Artificial intelligence in the defence context: Levels of employment, the DIKW hierarchy and human factors

Australia’s 2026 National Defence Strategy (NDS26) acknowledged this directly, identifying AI as presenting “the most significant potential for technological disruption in the coming years” and as already transforming war and changing the role of human decision making.

Understanding where AI is most productive – and where human judgement remains indispensable – requires mapping its application against both the levels of warfare and the data–information–knowledge–wisdom (DIKW) hierarchy.

Underpinning all of this is a sovereign dimension: assured access to AI systems, the data they depend on, and the infrastructure they run on is not merely a technical matter but a prerequisite for strategic self-reliance.

The weapons system level

At the weapons system level, AI performs the first and most fundamental step in the DIKW hierarchy: transforming raw data into information.

Sensors embedded in individual platforms generate signals – radar returns, acoustic signatures, infrared emissions, electronic fingerprints – that AI algorithms sort, classify, and interpret in real time at speeds and volumes beyond any direct human involvement.

VIEW ALL

Examples range from the signal processors in radar systems that classify returns as threats or non-threats to the acoustic sensors in torpedoes that match vessel signatures against trained parameters, to the engagement logic in close-in weapon systems that track and engage incoming threats within milliseconds.

Reports surrounding Iran’s deployment of highly autonomous drones – including systems alleged to have selected and engaged targets with minimal or no real-time human input – represent the upper end of this level, where AI has assumed not just the data processing function but aspects of the engagement decision itself, a development that generated significant international concern.

NDS26 explicitly noted the disproportionate battlefield impact of autonomous and uncrewed systems.

A less visible but analytically significant dimension is the use of AI – including machine learning techniques – during the design and development of individual weapons systems, shaping the decision logic, engagement envelopes, and classification algorithms that the deployed weapon then executes as deterministic code.

The weapon in the field may operate on fixed rules, but those rules may themselves be the product of a training process rather than explicit human engineering. This has direct implications for Australia’s Responsible AI Policy requirement that the function and relationship between inputs and outputs of AI technologies must be traceable: a rule set derived from machine learning may satisfy output-level traceability while remaining less transparent at the level of its origins and the training process that produced it.

The policy also requires that a named accountable officer bears personal responsibility for AI decisions and outcomes at each stage of the technology life cycle – a requirement that the ML-in-development dimension makes more complex to operationalise.

Tactical employment

At the tactical level, AI converts information into knowledge: synthesising multi-sensor feeds, friendly force tracking, and threat assessments into an integrated operational picture that enables small-unit commanders to make faster, better-informed decisions.

Israel’s AI targeting tools Gospel and Lavender, refined through the Gaza and Lebanon campaigns and deployed at scale during the 2026 Iran campaign, exemplify this conversion: they ingest and correlate information from multiple sources – signals intelligence, imagery, pattern-of-life analysis – and nominate targets for human approval.

Ukraine similarly uses AI to process more than 50,000 raw video feeds from the front line each month, converting distributed sensor data into a structured, actionable targeting picture – an information product that feeds directly into tactical decision making. These are not weapons systems; they are targeting support systems operating across a theatre, processing data and converting information into knowledge at machine speed.

The US Maven Smart System, deployed across CENTCOM operations during the Iran campaign in early 2026, reportedly enabled highly compressed operational tempos, contributing to debate over whether effective human supervision can be sustained under such conditions. NDS26 draws on an Australian Defence Force operational analysis team deployed to Ukraine, incorporating these lessons into Australia’s strategy of denial.

Human factors are a significant issue at this level: automation bias – the tendency to defer uncritically to algorithmic outputs – is a known and serious risk that compounds as tempo increases, and NDS26 explicitly identifies AI literacy as a future ADF workforce requirement.

Operational level

Operational commanders transform knowledge into campaign coherence. In DIKW terms, the operational level sits at the knowledge-to-wisdom transition.

AI contributes most effectively at the knowledge end – synthesising intelligence products, correlating logistics models, and optimising force allocation – but campaign design requires the kind of contextual judgement, appreciation of political and cultural constraints, and tolerance for irreducible uncertainty that characterises wisdom.

This is precisely where human judgement becomes indispensable: not merely as a check on machine outputs, but as the human capacity to reconcile military effectiveness with political intent, alliance management, and uncertainty.

NDS26 specifically cited Palantir’s AI-enabled targeting as having provided compressed targeting cycles in Ukraine. Coalition management, legal compliance, deception planning, and the navigation of political constraints each require the appreciation of strategic culture, adversary personality, and political constraint that no current AI system can reliably provide.

AI is a powerful support tool at this level; the judgement that gives it direction and purpose remains fundamentally human. AI may also enhance operational and strategic decision making by expanding the volume of information senior commanders can meaningfully interrogate, even where final judgement remains human.

Strategic level

At the strategic level, AI enables senior leaders to integrate knowledge towards wisdom – aligning intelligence assessments, modelling adversary decision making, and optimising force generation with political goals across extended time horizons. NDS26 identified China’s AI-enabled military systems as a primary driver of Indo-Pacific security dynamics and called out Beijing’s continued research and development across AI, quantum technology, and advanced semiconductors as a compounding strategic challenge.

