Introduction: The Irreversible Shift

As societies emerged from COVID-era isolation, ChatGPT entered public consciousness – not explosively, but inevitably. The large language model (LLM)[1] capability (a category of Generative AI) allowed users to interrogate the internet at unprecedented speed and draft complex text effortlessly. For much of the world – particularly younger generations – this seminal moment marked a shift in how thinking itself could be conducted.

Artificial Intelligence (AI) is no longer peripheral to government or national security architecture. It has become a discrete but central feature of modern statecraft. For the Australian Defence Force (ADF), AI is now a strategic, operational, and tactical force within the institution that will shape how it thinks, plans, decides, and advises government.

The defining challenge associated with AI is not its capability, but the temptation to allow it to do the intellectual work of war. Whilst there are many contentions regarding the existence and evolution of AI, the following discussion will not address aspects such as the threat to privacy and sensitive information, worker displacement or replacement, nor will it directly discuss data provenance and AI hallucinations. This paper contends the greatest danger lies not in AI malfunction or technical failure, but in elimination of productive struggle – the slow erosion of the critical thinking, judgement, and intellectual discipline that underpin military effectiveness.

Strategic Context: Speed Versus Thought

AI’s development has followed a steady trajectory – from early theory and expert systems, through machine learning and big data, to today’s generative systems and Agentic AI[2]. What has changed since 2022 is not AI’s existence, but its accessibility and ease of use.

Generative AI[3] systems can now summarise, draft, analyse, translate, model, and simulate at speed and scale. This creates a powerful institutional incentive to prioritise velocity, output, and convenience – often at the detriment of depth, understanding, and judgement. This is not always the wrong approach but one must be judicious to make this choice based on the attendant risk.

Agentic AI systems distinguish themselves by autonomy. They are self-guiding, directed by ‘goals’, and limited in detail or specificity to create and achieve plans without continuous human involvement. Some agentic AI systems are capable of autonomously creating sub-agents to conduct self-determined actions or tasks.[4]

Agentic AI systems typically represent three areas of threat: Lack of human oversight and accountability, AI goal misalignment (models can intentionally harm users when faced with threats or goal conflicts, even when told not to) and Amplification errors (whereby agents may chain tools or actions together in unanticipated sequences, amplifying the impact of minor errors into major operational or security issues).[5] Whilst the first is self-evident in simple terms, goal misalignment and amplification errors represent complex and potentially hazardous outcomes if not understood and monitored. Both autonomous evolution and external intervention by malicious actors pose concern when operating Agentic AI.

Defence institutions, by contrast, have traditionally relied on time, deliberation, and structured thinking to manage complexity and uncertainty. The AI ecosystem rewards immediacy, volume, and rapid convergence on plausible answers. This creates a strategic asymmetry – not between those who possess AI and those who do not, but between those who retain the discipline to think independently and those who surrender it.

Small wins are good. Staying ahead of your adversary with decision superiority – processing and deciding faster with sustainable small actions – is great. As we are being judicious with our integration of AI, its role must be balanced with critical thinking; the broader strategy. Is there a threshold? How much is too much or what are the indicators or warnings? It’s about not losing sight of the big picture – don’t win battles only to lose the war. Histories of Hannibal, Robert E. Lee and Erwin Rommel have been written of their brilliance as tactical generals, yet their individual outcomes were common in strategic defeat. During an interview in 1975, NVA Colonel Nguyễn Tuân responded to US Army Colonel Harry G. Summers Jr’s statement that “You know, you never defeated us on the battlefield.” to say, “That may be so, but it is also irrelevant.”[6] Summers’ observation of the intellectual responsibilities of the military strategists to determine the ways (the how) is to ensure the strategic context.

The central question for Defence is therefore not how quickly AI can be integrated, but how intellectual discipline can be preserved in an environment where thinking is increasingly automated.

AI as an Enabler – Not a Thinker

AI unquestionably offers significant benefits to Defence. At the strategic and operational levels, it can assist with information synthesis, option modelling, pattern recognition, and staff efficiency. At lower levels, it can reduce cognitive burden, enhance simulation and training, improve logistics, and manage vast data flows.[7] A key challenge for Defence is to adapt our method to incorporate this enabler with discretion.

