Program discussion"},{"comment":"The conclusion extends the argument to the 'military-industrial-commercial-academic complex,' but the corpus contains DoD solicitations only; the commercial side of the claim should be flagged as an interpretive extension rather than an empirical finding of this paper.","section":"Conclusion"}],"recommendation":"major_revision","confidential_remarks":"The paper has good primary-source material and several sharp interpretive observations. Its main problem is a mismatch between the categorical language of the abstract and the explicitly 'indicative' method, especially for the causal 'enlistment' claim. I see no novelty or citation problem, and no need for additional data collection if the authors are willing to reframe the central claims to match the evidence. With that reframing and the methodological details requested, the paper could be a solid contribution to a critical AI-studies venue."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know: this paper gives you the best close reading I've seen of how DoD 'basic research' solicitations are saturated with military framings, and it names a real recurring rhetorical move ('one small problem'). The central claim that solicitations actually enlist researchers into military problem formulations, and that the basic-research label offers moral shelter, is plausible but not directly demonstrated. The stress-test note lands: the paper shows the texts frame research militarily; it doesn't show researchers adopt that framing.\n\nWhat's genuinely new: the corpus (7,187 DoD PDFs, 2007-2023) and the diachronic reading of MURI, VBFF, and Minerva series, tracing how AI techniques appear in solicitations shortly after academic prominence. The 'flexible tether' metaphor for the basic/applied distinction is useful. The DARPA battlefield reading is careful and well-cited, and the authors are honest that the analysis is indicative rather than comprehensive.\n\nSoft spots: the performative mechanism of enlistment is asserted rather than shown. The closest is 'It enrolls them into thinking like the DoD...' but no researcher interviews, no awarded-proposal analysis, no before/after comparison. The 'mutual' side of mutual enlistment is thin—mostly timing correlations. Also, the computational choices (LDA topic count, DBSCAN parameters) are underspecified and no code is shipped, though the authors do provide a citation to the original dataset and a manifest of cited documents. A prevalence count or negative cases would strengthen the 'one small problem' claim, which is otherwise selected examples.\n\nWho it's for: critical AI studies, STS, research ethics, university administrators who care about the terms of military funding. It deserves a serious peer review; the empirical corpus and the rhetorical pattern are worth debating. I'd recommend publication after revision, with the causal claims softened or supported.\n\nI'd bring this to a reading group—the 'flexible tether' and 'one small problem' are citable concepts that sharpen discussions of military-academic funding.","headline":"A solid close reading of DoD solicitation rhetoric, but the 'enlistment' claim is stronger than the evidence supports.","tokens_in":17393,"tokens_out":2064,"would_cite":true,"duration_ms":18620,"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":"The paper argues that DoD grant solicitations framed as 'basic research' function as a vehicle of mutual enlistment, steering academic AI research toward military ends while giving researchers moral shelter.","keywords":["U.S. Department of Defense","military AI research","basic vs applied research","grant solicitations","academic research funding","enlistment","DARPA","warfighting"],"falsifier":"A concrete test would be to interview or survey researchers who received DoD grants framed as basic research about whether they understood their work as militarily directed and whether the 'basic' label provided moral comfort, and to compare funded proposals against the solicitations to see whether the performed work actually aligns with the military priorities expressed in the calls. If funded research does not track the solicitations and researchers do not adopt the DoD's framing, the central claim would be contradicted.","tokens_in":16238,"feed_emoji":"🎖️","tokens_out":7559,"duration_ms":61711,"temperature":0.7,"pith_summary":"This paper examines 7,187 U.S. Department of Defense grant solicitations from 2007 to 2023 and argues that calls framed as 'basic research' nonetheless enlist academic AI researchers in a warfighting agenda. The authors contend that the basic/applied distinction is not a clean line but a 'flexible tether' that binds the military and the research community: proposals are steered toward defense priorities while the 'basic' label gives researchers moral wiggle room. The paper supports this claim through three analyses: a critique of the basic/applied framing in the calls, a longitudinal study of a recurring 'one small problem' caveat that justifies further investment, and a reading of DARPA solicitations for battlefield AI. If correct, the paper shows that the moral comfort researchers take from doing 'basic' military-funded work is largely illusory, and that the research community is more deeply implicated in U.S. militarism than the funding label suggests.","feed_headline":"Basic research label recruits AI scientists to military aims","feed_subtitle":"Analysis of 7,187 DoD solicitations shows the basic/applied split is a 'flexible tether' that gives researchers moral cover.","key_machinery":"The central mechanism is the rhetorical use of the basic/applied research distinction, which the paper characterizes as a 'flexible tether' that binds military and academic interests without making the link explicit. Supporting this are the 'one small problem' caveat—a rhetorical move where a solicitation affirms the promise of an existing technique, then identifies an outstanding problem that requires more research—and the 'durable frame' of AI, which recasts an ever-widening set of problems in machine-learning terms. These devices operationalize the enlistment: the language of the call steers proposers to align their work with DoD priorities while preserving the