by Dr. Mie Augier and Dr. Jamie Porchia

Introduction

The Department of War (DoW) is seeking a responsible AI-ready force and DoW educational institutions play a role in ensuring the DoW can meet this objective. As AI continues to evolve and get adopted into warfighting and our warfare organizations, including in our educational institutions, education of warfighters can influence both the short-term skills and traits of warfighters (making them warrior-scholars comfortable with AI across domains and regardless of careers), as well as the longer-term adoption of AI in our military institutions. At the same time, important discussions are happening about the roles, potentials and risks of AI, and the complementarities of machine intelligence and human judgment.¹

The aim of this paper is to provide some suggestions for how educators can incorporate AI to educate warrior-scholars, in the context of the potential (and risks) of AI use. Building on research and findings from other disciplines we present some of the challenges AI brings to education and learning, before suggesting how incorporating AI into the classroom can help mitigate these challenges and build more AI ready warriors.² The next section introduces a few promises and pitfalls of AI when it comes to learning (and the centrality of cognitive engagement), before offering a few suggestions for embracing teaching with AI.

 AI and Learning: The centrality of Cognitive engagement

AI has become an essential part of most organizations, across different sectors and industries. The 2025 McKinsey report on the State of AI notes that AI is continuing to be integrated into organizational processes, with more than 80 % responding that their organization use AI for at least one business function, also noting increased curiosity when experimenting with different AI agents. At the same time, they also report increasing concerns with AI related risks, such as inaccuracy and regulatory compliance.³

One well known risk / vulnerability associated with AI is hallucinations. AI’s reputation for producing hallucinations has generated skepticism resulting from several well-known examples of faulty outputs. Within the legal profession, an attorney who used ChatGPT for legal help cited a set of cases, as lawyers do, complete with citations, quotes, and convincing legal reasoning. The cases and data, however, did not exist! This has led the legal profession to recommend safeguards such as manual reviews of legal documents.⁴ Researchers from Stanford’s Human Centered AI found that LLM’s produce hallucinations in more than 70 percent of the time (dependent on task).⁵

But hallucination is not the only risk. A subtler and perhaps more consequential danger is what researchers across disciplines have called ‘cognitive offloading’ – the delegation of cognitive effort to AI in ways that reduce active engagement, reflection, thinking, and learning. For example, within higher education, while having potential to improve efficiency in teaching and generating material, the utilization of AI can also lead to ‘meta cognitive laziness’ in both teachers and students.⁶ One study found that people using AI in writing tasks risk ‘cognitive laziness’ due to lack of thinking and engagement when using LLM’s for writing, raising essential questions for academic institutions about the integration of AI into research and education and arguing for explicit strategies for ensuring active engagement to mitigate the potential detrimental effects on learning.⁷ Another study on AI in the educational realm argues for bridging cognitive load and the engagement of learners and the need for active roles of teachers to ensure effective learning  (knowledge acquisition, retention and application) as AI by itself can lead to lower knowledge retention for the learners.⁸ A recent study on STEM education has also found risks such as mistaking learning with short term performance, and ignoring the need for active and reflective engagement needed for learning in the STEM disciplines.⁹

A recent paper from the medical profession has warned against both cognitive offloading as well as more specific dangers of never-skilling, mis-skilling and de-skilling.¹ In particular, offloading complex decision making and tasks they found can lead to failure to develop essential skills (never-skilling), enforcement of incorrect behavior due to AI use (mis-skilling), and loss of previously developed skills (deskilling). Thus, across disciplines and professions, it is essential for educators and educational institutions to strategically employ AI in education in ways that help develop essential skills of problem solving with AI, while mitigating the risks (such as the lack of deep learning and the off-loading of essential skills such as critical thinking).

