Prime Minister Scott Morrison has added science and technology to the portfolio of the defence industry minister, Melissa Price, signalling closer alignment of the research sector with the defence organisation.

As that’s been done in the context of the new AUKUS partnership, it’s worth noting the United States and British governments’ sharp focus on the defence and national security implications of artificial intelligence. A review of defence innovation in Australia may reveal a need to strengthen policy areas in which we have lagged behind, in particular our lack of strategic direction with a national AI strategy.

The media coverage of the AUKUS pact has mostly focused on the nuclear-powered submarine announcement, but the agreement also emphasises the importance of AI to defence and national security. Australia’s innovation ecosystem will need to take on board developments in the US and the UK.

Britain’s new AI strategy puts defence front and centre after a 2017 industrial strategy and a £1 billion AI sector deal in 2018. By the end of this year, the UK Ministry of Defence intends to publish its own strategy on how it will adopt and use AI. Over the longer term, the National Security and Investment Act will protect national security and monitor investments in AI technology.

In the US, the necessity for speed has been articulated by the National Security Commission on Artificial Intelligence, which was tasked with accelerating the adoption of AI for national security. ‘By 2025, the Department of Defense and the Intelligence Community must be AI-ready,’ it says. This represents a new paradigm in which AI and national security go hand in hand.

Overseen by former Google CEO Eric Schmidt, the AI commission is championing partnerships between technology firms and the Defense Department. A cynic may say that this is to avert the gaze of Congress from antitrust questions in the tech sector, but it is more a sign that technology is increasingly seen as geopolitical.

Elements of the linking of AI and national security most important for Australia are the US treatment of its research enterprise as a national asset, and how the maturation of AI—and its monopolisation by a smaller number of firms—is affecting investment.

Former chief defence scientist Robert Clark and ASPI executive director Peter Jennings have pointed out that US universities are being redefined as an important part of the national security enterprise, making them subject to stronger regulation and oversight. They noted that ‘the current largely open approach of Australian research universities to their international links is significantly exposed’.

If the US positioning of research as a national asset is anything to go by, Australian universities are likely to be repositioned as an element of critical infrastructure. Australian AI research ranks highly in terms of per capita output and quality, in particular the frequency with which research papers are cited. But we have a weak venture capital system and longstanding issues in commercialising our research.

Although there are credible calls to have a wholesale redesign of science and technology research along the lines of the US Defence Advanced Research Projects Agency, or DARPA, some trends will affect these solutions, particularly the sector’s maturation and its monopolisation by fewer firms. How we forge partnerships with international tech firms while we develop our own sector is increasingly important.

The shrinking number of firms involved in AI presents national security risks and compounds problems in defence acquisition, where procurement risk is spread by having diverse suppliers.

Fewer firms possess the human capital and brute computer-processing power to handle the data-processing requirements of AI projects. Graduates are increasingly taking jobs with industry rather than with public research institutions. Investors are putting their money into established companies or, as the OECD put it, making ‘bigger bets on a smaller number of companies as the AI sector matures’.

While technology seems to be getting bigger and to require an endless flood of private and public cash to take it to the stratosphere, it’s clear that the need for speed in defence acquisition is increasing.

In the US, subtle changes are happening in defence innovation to make processes faster and to navigate the procurement labyrinth.

The US Air Force started its own venture capital division, AFVentures, as a division of its technology and acquisition and development group, AFWerx. It has funded an Australia-based AI company, Curious Things, which is trying to redesign the USAF recruitment process. With the Australian defence innovation system under review, it will be interesting to see how these big and small challenges will be incorporated.

As we race towards having an AI edge, there hasn’t been a mature conversation about what the next wave of convergent technology looks like and who should benefit.

The new paradigm of AI and national security will bring out all sorts of arguments that the next wave involves smart dust, photon programming, biotech–AI interfaces, lethal autonomy and the multiverse. But it’s more important to locate Australia’s strategic vision alongside the areas of interoperability and cooperation that are emerging with the US and UK.

Originally published 29 April 2019.

The 2016 defence white paper and the decades-long integrated investment program will deliver a future force that includes 72 joint strike fighters, several hundred infantry fighting vehicles, nine new frigates and 12 new submarines. F-35 deliveries have started but the ‘future’ frigate and submarine programs were well named: the Hunter-class frigates will turn up, all going well, between 2028 and the early 2040s, and the first Attack-class submarine is scheduled to enter service in 2035, with the 12th in the mid-2050s.

