Home : News : Display
Dec. 17, 2025

Responsibly Pursuing Generative Artificial Intelligence (GenAI) for the War Fighter

A. Blair Wilcox and C. Anthony Pfaff
©2025 C. Anthony Pfaff

ABSTRACT: This special commentary argues that the US Army must adopt a sober and methodical approach to integrating GenAI into military decision-making processes. Drawing parallels to the historical introduction of tanks and airplanes, the authors caution against both underutilizing the technology and being misled by oversold capabilities. Using the GenAI system Donovan as a case study, the commentary highlights critical deficits in current systems, such as a lack of computational and geospatial reasoning and points to procurement challenges that hinder improvement. The authors contend that simply using GenAI to speed up legacy processes will waste its potential. Instead, they advocate for using war gaming and experimentation within professional military education as a stress test to define requirements properly, manage cognitive loads on personnel, and guide the private sector to develop solutions that are truly aligned with the war fighter’s needs, ultimately enhancing military decision making.

Keywords: generative artificial intelligence, GenAI, technology, Center for Strategic Leadership (CSL), Donovan

 

As the Army begins to adopt artificial intelligence into nearly every aspect of its operations, it is worth considering several lessons from integrating previous disruptive technologies.1 Both the tank and the airplane were initially used in roles that did not fully realize their potential. Tanks, for example, could not exploit their superior mobility because they were initially assigned to infantry, who, in combat, moved only as fast as soldiers could walk.2 Aircraft were relegated to reconnaissance roles because commanders were skeptical of their offensive capabilities. Even military aviation pioneer Brigadier General Billy Mitchell originally considered aviation subordinate to the Signal Corps.3

On the other hand, both technologies were also oversold. During the interwar years, British theorists Sir Basil Liddell Hart and J. F. C. Fuller believed the tank would make infantry largely obsolete.4 Italian Air Marshal Giulio Douhet famously argued that bombers could force an enemy to surrender without the necessity of costly ground battles.5 In the end, these technologies found their own way, in a sense, as militaries experimented, innovated, and most importantly, developed doctrine on how best to use them.

In both cases, advocates extrapolated too much from the innovation’s novelty without fully considering the doctrinal, organizational, and logistical resources required to make full use of it. Their oversell also reflected the fact that advocates often viewed the innovation as a solution to the fog and friction of war without considering how it might create fog and friction of its own.6 As militaries gained experience with these weapon systems, they learned that tanks and aircraft demanded an extensive support system, increasing the potential for friction and a new doctrine—not so much to reduce the fog of war, but to create it for the other side.7 For generative artificial intelligence (GenAI), the lesson is that early claims about disruptive, war-winning effects are likely to mislead practitioners unless paired with a sober, methodical understanding of its effects on doctrine and resourcing.

The integration of GenAI is at a similar point to the early days of armored and air warfare. Recent reports indicate that even where the technology has been adopted, the US military has not yet fully optimized its use. According to a report by Georgetown’s Center for Security and Emerging Technology, AI integration by the XVIII Airborne Corps enabled a targeting process that, during Operation Iraqi Freedom, drastically reduced the number of humans required.8 Despite this fact, those organizational efficiencies have not yet been reflected in doctrine or adjustments in the Corps Fire Support Element’s Army-wide. In another case, as a recent Center for Strategic and International Studies report observes, the failure to invest in computational infrastructure and develop AI-literate personnel risks fragmenting legacy processes and increasing America’s vulnerability to adversary attack.9

At the same time, AI technologies are often oversold. For example, the GenAI system Donovan advertises that it can “put advanced warfighting capabilities in the hands of warfighters when and where they need it the most.”10 However, when war gamers at the Center for Strategic Leadership (CSL) integrated it into a weeklong, secret-level, operational war game in late May 2025, they identified three critical deficits in the system’s functioning that limited its ability to perform certain essential planning functions, including a lack of computational and geospatial capabilities critical to military planning and a procurement system that made it difficult for the vendor to adapt to observed shortcomings.

Donovan, like most GenAI systems, returns probabilistic results and lacks an innate capability to perform computations. So, if asked to calculate movement times or other critical planning information, it would use its database to guess an answer rather than perform the calculations itself. It also lacked geospatial reasoning capabilities, making it even less reliable when calculating, for example, movement tables. Further, because it lacked the memory necessary to recall previous iterations without directed management, its ability to learn from user interactions was limited.11

Unfortunately, the procurement system limits what a vendor can do to address any concerns. At a September 2025 Chief Data and Artificial Intelligence Office (CDAO) defense conference in Washington, DC, vendors and industry leaders alike recognized the difficulty of obtaining valid, operational-level feedback to improve their products.12 In this case, the Global Integration Dominance Experiment (GIDE), which funds Donovan and other vendor licenses, did not enable the government to fund further development to address the identified issues. As a result, Donovan’s performance did not improve. In such a procurement environment, operational units, like the XVIII Airborne Corps, are compelled to continue experimenting with systems insufficient for the required tasks and publish results in hopes of attracting vendors who claim valid solutions. Currently, the CSL is exploring innovative funding pathways through Cooperative Research and Development Agreements (CRADAs), the Small Business Innovation Research (SBIR) program, and existing Army contract vehicles to promote experimentation and development aligned with Army capabilities.