Australia’s Advanced Strategic Capabilities Accelerator (ASCA) and AUKUS Pillar 2 advanced capabilities are the primary domestic vehicles for strategic AI investment. Strategic wargaming exercises – sometimes themselves AI-assisted – have raised concern that under some conditions, AI reasoning models may favour escalatory responses, making human strategic judgement not merely important but irreplaceable.

The political, cultural, and ethical weight of strategic decisions demands a quality of wisdom that AI can inform but cannot supply.

Grand strategic level

Grand strategy concerns the integration of all instruments of national power – diplomatic, informational, military, and economic – over extended time horizons and broad geographic regions. Here, wisdom in the DIKW sense is paramount. NDS26 broadened the concept of National Defence to encompass economic security, national civil preparedness, and integrated statecraft – domains where AI can surface patterns but cannot substitute for political leadership or democratic accountability.

The Lowy Institute has observed, in an independent assessment, that NDS26’s treatment of AI-enabled cognitive warfare amounts to little more than a gesture towards strategic communications, despite this domain having been central to both the Ukraine and Iran conflicts.

AI can assist grand strategic analysis – surfacing patterns in adversary behaviour, modelling influence networks, and supporting strategic communications – but the judgements about values, legitimacy, and political will that ultimately shape grand strategy remain fundamentally human. Human judgement is dominant and non-delegable at this level: the considerations of strategic culture, alliance relationships, domestic political constraints, and the adversary’s capacity and will to endure that shape grand strategy lie beyond the reach of current AI systems.

Sovereign AI: Data, protection, and assured access

Running across all levels of warfare is a dimension NDS26 addresses in general terms but does not resolve for AI specifically: sovereign capability.

NDS26 defines self-reliance not as self-sufficiency but as the ability to distinguish what Australia can do independently from what can be assured through trusted partnerships – and to invest in a domestic industrial base capable of producing, adapting, and sustaining the capabilities needed in conflict. For AI, this framing is necessary but insufficient.

Sovereign AI in the defence context requires three things that go beyond platform acquisition. First, access to training data: AI systems are only as capable as the operational and environmental data on which they are trained. Dependence on allied data sets introduces both classification constraints and the risk that Australian operational patterns are shaped by others’ priorities.

Second, protection of that data: NDS26 acknowledges that espionage and foreign interference are already at extreme levels, that authoritarian regimes are increasingly capable of disrupting critical digital infrastructure, and that seabed communications cables remain vulnerable to sabotage. An AI system whose training data, model weights, or inference infrastructure can be compromised or degraded is a strategic liability.

Third, assured access to the systems themselves: the 2026 Iran conflict illustrated both the operational importance of AI-enabled systems and the vulnerability created by dependence on commercially provided AI services, including the risk that access could be constrained by political or regulatory developments. Israeli defence officials drew the explicit lesson, noting publicly that Israel needed its own AI systems precisely to avoid the moment when a commercial provider is “taken away” by political events in San Francisco.

Australia’s current sovereign AI strategy – focused on signals intelligence, autonomous maritime surveillance, electronic warfare, and logistics optimisation through ASCA and Project REDSPICE, which encompasses signals intelligence and cyber capabilities with significant AI components – is well-directed but faces material constraints.

Specialised AI hardware remains largely manufactured overseas, creating supply chain vulnerability that mirrors the munitions dependency NDS26 addresses elsewhere. The talent pipeline in Australia is thin, security clearance timelines are slow, and Australia lacks a dedicated AI testing range at operational scale. NDS26’s commitment that “off-the-shelf procurement no longer offers a guarantee of speed to capability” applies with particular force to AI, where the technology evolves in months while procurement cycles historically span years.

Conclusion

AI’s defence utility is highest where the volume and velocity of data overwhelm human cognition: sensor fusion, target nomination, logistics routing, and intelligence triage.

As decisions ascend through tactical, operational, strategic, and grand strategic levels, the DIKW requirements shift towards knowledge and wisdom, human judgement intensifies, and the consequences of error extend beyond the immediate engagement – potentially affecting campaign outcomes, alliance relationships, and strategic stability in ways that may be irreversible.

NDS26 acknowledged AI’s transformative significance. The 2025 Responsible AI Policy for Defence established an important governance foundation – including accountable officers, values-based principles, proportionate controls, and Article 36 review requirements – but the operational doctrine for AI employment across the levels of war remains underdeveloped relative to the investment being committed.

Most critically, the document does not yet address sovereign AI in its full dimensions – data access, data protection, and assured operational availability – as discrete strategic requirements.

Bridging that gap is the most pressing near-term challenge for Australian defence AI policy, and one that the next NDS cycle should prioritise.

John Harvey is a former Air Marshal in the RAAF and has a PhD in computer science from UNSW Canberra. His postings have included Chief Capability Development Group, F-35 project manager, director Military Strategy and director Air Power Studies Centre.