Table 1 captures some of the specific risks that researchers have found, along with manifestations / vulnerabilities particularly relevant in a military context:

Table 1: Cognitive Offloading and Skilling risks in AI-enabled learning:

Risk Dimension / Vulnerability	Underlying mechanism	Manifestation in other disciplines / professions	Potential vulnerability manifestation in warfighter context
Cognitive offloading / meta cognitive laziness	When delegation of cognitive effort to AI reduces active engagement, effort, reflection and thinking	Students relying on AI generated summaries or analysis, reducing deep learning and retentions; practitioners deferring to AI generated output, relying on AI for writing and analysis resulting in cognitive engagement	Warfighters relying on AI for analysis or planning, reducing engagement and experience with ambiguity, uncertainty, and complexity in operational problems leading to lack of judgment and critical thinking skills
Never skilling	When substitution of core learning prevents development of core competencies	Students never develop full problem solving, writing or diagnostic reasoning skills due to early reliance on AI; professionals relying on AI tools before building expertise in field / domain	Embedding AI into initial skills and other related training before sufficient competency without AI has been achieved
Mis skilling	Shifts in skills development towards tool interaction instead of reasoning and judgment	Learners becoming proficient at prompting or reading AI outputs without understanding underlying mechanisms; professionals optimize for tool use instead of reasoning	Warfighters use an AI tool to help accomplish their tasks, but they have limited understanding of what the output of the tool means and how to utilize its findings
Deskilling	When previously acquired skills atrophy due to repeated reliance on AI	Reduced ability to diagnose problems; loosing ability to perform core analytic or writing tasks	AI eroding skills that warfighters used to have, leading to disadvantages in conflict when those essential (e.g. old school navigation)

Researchers have suggested cognitive engagement be a critical component to reduce the risks. Educators can help facilitate this by using active learning approaches to ensure cognitive engagement and thinking with AI, not outsourcing the essential skills. How, then, do we both embrace AI in our classrooms (and beyond), and cultivate curiosity about the potential as well as its limitations? Below are a few suggestions from our own recent experiences.

Experimenting with AI in the classroom

Experimentation with AI in the classroom involves commitment from instructors to examine how best to incorporate AI considering the subject, learning modality, and learning objectives. For example, half hazard applications of AI to unstructured learning objectives will undoubtedly lead to poor outcomes. Yet, designing activities that can both bring about experiences with AI as well as help mitigate some of the risks is possible. Co-learning with AI systems offers individual level learning benefits such as utilizing AI to examine an existing situation to offer new observations or identifying decisions that may have been overlooked, both with the intent to prompt ideas that may have not been considered before.

A few ways that we have experimented with introducing AI in the classrooms (with an eye for learning and engagement) are: teaching cases with AI content; experimenting with different AI tools; and experimenting with different usage levels of AI. A few words on each:

AI related cases

One way to bring AI into the classroom applicable to a variety of different disciplines and courses is to teach AI related cases. Case teaching in general has been embraced by warrior scholars such as General Gray and is a form of active learning which encourages students to think through uncertain and ambiguous situations.  In addition to teaching cases relating to adoption of AI within military organizations, one can bring in other examples of cases to discuss issues, such as, what are enablers / barriers do students see with adopting AI within their own organizations? What have been the challenges to adopting technology in military organizations in the past and how might we learn from those histories? How can they overcome resistances and barriers?

By thinking through some of the enablers and barriers to adoption of technologies in different organizations, students develop familiarity and analytic understanding with the concepts and theories used, as well as development of strategic thinking skills when facilitating a discussion of the similarities, dissimilarities and analogies between the organization discussed, and their own. In addition, having been part of thinking through problems of AI adoption, they might be less likely to be part of the resistances in their own organizations later.

Using different AI tools

Another way to bring AI into classroom is more direct; less about thinking about AI, and more thinking with AI. Students have varying levels of comfort and understanding of AI but providing students with a space to challenge their own assumptions and explore the bounds of AI capabilities is one of the advantages of incorporating AI into the classroom. We crafted an activity to help students learn how to iterate with AI and do so across different AI tools. For the activity, we developed weekly prompts for the students to explore, utilizing a different AI tool each week. The students were required to engage in inquiry with the tool at least 5 times to challenge not only the AI tools responses, but their own assumptions based on the feedback from their selected AI tool. Further learning and discussion were stimulated when the students presented their findings to the rest of the class each week. The cross section of identifying strengths and limitations of each tool boosted the student’s confidence in knowing which AI tool would be the most effective for a given task.  The collective learning across AI systems emphasizes the dynamic learning that can be achieved when incorporating AI into the classroom.  By conducting an activity such as this, the risk of cognitive offloading is reduced as students are being asked to employ judgement on the outputs of the AI tool and to iterate with the tool based on their assessment of the accuracy and usefulness of the tool.