appearance of investigator-driven basic science.","core_discovery":"The paper argues that U.S. Department of Defense grant solicitations, including those explicitly labeled 'basic research,' are a vehicle for the mutual enlistment of DoD funding agencies and the academic AI research community in setting research agendas. The 'basic/applied' distinction is not a clean separation but a 'flexible tether' that lets the military steer university research toward its warfighting imaginaries while giving researchers a way to tell themselves their work has no military end use. The paper shows this through three analyses: examination of the basic/applied framing in calls, a diachronic study of a 'one small problem' caveat across three recurrent funding programs, and a reading of DARPA solicitations for battlefield AI. It concludes that the trope of basic research offers shelter from moral questions while obscuring researchers' implication in U.S. militarism.","pith_inferences":["The paper's mechanism suggests that even researchers who consciously avoid weapons work can be enrolled in military agenda-setting simply by framing proposals to fit a DoD call; the enlistment happens at the level of problem formulation, not just end use.","A direct testable extension would track funded proposals from the same solicitation cycles to see whether the work performed matches the military priorities expressed in the calls, and whether researchers describe their own projects as 'basic' in defense of the work.","The 'flexible tether' concept could generalize to other mission agencies (e.g., health, energy, space) as a way to study how civilian research gets aligned with state priorities while maintaining the appearance of researcher autonomy.","If the argument holds, the current debate over military AI ethics is incomplete: it focuses on weapons applications, while the more pervasive influence may be the subtle recruitment of the research community's imagination and problem choices."],"forward_implications":["University researchers who accept grants framed as basic research should assume their work is being evaluated for its relevance to military priorities, even if no weapon is named.","The moral reassurance that 'basic research' is unconnected to war is illusory; researchers and universities may be implicated in militarism through the agendas they help advance.","If solicitation language steers research, then diversifying the sources of academic funding or changing the language of calls could shift the direction and ethical valence of AI research.","The recurring 'one small problem' caveat creates a self-perpetuating funding cycle: affirmed progress plus an identified shortfall justifies the next round of investment, so 'basic research' grants do not wind down as problems are solved."],"supporting_citations":[{"why":"Supplies the corpus of 46,175 federal grant solicitations that is the paper's data foundation.","marker":"(Bateyko & Levy, 2024)"},{"why":"Provides the concept of 'enrolment' that the paper adapts as 'enlistment' to analyze how problems and actors are recruited into a shared agenda.","marker":"(Callon 1984)"},{"why":"Shows parallel evidence that DoD funding shaped the culture and direction of NLP research, supporting the claim that funding steers research norms.","marker":"(Gururaja et al. 2023)"},{"why":"Gives the formal budget-activity categories and numbers showing 'basic' research is a small fraction of DoD R&D, used to contextualize the basic/applied distinction.","marker":"(Sargent, Jr, 2022)"},{"why":"Source of the 'moral wiggle room' concept that explains how the basic-research label shields researchers from ethical scrutiny.","marker":"(Widder and Nafus 2023)"},{"why":"Documents the expansion of DoD funding into commercial tech and academia, framing the military-academic complex the paper describes.","marker":"(Whittaker 2021)"},{"why":"Provides the critique of human-machine teaming and the 'machine in the middle' imaginary that the paper applies to DARPA's human-machine collaboration calls.","marker":"(Suchman 2007)"},{"why":"The Defense Science Board report on autonomy that the paper uses to show how 'trustworthiness' language justifies increased automation of weapons.","marker":"(Office of the Undersecretary of Defense 2016)"}],"fun_headline_variants":["DoD 'basic research' calls draft AI academics into war aims","How 'basic research' grants enlist AI scientists in militarism","Military AI funding hides enlistment behind 'basic research' label","The 'flexible tether' connecting basic AI research to war"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument rests on the premise that the language of a grant solicitation has performative force—that reading a call framed as 'basic research' actually shapes researchers' problem formulations and shields them from moral questions—and if that premise fails, the enlistment claim is not established.","fun_headline_variants_meta":{"raw":{"variants":["DoD 'basic research' calls draft AI academics into war aims","How 'basic research' grants enlist AI scientists in militarism","Military AI funding hides enlistment behind 'basic research' label","The 'flexible tether' connecting basic AI research to war"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000251,"raw_usage":{"total_tokens":1571,"prompt_tokens":976,"completion_tokens":595,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":592,"completion_tokens_details":{"reasoning_tokens":522}},"tokens_in":592,"tokens_out":595,"duration_ms":4903,"temperature":1.0,"reasoning_tokens":522,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:47:34.066916+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete test would be to interview or survey researchers who received DoD grants framed as basic research about whether they understood their work as militarily directed and whether the 'basic' label provided moral comfort, and to compare funded proposals against the solicitations to see whether the performed work actually aligns with the military priorities expressed in the calls. If funded research does not track the solicitations and researchers do not adopt the DoD's framing, the central claim would be contradicted.","supporting_citations":[],"review_version":1}