Meanwhile, technology is advancing quickly. There’s been lots of anxiety and hype about lethal autonomous weapon systems—the feared killer robots. But there’s also been a huge amount of experimentation and investment in autonomous and semi-autonomous systems, land, air, sea and undersea.

The US, for example, has demonstrated self-navigating surface and undersurface vessels and numerous land and air systems. The US Navy’s Project Overlord is aiming to deliver 10 large unmanned surface ships by the mid-2020s—before our first frigate enters service. China and Russia are developing similar systems.

The Australian Army is experimenting with unmanned aerial vehicles (UAVs) and the RAN with various unmanned systems. Boeing’s ‘Loyal Wingman’ concept UAV is being developed for the RAAF, with a refreshingly rapid timeline that aims for a first flight in 2020. But I don’t see a shift in where Defence’s investment dollars are going that indicates this is more than a hobby for now. That needs to change fast.

Such systems will, at the very least, be essential ‘complementary capabilities’ to the big, complex manned platforms lumbering into the ADF’s inventory over the next 35 years. They may make some of these platforms obsolete, or turn them from frontline combat systems into motherships, providing power and logistics support.

The design concepts of our ‘future force’ need to be worked on now with this in mind.

The other big area in which technological developments and our defence capabilities look badly out of step is in applying ‘narrow artificial intelligence’ to how defence systems are used and to how decisions—tactical, operational and strategic—are made. Supermarkets use this type of AI to optimise inventories to meet customer needs. It displaces humans in factories and call centres and helps us with our internet searches. So, narrow AI is not the potentially species-threatening superintelligence or ‘artificial general intelligence’ Elon Musk has warned about. Think more along the lines of high-speed trading algorithms or Google’s AlphaGo than Will Smith’s enemy in I, Robot.

Technological change is redefining warfare by the attributes of speed, autonomy and precision. Dispersal and rapid reconstitution of combat losses will be required for the future ADF to thrive in such an environment. Speed will be essential not just for weapon systems (for example, hypersonics) but also for command, control and decision-making.

But speed and hierarchies don’t coexist easily or naturally. Hierarchies, by their nature, involve delayed decision-making and sequential decision-making at multiple levels—and they’re the enemies of speed of response. So the organisational structures and processes in ships, squadrons, platoons, brigades and joint taskforce headquarters, across the Defence organisation, in every part of Russell Offices, and right across the lake into the cabinet room will need to change to be fit for future conflict. It will be a wrenching and threatening package of reform and change, as disruptive to current military hierarchy and structures as Amazon has been to the retail industry.

For AI to take back the time advantage from adversaries using high-speed precision weapons, it will need to find its way into the ADF’s existing systems and platforms soon—well before the first new frigate or submarine gets wet.

Warfare through history is a story of measure and countermeasure. While the combat systems that are on our Anzac-class frigates and air warfare destroyers can track and defeat numerous simultaneous missile threats, the speed of decision-making on board the ships depends on multiple human intervention points. Myriad pieces of information from shipboard and offboard sensors, intelligence and data need to be fused into a coherent picture for the crew and captain.

No matter how skilled the console operators, principal warfare officers or commander, and no matter how highly developed the combat systems are, the time-and-motion problem posed by numerous supersonic and hypersonic threats will be intense.

AI will be essential for supporting high-speed data analysis and decision-making. Training AI decision-support algorithms on the mass of data that a ship’s combat system and command chain will have available won’t just allow the vessel to use its onboard defence systems to defeat attacks; it will also facilitate rapid cueing of offboard systems—say, on a nearby F-35, Growler electronic attack aircraft, P-8 patrol aircraft, accompanying surface vessels or even unmanned systems—to defend the ship and attack the adversary’s weapons, launch sites, targeting systems, and command-and-control chain. And it will allow the AI-enabled decision system to draw on masses of historical data and intelligence relevant to the environment and threats being faced.

In the world of high-speed, autonomous and precision weapon systems, AI can return the advantage to the defender by enabling its systems to make faster and better informed decisions than the attacker’s systems. It can support an asymmetric response that will give the defender more options than simply shooting faster than the attacker.