The difficulty with integrating disruptive technologies is finding ways to make them sustainable. Unfortunately, sustaining innovations are those that improve performance in warfighting functions that have traditionally been valued. However, disruptive innovations are those that improve performance in a war-fighting function not previously valued.13 This issue is clearly the case with GenAI. If GenAI is only used to speed legacy processes, it will remain underutilized. Finding which other functions GenAI can bring value to will require experimentation and innovation at lower levels—much as Billy Mitchell did for airpower—that inform institutional reform.

Getting to that point will require senior leaders and acquisition professionals to improve their engagement with the private sector to align its technologies with Army needs. Programs like “FUZE,” which enable services to integrate current technologies more rapidly to suit Army requirements, are a good start.14 However, as the Donovan example illustrates, those involved in the procurement and acquisition processes must understand the capabilities the technology needs to achieve the desired results: If you want to create movement tables, for example, your AI assistant must perform the necessary calculations.

Additionally, they must understand how commercial incentive structures impact AI design. Doing so requires keeping sight of the strategic concept for AI integration amidst the glamour of the perceived technical superiority.15 Commercial incentives prioritize the broadest applications for cost control, and the result is that the military is the recipient of the same GenAI models that high schoolers use to write school reports. Such “dual-use” approaches, in which the military adapts civilian applications to its needs, will not likely meet all military requirements.16

This point is not to say that the military should not utilize civilian applications. Often, they are capable enough that it is the most economical path to military acquisition. However, those responsible for its acquisition, procurement, and integration need to understand enough about how the technology will be used to assess its suitability and sustainability. For example, the military requires highly trained models essential to enhancing the cognitive performance of command and staff organizations in managing violence.17 To optimize those processes, Army leaders need an understanding of how AI technologies redistribute cognitive load and, in turn, the capabilities required to bear it. One does not go from requiring many brains to just a few without changing how those brains think and are organized. The XVIII Airborne Corps personnel involved in AI experimentation and application refer to this redistribution as “fighting with the data,” rather than “with the weapons,” as characterized by the legacy practice.

The Army and Joint Force, therefore, must build and leverage procurement systems that incentivize private-sector expertise to produce what the war fighter needs across all echelons of command. One way to achieve this goal is to invest in war-gaming research and development within professional military education.18 In the cognitive realm, these environments expose systems to human-level, operational evaluation. The evolution of AI systems within Army staff structures will require individual tailoring to prevent cognitive heuristic errors.19 Models must facilitate human ownership of plans and processes by accepting uncertainty, being transparent about data sources, and managing cognitive loads to optimize human and machine contributions. This “stress test” is needed to ensure AI solutions meet the needs of the war fighter, not the other way around. As General Dwight D. Eisenhower once pointed out, “In preparing for battle I have always found that plans are useless, but planning is indispensable.”20 The implication for AI systems is that if the act of planning is valuable to humans, removing them from key elements of the process will reduce its benefits.21

To promote those benefits, the CSL continues to test existing solutions in theater-level war games to first, evaluate their performance in an operational setting; and second, better define what the Army needs at echelons above corps to facilitate decision making. In spring 2026, CSL will lead a strategic-level war game for the newly established Western Hemisphere Command. This game will enable a newly formed headquarters to defend the homeland and establish the standards by which GenAI systems are evaluated.

Empowering continued experimentation in professional military education is an important adaptation in procurement. In the cognitive dimension of warfare, PME retains an advantage in understanding the proper conceptual integration for cognitive enhancement across Army and Joint planning processes to promote technical solutions.22 Up to this point, the Army has been overly focused on technical pageantry rather than on demonstrated performance aligned with how the Army fights. GenAI offers potential, but in its current form, it is ill-suited for enhancing senior leader cognition at the strategic level.

The US industrial base is prepared to respond, and the Army has the capacity and venue to empower Silicon Valley to help the war fighter maintain the edge in the cognition dimension of warfare. GenAI integration will find its own way in the Army, much as the tank and the airplane did. Unlike the tank and the aircraft, however, the stakes are potentially higher, as these new systems interact with almost every Army system and impact the Army’s greatest strength—how leaders think and make decisions. The obligation now lies with the Army to regain the narrative and empower the commercial sector to enhance American decision making and lethality through war gaming.

 
 

Disclaimer: This abstract was written, with light editing from an author, by Gemini, accessed through the new Department of War GenAi website.

 
 

A. Blair Wilcox
Lieutenant Colonel A. Blair Wilcox is the deputy director of the Strategic Landpower and Futures Group and assistant professor of concepts and doctrine at the US Army War College.

C. Anthony Pfaff
Dr. C. Anthony Pfaff is currently the director of the Strategic Studies Institute at the US Army War College, where he was formerly the research professor for strategy, the military profession, and ethic. He has written widely on the integration of artificial intelligence and other disruptive technologies, focusing on strategic and ethical implications.