Different usage levels

An additional way to introduce AI – also to learn with, as well as about — is to structure discussions around small experiments in different groups in class, with different usage levels mandated to solve the same task or problem. For example, after a general discussion about concepts related to one’s core learning objectives, teachers can lead a discussion with small groups that first spend time in the groups to solve a problem with different levels of AI use (no AI, a little AI, and high AI). It is important that the groups don’t ‘cheat’ and stick to their assigned usage level. One group member is tasked with developing ‘think aloud’ notes on the group’s problem solving. Each group debriefs the results.

A discussion after the experiment can include not just the ’what’ that was solved; but also reflecting on ‘how’. How did the interaction with AI’s influence how students solved the problem? What learning was done by humans, and what by machines? Depending on the topic of the class, one can also include topics such as, what was the most creative or innovative solution? (surprise: it is usually not the high-AI use groups). Experimentation with different usage levels also helps develop an understanding of the usefulness of AI – not as a universal fix but as a tool which we need to understand both the potential and limitations thereof.

As research has found, excessive use of AI has some negative effects such as increasing digital dependencies and loss of higher-level thinking and learning skills. A study by MIT’s Media Lab found that excessive use of AI for writing can lead to weaker memory recall and coined the term “cognitive debt” for when weaker neural engagement and memory encoding result from LLM use.¹¹ Teachers can bring in such external research as part of the discussion, helping students learn the broader strengths and weaknesses of the digital tools they used.

Using AI to ‘red team’ our thinking + adversarial decision making

In addition, we can bring AI into the classroom to help do adversarial decision making (red-teaming both our own analysis as well as helping us develop realistic competitor thinking). This requires a more active use of AI more as a ‘cognitive partner’, but one that we don’t blindly adopt suggestions from; rather, engage / iterate over time, using our AI partner for what it is good for (information processing of large amounts of historical data), but engaging human thinking for developing judgment. When the roles within the human-AI cognitive partnership are clear, activities that utilize AI in this manner can minimize mis skilling and cognitive offloading by asking students reasoning and judgment related questions, while letting AI support the human with a synthesis of pertinent data and historical context that frames the activity.

For example, after an analytic discussion or activity (human thinking only), students can use AI to red-team their own thinking, surface assumptions in their own analysis, thus using AI to help mitigate common cognitive biases such as group thinking and status quo thinking.

Another example is leveraging AI to capture some of the many differences in thinking between us and competitors. We know for instance that the difference between western and eastern thinking is stark; asking AI to build that into thinking through decision making scenarios with partners and adversaries can help capture some of the challenges to getting stuck in thinking that ‘everyone thinks like us’. Although one must be particularly attentive here to possibility of AI generated hallucinations.

The centrality of reflection

Central to all these ways to bring AI into the classroom is to have a specific time for reflecting and discussing the student’s experiences with AI. Without explicit reflections, we risk students focusing on just the use or the technology part of the class, not the understanding of the ‘why’ and ‘how’ behind it. For example, in a class experimenting with different kinds of AI, teachers can include discussion on why there are differences in the results and what that could mean in a warfighting context (e.g. targeting, intel gathering, using AI for optimizing workflow), which would also help students learn not just the ‘what’.

Table 2 captures some of the dimensions and learnings from these different types of AI activities in the classroom:

Table 2: AI in the classroom, examples

Dimension	AI as object of study	Learning with AI	AI as ‘cognitive partner’
Core activity	Cases on AI Adoption; analysis of failures	Using multiple AI tools; different usage levels in tasks with AI	Red teaming; generating counterarguments to challenge assumptions; adversarial decision making
Learning mechanism	Sensemaking; Causal reasoning	Iterative thinking; critical thinking; judgment	Metacognition; judgment; thinking critically
Educational value	Understanding AI limitations and the organizational challenges implementing AI	Developing practical AI interaction skills and awareness of the potential and limitations of AI	Enhancing decision quality and surfacing biases with active engagement
Human-AI relationship	Indirect / conceptual	Interactive tool use	Collaborative cognitive partnership
Key risk & Mitigation	Can be abstract (but structured discussions can mitigate)	Cognitive offloading risks with high usage level (mitigated with required reflection and discussion)	Overreliance / artificial certainty risk (mitigated with structured challenges and judgment)

Closing

AI is here to stay, and it is shaping the world and way of warfighters, warfighting, and warfighting organizations. Educators in military educational institutions can help incorporating AI into the classroom, not just the technical aspects of AI, but building in human – machine interaction and collaboration as part of their courses and curriculum.

