Across sensors and platforms, decision-making suites and command centres, artificial intelligence and autonomous systems are not merely enhancing existing capabilities – they are creating entirely new operational possibilities that, until recently, military commanders didn’t realise were achievable. This transformation is moving at a pace that fundamentally challenges how Defence tests, evaluates and deploys new technologies.
The scale of this shift becomes apparent when one considers the information landscape confronting contemporary military organisations.
As Dr Paul Robards, Defence’s chief data integration officer, observes, military commanders have historically struggled with information scarcity. That challenge has inverted entirely.
“Today, it’s a case of information overload,” he says. “There is just information coming to people from everywhere and all our systems generate enormous volumes of data, everything. Satellites, sensors, drones, communications networks and so on.”
This is the genuine problem statement for 21st-century defence operations. The ADF is simultaneously drowning in data and starving for actionable insight. Autonomy and artificial intelligence represent the practical response to this paradox. Yet the integration of these systems carries risks and complexities that extend far beyond simple capability development.
For Andrew Seedhouse, chief space intelligence at Defence Science and Technology Group, the contemporary challenge sits at the intersection of speed and responsibility. His team operates at the frontier of what’s technologically possible, running experiments with systems that didn’t exist months earlier. Yet the innovation cycle he inhabits moves at a profoundly different tempo to the enterprise integration cycle that Robards manages.
“My world is fast, but you’d hope and it does is expect that the testing and then production is slower,” Seedhouse says. “When you’re deploying an information system across a large enterprise, the testing of the implications of that on the whole enterprise does slow you down.”
This temporal dislocation creates the defining tension of contemporary defence innovation. Frontier artificial intelligence capabilities can be prototyped, tested and validated at extraordinary speed. Yet integrating those capabilities into the operational force structure of a 50,000-person defence organisation requires rigorous testing against thousands of potential second and third-order effects.
The frontier moves faster than the enterprise can follow.
There is also the question of what autonomy actually means in operational context. Commonly, civilian observers conflate autonomous systems with independent decision makers – machines that operate without human involvement.
This is a fundamental mischaracterisation. Cruise missiles, which have existed for decades, represent semi-autonomous systems: they can navigate to a target independently, but humans make the decision to launch them.
Modern uncrewed systems across Defence, whether in the air, maritime or information domains, operate within precisely defined parameters established by human commanders.
What is genuinely new is the rate at which decisions can be made and the speed at which information can be processed to support those decisions.
As Robards notes, the military advantage will accrue not to those with the most platforms or the largest budgets, but to those who can “move through that sense to side effect cycle much more rapidly and confidently”.
This insight reframes how the ADF should conceptualise its investment priorities across the coming decade.
The question of human accountability remains paramount. Both Seedhouse and Robards emphasise that despite the sophistication of AI systems, the fundamental principle governing their use within Defence is unambiguous: people must remain accountable for decisions and the outcomes of those decisions.
This is not rhetorical. It is an operational requirement grounded in understanding how AI systems actually function and their inherent limitations.
Large language models exemplify these limitations. “It’s a really eager bunny,” Seedhouse says. “You ask a question and it wants to find an answer. And if it doesn’t really know, it’s very happy to make up stuff to make you happy.” This is not a failure of the technology. It is intrinsic to how these systems are designed. They are optimised to generate plausible-sounding responses, not to acknowledge uncertainty or decline to answer.
“If you’re not happy and ask another question, it’ll try harder to make you happy,” Seedhouse says. “So, you’ve got to know that if you’re using large language models in your decision tools, that that’s a feature of them and they will always give you an answer even if the odds are too low to make a decision.”
This understanding shapes how Defence is deploying AI systems. They function as decision aids – sophisticated tools that process vast datasets and surface options for human consideration – rather than decision makers.
The art lies in presenting information in ways that support commanders without overwhelming them, that surface options without biasing their choices, and that remain explainable when scrutinised.
“If you go, someone comes to you and say, ‘The best car to buy is a blue one. Can I buy it?’ The question is, can I buy it?” Seedhouse asks. “You’re going to ask why? Why blue one? So, the system decision aid system has to be able to roll you back to tell you why it believes that decision is good.”
Beyond the operational domain, there is an industrial imperative that Defence and government cannot ignore.
The National Defence Strategy explicitly identifies the requirement for greater self-reliance and expanded industrial capacity.
This technology helps us understand other people and what they’re planning to do in their thinking and then we can see danger ahead and then this wonderful tool called diplomacy can kick in and you can prevent disasters.”
- Andrew Seedhouse
Yet, as Robards cautions, “Defence won’t, as a nation, be able to generate all of the demand or keep these industry sectors afloat just in our own right.” This creates a strategic puzzle for Australian policymakers: Defence cannot sustain an autonomous systems industry through defence contracts alone, yet civilian demand for autonomy is growing rapidly.
For Australia, however, there exists an unexpected advantage. Unlike larger defence powers saddled with legacy investments in conventional weapons production, Australia is not locked into established trajectories. “We haven’t got the infrastructure that others have had to build up,” Seedhouse says.
“We haven’t got any of that ... but we might be able to do small systems and be more agile.” Australia’s strength in academic AI research, combined with pockets of excellence in Defence laboratories, provides foundation stones.
The question is whether government policy and industrial strategy can accelerate the translation of that research into operational capability and commercial viability.
The stakes extend beyond defence. As Seedhouse argues, AI and autonomy technologies can fundamentally reshape how nations understand one another and their strategic intentions. “There is an opportunity for this technology actually to put us in a position where we can sustain peace,” he suggests.
“This technology helps us understand other people and what they’re planning to do in their thinking and then we can see danger ahead and then this wonderful tool called diplomacy can kick in and you can prevent disasters.”
This framing, viewing autonomy and AI as tools for strategic understanding rather than merely as weapons may represent the true measure of how seriously Defence is grappling with their implications.
Technology moves at machine speed. Wisdom, however, remains a distinctly human commodity.