 
 

Endnotes

  1. 1. Alexandra Kelley, “Pentagon CTO Wants AI on Every Desktop in 6 to 9 Months,” Defense One, September 16, 2025, https://www.defenseone.com/technology/2025/09/pentagon-research-official-wants-have-ai-every-desktop-6-9-months/408155/?oref=d1-featured-river-top&utm_campaign=dfn-ebb&utm_medium=email&utm_source=sailthru.
  2. 2. Allan R. Millet et al., For the Common Defense: A Military History of the United States from 1607 to 2012 (Free Press, 2012), 358–59.
  3. 3. John H. Morrow Jr., The Great War in the Air: Military Aviation from 1909-1921 (University of Alabama Press, 1993), 90.
  4. 4. Bruce Oliver Newsome, Sir Basil Liddell Hart and Tanks (Perseus Publishing, 2024).
  5. 5. Giulio Douhet, The Command of the Air, trans. Dino Ferrari (Air University Press, 2019), x.
  6. 6. Charles A. Pfaff, “Chaos, Complexity, and the Modern Battlefield,” Military Review (July-August 2000), 83–85, https://apps.dtic.mil/sti/tr/pdf/ADA512069.pdf, accessed December 1, 2025. This article observes that one lesson of complex adaptive systems is that adding components, even those intended to provide greater clarity, makes the system more complex, increasing the likelihood of chaotic effects.
  7. 7. C. Anthony Pfaff, “Anticipation in Asymmetric Warfare,” in Handbook of Anticipation: Theoretical and Applied Aspects of the Use of Future in Decision Making, ed. Roberto Poli (Springer, 2019), https://doi.org/10.1007/978-3-319-91554-8_75, 1479–1503, accessed December 1, 2025. This chapter points out that the German blitzkrieg, as an example of a doctrinal battlefield innovation, was designed to increase the fog and friction of war for the enemy rather than reduce it for friendly forces.
  8. 8. Emelia S. Probasco, Building the Tech Coalition (Center for Security and Emerging Technology, 2024), 4.
  9. 9. Benjamin Jensen and Matthew Strohmer, Rethinking the Napoleonic Staff (Center for Strategic and International Studies, 2025), 3.
  10. 10. “Scale Donovan,” Scale, n.d., accessed December 1, 2025, https://scale.com/donovan.
  11. 11. William J. Barry and Aaron “Blair” Wilcox, Centaur in Training: US Army North War Game and Scale AI Integration, Issue Paper 2-25 (Center for Strategic Leadership, 2025), https://media.defense.gov/2025/Aug/05/2003773179/-1/-1/0/CENTAUR%20IN%20TRAINING%20ISSUE%20PAPER_2025%2006-26_MD.PDF, n.d., accessed September 25, 2025.
  12. 12. CDAO Defense Conference 2025, Washington, DC, September 16–17, 2025.
  13. 13. Terry C. Pierce, Warfighting and Disruptive Technologies (Frank Cass, 2004), 25.
  14. 14. Jen Judson, “Army Adopts Venture Capital Model to Speed Tech to Soldiers,” Defense News, September 15, 2025, https://www.defensenews.com/land/2025/09/15/army-adopts-venture-capital-model-to-speed-tech-to-soldiers/.
  15. 15. William J. Barry and Blair Wilcox, “Neocentaur: A Model for Cognitive Evolution Across the Levels of War,” Modern War Institute, September 5, 2025, https://mwi.westpoint.edu/neocentaur-a-model-for-cognitive-evolution-across-the-levels-of-war/.
  16. 16. “Dual Use Technology,” Global Alliance for ICT and Development, n.d., accessed September 18, 2025, https://www.gaid.org/publications/ai-and-international-security/dual-use-technology.
  17. 17. William Barry and Aaron “Blair” Wilcox, “Hybrid Intelligence: Decision Dominance at the Strategic Level,” War Room, October 17, 2024, https://warroom.armywarcollege.edu/articles/hybrid-intelligence/.
  18. 18. Barry and Wilcox, Centaur in Training.
  19. 19. “Cognitive Biases,” The Decision Lab, n.d., accessed September 26, 2025, https://thedecisionlab.com/biases.
  20. 20. Jason Evanish, “41 Amazing Military Leader Quotes Any Manager Can Learn From,” Lighthouse Blog, https://getlighthouse.com/blog/military-leader-quotes-manager-learn/, n.d., accessed December 1, 2025.
  21. 21. The authors owe this point to an anonymous reviewer.
  22. 22. For an example, see C. Anthony Pfaff and Christopher J. Hickey, Integrating Artificial Intelligence and Machine Learning Technologies into Common Operating Picture and Course of Action Development (US Army War College Press, 2024), https://press.armywarcollege.edu/monographs/980/.
 
 

Disclaimer: Articles, reviews and replies, review essays, and book reviews published in Parameters are unofficial expressions of opinion. The views and opinions expressed in Parameters are those of the authors and are not necessarily those of the Department of War, the Department of the Army, the US Army War College, or any other agency of the US government. The appearance of external hyperlinks does not constitute endorsement by the Department of War of the linked websites or the information, products, or services contained therein. The Department of War does not exercise any editorial, security, or other control over the information readers may find at these locations.