Concepts for employing this aid to high-speed decision-making in existing ADF platforms and systems are urgently needed. To put these concepts into action, equally urgent design, development and investment is required to retrofit systems like the Anzacs and AWDs, and also the command-and-control systems already operating across the ADF. The Anzacs, the Collins submarines and the AWDs will be frontline combatants for decades, so it’s essential to equip them to survive and be effective over that timeframe. AI-enabled onboard decision-support systems must be a part of the plan.

A defining opportunity is the planned $2.5 billion investment in AIR 6500 to acquire a command-and-control system for integrated air and missile defence that will operate from the level of protecting deployed army units right up to protecting whole joint taskforces, combining army, air force and navy units and systems. This system must have AI built in from the concept stage.

That will be key to taking advantage of the masses of data collected and required by ADF units, along with historical data that enables AI systems to learn and to apply that knowledge to new environments.

It’s absolutely feasible to retrofit autonomy and AI algorithms into military platforms and control systems. The US Air Force is converting more F-16 fighters to autonomous aircraft, while DARPA is bringing AI-enabled tools into existing aircraft so that each can control multiple drones. Lockheed Martin’s Sikorsky now has years of experience in developing ‘software and hardware that bring autonomous flight to the cockpit’ of its Black Hawk helicopters.

It’s great to muse about the wonders of Australia’s $200 billion ‘future force’, but our time would be much better spent on the urgent new work and thinking required to make our existing force effective in the environment it will face in the early and mid-2020s.

That means loving the force you have—acting now to develop the concepts for use of AI across the ADF, and investing to retrofit the systems and platforms that will remain in service for decades.

That will prepare us for those bright sunlit days when the Hunter and Attack classes become real.

The 2016 defence white paper and the decades-long integrated investment program will deliver a future force that includes 72 joint strike fighters, several hundred infantry fighting vehicles, nine new frigates and 12 new submarines. F-35 deliveries have started but the ‘future’ frigate and submarine programs were well named: the Hunter-class frigates will turn up, all going well, between 2028 and the early 2040s, and the first Attack-class submarine is scheduled to enter service in 2035, with the 12th in the mid-2050s.

Meanwhile, technology is advancing quickly. There’s been lots of anxiety and hype about lethal autonomous weapon systems—the feared killer robots. But there’s also been a huge amount of experimentation and investment in autonomous and semi-autonomous systems, land, air, sea and undersea.

The US, for example, has demonstrated self-navigating surface and undersurface vessels and numerous land and air systems. The US Navy’s Project Overlord is aiming to deliver 10 large unmanned surface ships by the mid-2020s—before our first frigate enters service. China and Russia are developing similar systems.

The Australian Army is experimenting with unmanned aerial vehicles (UAVs) and the RAN with various unmanned systems. Boeing’s ‘Loyal Wingman’ concept UAV is being developed for the RAAF, with a refreshingly rapid timeline that aims for a first flight in 2020. But I don’t see a shift in where Defence’s investment dollars are going that indicates this is more than a hobby for now. That needs to change fast.

Such systems will, at the very least, be essential ‘complementary capabilities’ to the big, complex manned platforms lumbering into the ADF’s inventory over the next 35 years. They may make some of these platforms obsolete, or turn them from frontline combat systems into motherships, providing power and logistics support.

The design concepts of our ‘future force’ need to be worked on now with this in mind.

The other big area in which technological developments and our defence capabilities look badly out of step is in applying ‘narrow artificial intelligence’ to how defence systems are used and to how decisions—tactical, operational and strategic—are made. Supermarkets use this type of AI to optimise inventories to meet customer needs. It displaces humans in factories and call centres and helps us with our internet searches. So, narrow AI is not the potentially species-threatening superintelligence or ‘artificial general intelligence’ Elon Musk has warned about. Think more along the lines of high-speed trading algorithms or Google’s AlphaGo than Will Smith’s enemy in I, Robot.

Technological change is redefining warfare by the attributes of speed, autonomy and precision. Dispersal and rapid reconstitution of combat losses will be required for the future ADF to thrive in such an environment. Speed will be essential not just for weapon systems (for example, hypersonics) but also for command, control and decision-making.

But speed and hierarchies don’t coexist easily or naturally. Hierarchies, by their nature, involve delayed decision-making and sequential decision-making at multiple levels—and they’re the enemies of speed of response. So the organisational structures and processes in ships, squadrons, platoons, brigades and joint taskforce headquarters, across the Defence organisation, in every part of Russell Offices, and right across the lake into the cabinet room will need to change to be fit for future conflict. It will be a wrenching and threatening package of reform and change, as disruptive to current military hierarchy and structures as Amazon has been to the retail industry.