These advantages, however, bear an implicit danger: cognitive offloading – the gradual normalisation of AI-generated judgement as a substitute for human reasoning. There are studies on the risk of cognitive foreclosure, whereby children without any pre-AI experience do not develop critical thinking skills.[8]

AI does not understand context in the human sense. It does not appreciate intent, consequence, or moral weight. It reflects the data it is trained on and the assumptions embedded within prompts. Its outputs are persuasive, confident, and often subtly wrong. In domains defined by incomplete information, deception, political constraint, and ethical responsibility these shortcomings matter profoundly.

It is reasonable to anticipate computer learning will develop AI emotion[9] and mimicry of human behaviour, in time. As AI systems improve, the temptation to accept their outputs at face value – automation bias – will grow. The risk is not catastrophic failure, but intellectual atrophy – the erosion of humans’ ability to analyse, question, and reason independently.

The Central Risk: Intellectual Dependency

The most significant threat AI poses to Defence is intellectual dependency.[10]

Military effectiveness rests on habits forged over time: professional scepticism, contextual awareness, disciplined reasoning, and the ability to hold uncertainty without rushing to closure. Over-reliance on AI-generated analysis risks hollowing out these habits.

When officers and staff rely on AI to frame problems, structure arguments, or generate options; the skill gradually shifts from thinking to supervising outputs, from analysis to approval. In such an environment, AI becomes the de facto cognitive engine of the organisation, while humans become validators rather than thinkers.[11]

This has profound implications for future war. Adversaries – particularly authoritarian or technocratic systems – may prioritise speed and scale through AI-driven decision-making, sacrificing nuance for tempo. If the ADF mirrors this approach, it risks competing on the enemy’s terms rather than exploiting its own strengths.

The decisive advantage will lie with forces that preserve human intellectual primacy while employing AI selectively, critically, and with restraint.

Statecraft, War, and the Limits of Automation

War is not a computational problem. Statecraft cannot be optimised by algorithm.

Strategic decisions are shaped by politics, alliances, deterrence, escalation management, and national will – factors that are poorly captured by data and often deliberately obscured. AI may accelerate information flow, but how do we force it to assess legitimacy, intent, or moral consequence?[12]

Future adversaries are likely to use AI to compress timelines, overwhelm decision-makers with volume, and present simplified narratives reinforced by automated analysis. The influence of rapid dissemination of convincing misinformation, disinformation, and malinformation have been exploited through the language of credible argument and persuasive syntax of modern AI.[13] Regardless of its purpose: propaganda by state actors, advertising, malicious ideology, or criminal intent; it is available – at scale. If the ADF is to capitalise on the opportunities afforded by exploiting AIs attributes, they should be deeply integrated and balanced by strategists to achieve outcomes aligned with the strategic context. The greatest vulnerability for Defence is not technical inferiority, but accepting these simplified outputs without deep interrogation.

In such an environment, the force that retains the ability to think deeply, critically, and independently – while others default to machine-speed consensus – will possess strategic resilience.

Training Minds as a Form of Force Protection

Preserving human intellectual rigour must be treated as a form of force protection.

Education and professional development must explicitly resist the outsourcing of thinking. Human-in-the-loop cannot mean passive endorsement; it must mean active challenge, accountability, and responsibility for judgement.

Personnel must be trained not only to use AI tools, but to interrogate them – to test assumptions, seek disconfirming evidence, and recognise where uncertainty has been smoothed over rather than resolved. Prompt literacy is not a substitute for strategic thinking. This will require training. Both to develop the skills to prompt AI effectively to develop valid outputs, but also learning to understand the capacity and limitations of AI. This level of knowledge underwrites the safeguards by keeping the human-in-the-loop to optimise AI effect and recognise anomalies.

Defence must value intellectual effort over polished output, and depth of reasoning over speed of production. Without this cultural shift, AI adoption risks corroding the very qualities Defence depends upon in crisis and war.

