Pith. sign in

REVIEW 3 major objections 5 minor 2 references

Basic Research, Lethal Effects: Military AI Research Funding as Enlistment

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read 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.

desk verdict A solid close reading of DoD solicitation rhetoric, but the 'enlistment' claim is stronger than the evidence supports. read the letter →

arxiv 2411.17840 v1 pith:K74BVZHQ submitted 2024-11-26 cs.CY cs.AI

David Gray Widder , Sireesh Gururaja , Lucy Suchman This is my paper · ORCID
classification cs.CYcs.AI
keywords U.S.DepartmentofDefensemilitaryAIresearchbasicvsappliedgrantsolicitationsacademicfundingenlistmentDARPAwarfighting
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

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.

What carries the argument

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.

What would settle it

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.

Watch

Extended reading notes

Core claim

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.

Load-bearing premise

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.

Editorial extensions

If this is right

  • 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.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • 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.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

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.

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 (3)
  1. [How basic research is directed to military ends] 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.
  2. [One small problem / AI as a durable frame] 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.'
  3. [Interpretive methods with models] 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.
minor comments (5)
  1. [Interpretive methods with models] 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.
  2. [One small problem] 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.
  3. [Footnote 8] The citation 'Hastie, Tibishrani, and Friedman (2009)' contains a typo; the correct name is Tibshirani.
  4. [Mind's Eye Program discussion] 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.
  5. [Conclusion] 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.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's interpretive reading of DoD solicitations is self-contained and does not reduce to its cited sources.

full rationale

This paper is a qualitative, interpretive study of an external corpus of DoD grant solicitations. It contains no fitted parameters, no equations, and no quantity predicted from fitted values. The central claim—that solicitations framed as 'basic research' nonetheless enlist researchers in a military agenda—is supported by primary-source quotations, e.g., the Vannevar Bush Faculty Fellowship call that pairs 'blue sky' research with 'long-term national security needs,' and DARPA calls describing AI for battlefield applications. The 'one small problem' pattern and 'durable frame' are analytic categories derived from close reading and topic modeling of the corpus, not defined in terms of the conclusion. Self-citations (Widder and Nafus 2023, Gururaja et al. 2023, Suchman 2007/2023) provide conceptual vocabulary and background, but the solicitation readings stand on the quoted texts. Gururaja et al. is cited as an independent empirical example of DoD funding shaping NLP research culture, not as a premise guaranteeing the enlistment conclusion. Suchman's critiques are used as critical context, not as hidden premises. The paper's potential vulnerability is evidentiary: the claim that solicitation language has a performative recruiting effect on researchers is an interpretive inference not directly demonstrated by interviews or proposal outcomes. That is a question of empirical support, not circularity. The paper also explicitly frames its analyses as 'indicative, opening lines of investigation' rather than comprehensive or predictive. Under the stated rules, no circular step can be exhibited, so the score is 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The paper has no fitted numerical model whose parameters determine a prediction, so no traditional free parameters appear. The three entries above are methodological choices that shape which documents were read closely, and they therefore shape the evidence for the central claim. The axioms capture the interpretive assumptions on which 'enlistment' rests: that solicitation language has performative effects, that the selected examples represent the corpus, and that topic model and clustering outputs are meaningful. No invented entities in the sense of new particles or forces are introduced; the only new analytic object is the 'one small problem' caveat, which is evidenced by quoted public documents.