For AI to take back the time advantage from adversaries using high-speed precision weapons, it will need to find its way into the ADF’s existing systems and platforms soon—well before the first new frigate or submarine gets wet.

Warfare through history is a story of measure and countermeasure. While the combat systems that are on our Anzac-class frigates and air warfare destroyers can track and defeat numerous simultaneous missile threats, the speed of decision-making on board the ships depends on multiple human intervention points. Myriad pieces of information from shipboard and offboard sensors, intelligence and data need to be fused into a coherent picture for the crew and captain.

No matter how skilled the console operators, principal warfare officers or commander, and no matter how highly developed the combat systems are, the time-and-motion problem posed by numerous supersonic and hypersonic threats will be intense.

AI will be essential for supporting high-speed data analysis and decision-making. Training AI decision-support algorithms on the mass of data that a ship’s combat system and command chain will have available won’t just allow the vessel to use its onboard defence systems to defeat attacks; it will also facilitate rapid cueing of offboard systems—say, on a nearby F-35, Growler electronic attack aircraft, P-8 patrol aircraft, accompanying surface vessels or even unmanned systems—to defend the ship and attack the adversary’s weapons, launch sites, targeting systems, and command-and-control chain. And it will allow the AI-enabled decision system to draw on masses of historical data and intelligence relevant to the environment and threats being faced.

In the world of high-speed, autonomous and precision weapon systems, AI can return the advantage to the defender by enabling its systems to make faster and better informed decisions than the attacker’s systems. It can support an asymmetric response that will give the defender more options than simply shooting faster than the attacker.

Concepts for employing this aid to high-speed decision-making in existing ADF platforms and systems are urgently needed. To put these concepts into action, equally urgent design, development and investment is required to retrofit systems like the Anzacs and AWDs, and also the command-and-control systems already operating across the ADF. The Anzacs, the Collins submarines and the AWDs will be frontline combatants for decades, so it’s essential to equip them to survive and be effective over that timeframe. AI-enabled onboard decision-support systems must be a part of the plan.

A defining opportunity is the planned $2.5 billion investment in AIR 6500 to acquire a command-and-control system for integrated air and missile defence that will operate from the level of protecting deployed army units right up to protecting whole joint taskforces, combining army, air force and navy units and systems. This system must have AI built in from the concept stage.

That will be key to taking advantage of the masses of data collected and required by ADF units, along with historical data that enables AI systems to learn and to apply that knowledge to new environments.

It’s absolutely feasible to retrofit autonomy and AI algorithms into military platforms and control systems. The US Air Force is converting more F-16 fighters to autonomous aircraft, while DARPA is bringing AI-enabled tools into existing aircraft so that each can control multiple drones. Lockheed Martin’s Sikorsky now has years of experience in developing ‘software and hardware that bring autonomous flight to the cockpit’ of its Black Hawk helicopters.

It’s great to muse about the wonders of Australia’s $200 billion ‘future force’, but our time would be much better spent on the urgent new work and thinking required to make our existing force effective in the environment it will face in the early and mid-2020s.

That means loving the force you have—acting now to develop the concepts for use of AI across the ADF, and investing to retrofit the systems and platforms that will remain in service for decades.

That will prepare us for those bright sunlit days when the Hunter and Attack classes become real.

In 2017, a US Department of Defense study of the future operating environment described artificial intelligence (AI) as the most disruptive technology of our time.

In this environment, says the report:

big data techniques interrogate massive databases to discover hidden patterns and correlations that form the basis of modern advertising—and are continually leveraged for intelligence and security purposes by nation states and non-state entities alike.

The potential applications of artificial intelligence, and the deep-learning capabilities it brings, may be one of the most profound artefacts of the fourth industrial revolution. Informed and experienced experts such as Secretary James Mattis and US Deputy Secretary of Defense Bob Work are questioning whether AI will change not just the character, but also the nature of war. That highlights just how disruptive this technology is likely to be for society and commerce, as well as for human competition and conflict.

In future conflicts, we can expect decision cycles to become faster than human cognition can process. Military command and control—and strategic decision-makers—will need AI that can process information and recommend options for decisions faster (or of higher quality) than can the enemy.