Students in military educational institutions sit at a unique intersection between academic research and warfighter problems. The experiments we have described – teaching AI as case study subjects; learning to work across multiple AI tools and varying usage levels and using AI as a cognitive partner – are a few starting points for building capacity to use AI as well as also knowing when not to use it. Used deliberately, such experiments can help develop:

• An understanding of human-machine complementarity, sharpened by exposing students to varied perspectives and disciplines.
• A clearer sense of the organizational structures needed to help unlock AI’s value (adoption is not easy, and by itself doesn’t create that capability).
• Interdisciplinary problem-solving skills and communication skills that translate across different warfighting domains and organizations.
• Comfort with experimentation itself, which often is a real barrier to adoption later in one’s career.

It is worth recognizing that none of this replaces judgment with automation, rather it’s complementary. The warrior scholars who learn to think with AI, rather than defer to it, will be the ones equipped to lead the AI-ready force in the capacity that the DoW needs.

Dr. Mie Augier is a professor at the Naval Postgraduate School.

Dr. Jamie Porchia previously served as a faculty member at Naval Postgraduate School. She currently serves as the Commander of Defense Contract Management Agency Korea.

Notes

1. Patrick Tucker, “The Real Danger of Military AI Isn’t Killer Robots; It’s Worse Human Judgement,” Defense One, March 25, 2026, (https://www.defenseone.com/technology/2026/03/military-ai-troops-judgement/412390/).

2. Thus we are focusing on how to use AI into the intellectual parts of the classroom, not to develop syllabi, rubrics and other matters of course development.

3. Dan Tinkoff, Lieven Van der Veken, Michael Chui, and Tara Balakrishnan, “The State of AI in 2026: On the Road to ROI,” McKinsey & Company, August 25, 2026, (https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai).

4. William A. Ryan, Allen Garrett, and Brad Sears, “Practical Lessons from the Attorney AI Missteps in Mata v. Avianca,” Association of Corporate Counsel (ACC), August 8, 2023, (https://www.acc.com/resource-library/practical-lessons-attorney-ai-missteps-mata-v-avianca).

5. Matthew Dahl, Varun Magesh, Mirac Suzgun, and Daniel E. Ho, “Hallucinating Law: Legal Mistakes with Large Language Models Are Pervasive,” Stanford Institute for Human-Centered Artificial Intelligence (HAI), January 11, 2024, (https://hai.stanford.edu/news/hallucinating-law-legal-mistakes-large-language-models-are-pervasive).

6. Fan, Y, et al. “Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance,” British Journal of Educational Technology, 56(2), (2025), 489-530.

7. Georgios P. Georgiou, “ChatGPT Produces More ‘Lazy’ Thinkers: Evidence of Cognitive Engagement Decline,” arXiv preprint, arXiv:2507.00181, submitted June 30, 2025, (https://arxiv.org/abs/2507.00181).

8. Meng, N., Mat Deli, M., & Abdul Rauf, U. A. (2025). Educational Technology and AI: Bridging Cognitive Load and Learner Engagement for Effective Learning. Sage Open, 15(4), 21582440251395930.

9. “Learning against the Machine: The double edged short of (Gen) AI in STEM education. International journal of STEM education, Dec 1, 2025.

10. Abdulnour, R. E. E., Gin, B., & Boscardin, C. K. (2025). Educational strategies for clinical supervision of artificial intelligence use. New England Journal of Medicine, 393(8), 786-797.

11. Kosmyna, et al, (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv preprint arXiv:2506.08872, 4.

Featured image: The U.S. Army Network Enterprise Technology Command’s (NETCOM) Data Science Directorate (DSD) demonstrate how AI tools could help develop applications and address challenges. (U.S. Army photo by Gabriel Archer)