free parameters (3)
  • LDA topic count = 40 (chosen via 'elbow optimum')
    The number of topics in the topic model was chosen by the authors to balance broad and specific topics; this shapes the thematic categories that guided close reading but does not enter the central claim numerically.
  • DBSCAN clustering parameters = Not reported (Levenshtein distance metric)
    Title clustering into recurring programs depends on an unreported distance threshold; the resulting clusters select the three program series analyzed.
  • DARPA AI subset filter = 'AI' keyword, agency DARPA, 2010-2023
    The set of 101 documents chosen for close reading depends on an unspecified search over 'AI'; different queries would yield different subsets, and this subset is the basis for the battlefield applications analysis.
assumptions (4)
  • domain assumption Grant solicitation language has performative force: the framing of calls shapes researchers' problem formulations and moral orientations.
    The paper's 'enlistment' argument assumes texts recruit researchers into military ways of thinking, as asserted in 'It enrolls them into thinking like the DoD' without direct evidence from researchers.
  • domain assumption The selected solicitations are representative enough to support general claims about DoD funding calls.
    The paper states its analyses are indicative, not comprehensive, yet the conclusion generalizes to 'grant solicitations work as a vehicle for mutual enlistment.' No negative case analysis or prevalence counts are provided.
  • standard math Topic model and clustering outputs are meaningful guides to thematic content under standard LDA bag-of-words assumptions.
    The computational methods rely on standard LDA and DBSCAN assumptions that documents can be represented as word distributions; the authors acknowledge interpretive judgment in building the model.
  • domain assumption The corpus of 7,187 PDFs from grants.gov is complete enough for the DoD solicitations of interest from 2007 to 2023.
    The pipeline excludes non-PDF documents and relies on automated download; the paper notes Word documents were omitted but judged mostly redundant.
invented entities (1)
  • 'One small problem' caveat as an analytic category independent evidence
    purpose: Names a recurring rhetorical move in solicitations: affirm progress, acknowledge an outstanding problem, then justify additional research funding for military settings.
    The pattern is evidenced by quotes from multiple MURI, VBFF, and Minerva solicitations in the public corpus; readers can verify the quotes, though the paper does not quantify prevalence.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Basic Research, Lethal Effects: Military AI Research Funding as Enlistment." pith.science (2026). https://pith.science/paper/K74BVZHQ

@misc{pith2026241117840,
  author       = {Pith},
  title        = {Pith review of: Basic Research, Lethal Effects: Military AI Research Funding as Enlistment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/K74BVZHQ}},
  note         = {Machine review of arXiv:2411.17840}
}
read the original abstract

In the context of unprecedented U.S. Department of Defense (DoD) budgets, this paper examines the recent history of DoD funding for academic research in algorithmically based warfighting. We draw from a corpus of DoD grant solicitations from 2007 to 2023, focusing on those addressed to researchers in the field of artificial intelligence (AI). Considering the implications of DoD funding for academic research, the paper proceeds through three analytic sections. In the first, we offer a critical examination of the distinction between basic and applied research, showing how funding calls framed as basic research nonetheless enlist researchers in a war fighting agenda. In the second, we offer a diachronic analysis of the corpus, showing how a 'one small problem' caveat, in which affirmation of progress in military technologies is qualified by acknowledgement of outstanding problems, becomes justification for additional investments in research. We close with an analysis of DoD aspirations based on a subset of Defense Advanced Research Projects Agency (DARPA) grant solicitations for the use of AI in battlefield applications. Taken together, we argue that grant solicitations work as a vehicle for the mutual enlistment of DoD funding agencies and the academic AI research community in setting research agendas. The trope of basic research in this context offers shelter from significant moral questions that military applications of one's research would raise, by obscuring the connections that implicate researchers in U.S. militarism.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

2 extracted references · 1 canonical work pages

  1. [1]

    ReplicationDatafor: OneBadNOFO?Federal Grantmakers’ SilenceonAIPolicies

    Abraham, Y(2023)‘Amassassassinationfactory’: InsideIsrael’scalculatedbombingofGaza. +972Magazine, November30.https://www.972mag.com/mass-assassination-factory-israel-calculated-bombing-gaza/ Bateyko, D., &Levy, K. (2024). "ReplicationDatafor: OneBadNOFO?Federal Grantmakers’ SilenceonAIPolicies", https://doi.org/10.7910/DVN/LPK0IN Boyd-Graber, J., Mimno, D...

    work page arXiv 2023

  2. [2023]

    AIsupplychain

    https://www.sipri.org/databases/milex Suchman, L. A. (2007). Human-machinereconfigurations:Plansandsituatedactions(2nded).CambridgeUniversityPress. Suchman, L(2023)Imaginariesofomniscience: AutomatingintelligenceintheUSDepartmentofDefense. Social StudiesofScience, 53(5), 761–786. Suchman, L(2024)TheAlgorithmicallyAcceleratedKillingMachine. Guestpost, AINo...

    2007

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.