And as I’ve written previously, military organisations will contain thousands or even tens of thousands of unmanned and robotic systems, all with some type of AI. These swarms will demand AI-assisted command and control, as will the other composite human-automated military formations that are likely to exist in future areas of conflict.

While there’s a need to build capacity within the Defence organisation, the guiding principles apply to a wider national security community. Defence exists in an ecosystem of government organisations working towards national objectives. It’s imperative to quickly introduce AI to support decision-making in this joined-up environment. Then, immediate action is required for Defence (the department and the ADF) to rapidly increase its understanding of the applications of AI, and to contribute to a national approach.

Frank Hoffman recently proposed that military organisations may be at the dawn of a seventh revolution in military affairs that he calls the ‘autonomous revolution’. Underpinned by exponential growth in computer performance, improved access to large datasets, continuing advances in machine learning and rapidly increasing commercial investment, the future application of AI and machine learning may change military organisations and, more broadly, how nations prepare for war.

However, as Max Tegmark has written, there’s debate among AI researchers about whether human-level AI is possible, and when it might appear. The nearest estimates are ‘in a few decades’, with others predicting ‘not this century’ and ‘not ever’.

But, assisted, augmented and autonomous intelligence capabilities are already in use or can be expected over the coming decade. AI needn’t replicate human intelligence to be a powerful tool. The intellectual preparation of Defence personnel, and that of the wider national security community, to effectively use AI must begin now.

The first of four key imperatives is to start educating Defence and other national security personnel about AI. A Belfer Center report finds that it’s vital for non-technologists to be conversant with the basics of AI and machine learning. The aim is to develop baseline AI literacy among more Defence and national security leaders to supplement the expertise of the few technical experts and contractors who design and apply algorithms. Reading lists, residential programs and online courses, as well as academic partnerships and conferences, will help.

Defence education must be adapted to deliver greater technical literacy so personnel better understand machine learning and AI. This will permit a wider institutional capacity to effect quality control and address the risks of misbehaving algorithms. Personnel must be educated about the ethical issues of using AI for national security purposes. The overarching aim must be to develop a deep institutional reservoir of people who understand the use of AI, and who appreciate how human and AI collaboration can be applied most effectively at each level of command.

The second imperative is for Defence and other agencies to move beyond limited experimentation and thinking in key areas to broader explorations of how to use AI. This might include finding new ways for the national security community to use AI to work together more effectively. It may also allow the use of AI to support decision-making that could drive changes to the organisation of military and non-military elements in national security.

Some aspects of this exploration will require leaps of faith in assessing how capable AI will be in the future. Between the wars, the German army used fake tanks to develop its combined arms operating systems. So too might we use anticipated future capabilities to build new integrated national security and Defence operating systems using AI.

Potentially, this exploration could even permit consideration of significant changes in strategic decision-making processes and organisations—mostly leftovers from second and third industrial revolution mindsets.

The third imperative is for Defence to deepen collaboration with external institutions working on AI applications. All of the Group of Eight universities in Australia conduct AI research and teach applications. A number have partnerships with international institutions, including some who do work for foreign military and security agencies. They could help Defence, and other government agencies, explore the use of AI to support operational capabilities, for decision-making in directing and running operations and in other strategic functions such as education. Broad collaborative research with our closest allies will  permit sharing of best practice and offers small nations like Australia the opportunity to develop bespoke applications that complement—not copy—overseas innovations.

Eventually, this could (and probably should) lead to a fourth imperative, the development of an AI equivalent of the Australian Naval Shipbuilding Plan, and a focus for a sovereign capability in AI research and development. Such a national approach is important because it might provide resourcing for further collaboration between universities and government and commercial entities. It may provide the basis for a larger national industry to support non-military AI functions.

Finally, if robotics is included, it may provide a basis to mobilise national effort if that’s necessary in the coming decades. Australia sits at the end of a long supply line for almost every element of sophisticated weaponry. Our successors may thank us if we have the forethought to develop an indigenous capacity to design and construct (using additive manufacturing) swarms of autonomous systems.

Sitting back and observing foreign developments isn’t an effective strategy. An aggressive national and departmental program of research, experimentation and educating Defence and other national security personnel is required. The knowledge for such a program exists in our universities and changes in the regional and global security environment provide the strategic drivers for action.

Leveraging potential capabilities, we must start to educate our people now, and we must develop a national sovereign capacity to harness AI for national security purposes. In this way we might develop a truly brilliant future ADF.