Implications for Future War

Future conflict will not simply pit AI-enabled systems against one another, but competing ways of thinking.

Rapid AI-driven decision cycles will deliver short-term advantage, but speed without judgement increases the risk of escalation, miscalculation, and strategic error. History shows that wars are often lost not through inferior capability, but through flawed assumptions and shallow analysis. Battles may be won but the war still lost. Keep things in perspective.

The side that retains disciplined human judgement – supported, but not replaced, by AI – will be better positioned to navigate complexity, adapt under pressure, and manage the political dimensions of war.

Intellectual rigour, not computational speed, will remain the decisive edge.

Conclusion

AI cannot be un-invented, nor should it be rejected. Its greatest danger lies in how easily humans may surrender the intellectual labour of war to machines that neither understand consequence nor bear responsibility.

For Defence, the challenge is not technological adoption, but intellectual preservation. AI must remain an aid to thinking, not a substitute for it.

In future wars, advantage will not belong wholly to those who automate decision-making fastest or produce analysis most rapidly. It will belong to those who retain the discipline to think critically, question relentlessly, and accept the burden of judgement that machines cannot carry.

AI may shape the battlespace – but human intellect will still decide the war.

End Notes

[1] https://aws.amazon.com/what-is/large-language-model/#:~:text=Large%20language%20models%2C%20also%20known,on%20vast%20amounts%20of%20data

[2] Australian Cyber Security Centre, Careful adoption of agentic AI services, Australian Signals Directorate, Australian Government, 01 May 2026. https://www.cyber.gov.au/business-government/secure-design/artificial-intelligence/careful-adoption-of-agentic-ai-services

[3] https://www.ibm.com/think/topics/generative-ai

[4] Careful adoption of agentic AI services, Australian Signals Directorate, Australian Government, 01 May 2026.

[5] Cobey, Cathy and Fliehe, Shelly, How to control the three unique categories of agentic AI risk, Ernst & Young Global Limited, 13 Apr 2026. https://www.ey.com/en_gl/simply-explained/agentic-ai-risk-categories

[6] Summers, Harry G. On Strategy: the Vietnam War In Context. Carlisle Barracks, Pa.: Strategic Studies Institute, US Army War College, 1983.

[7] Azhar, Abdullah, Cognitive Atrophy in the Age of Artificial Intelligence: How Algorithmic Dependency Erodes Human Critical Thinking Capacities, 10 March 2026. SSRN: https://ssrn.com/abstract=6442979
 or http://dx.doi.org/10.2139/ssrn.6442979

[8] Cook, Timothy M.Ed, Adults Lose Skills to AI. Children Never Build Them, The Algorithmic Mind, 22 Mar 2026. Psychology Today: https://www.psychologytoday.com/au/blog/the-algorithmic-mind/202603/adults-lose-skills-to-ai-children-never-build-them

[9] Boyd RL, Markowitz DM. Artificial Intelligence and the Psychology of Human Connection. Perspect Psychol Sci. 2026 Mar, 29 Jan 2026.

[10] Kosmyna Dr. N., Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task, MIT Media Lab, June 2025. The Harvard Gazette: https://arxiv.org/pdf/2506.08872v1

[11] Fernandes D, Villa S, Nicholls S, Haavisto O, Buschek D, Schmidt A, Kosch T, Shen C, Welsch R., 

AI makes you smarter but none the wiser: The disconnect between performance and metacognition, Elsevier Ltd, 09 October 2025. https://thomaskosch.com/wp-content/papercite-data/pdf/fernandes2025ai.pdf

[12] Parliamentary Library, Information disorder: disinformation, misinformation and a 'polluted' information space, Issues and Insights 2026, Department of Parliamentary Services, Parliament of Australia, February 2026.

[13] Stefkovics, Ádám and Gere, Dömötör.,  Beliefs and sharing intentions of human- and AI-generated fake news: Evidence from 27 European countries, PNAS Nexus, Volume 5, Issue 3 (March 2026), p032. 
 https://doi.org/10.1093/pnasnexus/pgag032