Modern warfare will be increasingly AI-driven. Here’s how Australia is planning to lead the charge.
With geopolitical tensions flaring, Australia’s national security is becoming increasingly uncertain.
The power dispute between China and the US has manifested in territorial disputes over the South China Sea, an important strategic and economic maritime route for Australia – which has a big waterfront and landmass, without a sizeable population to defend it.
Thinking outside the box to develop innovative defence solutions capable of overcoming these vulnerabilities is therefore key.
Uncrewed systems to assist with surveillance and force multiplication are the latest frontier, with Australia developing this technology quicker than other nations, according to Shane Arnott, Senior Vice President Engineering at American defence technology company Anduril Industries.
Arnott drove the development of both Boeing’s MQ-28 Ghost Bat for the Royal Australian Air Force and Anduril’s Ghost Shark autonomous undersea vehicle program (XL-AUV) for the Royal Australian Navy (RAN), which are designed to strengthen Australia’s maritime and airspace presence.
Speaking with create, Arnott explained the training and regulatory requirements of getting autonomous systems off the ground, and how they will function alongside humans in battle.
AI-powered modern warfare
Besides filling in missing workforce gaps, autonomous systems can be designed without needing to factor in accommodation and protection of human life – making them much more affordable.
“There’s a lot of costly technology that goes into protecting humans, be it in a submarine travelling at the depths of the ocean or screaming around at 600 kn in a fighter jet,” Arnott said.
“Albeit for uncrewed systems, you also don’t have to ensure the failure rate is at the same level because there are no human lives at stake to perform basic operations.”
But when building robots to do the dull, dirty and dangerous tasks instead of humans, safety is still high on the agenda to ensure they don’t cross paths with other systems when humans are present.
That entails ensuring the robots are smart and aware enough of their environment through the inclusion of AI functions such as sensor fusion and processing – which build the trust element in autonomous systems.
Uncrewed defence vehicles have a raft of different sensors, depending on whether they are geared for the air or maritime domain.
“Those designed for the air domain may have visual sensors and radar to perceive things in the same way as a human sitting in a fighter jet,” Arnott said.
In lieu of a heads-up display – the transparent display of vital information in a pilot’s forward vision – robots consume data through sensors to visually process nearby objects.
![](https://createdigital.org.au/wp-content/uploads/2024/07/Untitled-design-1-1024x718.jpg)
By contrast, autonomous submarines would use sonar technology in the traditional form so they can hear vessels on the surface or other submarines in their midst, or, when arrayed together, detect the bottom of the ocean.
Combining these different modalities – whether visual, sonar or radar – enables the robot to develop a world perception and make various decisions.
“Having that ability to determine a school of fish from another submarine, or waves versus a catamaran, ensures the robot knows that when it surfaces, it does so safely,” Arnott said.
The human element
When autonomous systems are deployed, the role of humans will be elevated to “orchestrators” – moving from in the loop to on the loop.
Semi-autonomous systems, such as the remotely piloted MQ-1 Predator used in the Iraq and Afghanistan wars, require humans to be in the loop – with similar crew requirements on the ground as if it was a crewed system in the air.
“The pilots literally have their hands on the stick and throttle, making similar movements throughout the mission,” Arnott said.
The “on-the-loop” control of fully autonomous systems such as the Ghost Shark or Bat are intended to only require human intervention when something goes wrong or requires attention.
“It’s much like being a conductor, who directs the flow and can stop and start actions underway, but isn’t in the detail of playing every note,” he said.
At this stage, it’s not known how many people will be required to operate either the Ghost Shark or Bat, neither of which are in operation. This will vary with the nature and use of the systems. For example, the Ghost Shark’s missions will last for months at a time below the waves, with the orchestrator only able to communicate with the system periodically.
“The Ghost Bat, on the other hand, as a fast jet will fly for hours per mission, with the intention of having a fighter crew controlling multiple Ghost Bats at any one time,” Arnott explained.
Preparing for battle
To train an autonomous system for deployment, a “mission design environment” that precedes mission planning, operation and debrief must be created.
The mission design environment can be likened to a “gym” for the AI to develop its muscles.
“To create a Top Gun or Perisher for AI – the training courses for the best submariners and fighter pilots – you need to design a testing environment from scratch,” Arnott said.
Using a digital twin with enough representation of both software and hardware in sufficient detail so it can pass a safety case can also help to determine whether an autonomous system is safe for the purpose of a particular mission and to work alongside warfighters in future conflict.
“The Ghost Shark and Bat are pushing the envelope of what those uncrewed safety cases look like,” he added.
That means regulations are evolving in real time to take these autonomous systems into account; with the Ghost Shark, the RAN’s Robotics, Autonomous Systems and Artificial Intelligence Strategy underpins the intent to rapidly and continually onboard technologies.
This agile process, known as “evergreening”, allows new systems to be created and accepted much faster.
“We’re working with defence to convert them to the tech way of thinking, like when Apple comes out with a new iPhone or Tesla comes out with a new model – with significantly new functionality every year,” Arnott said. “But that requires new systems and functionality to be approved on a 12-month rather than 10-year cadence.”
Expedition of defence system rollouts is a sign of precarious times, much like in World War II when new aircraft were built based on brand new designs every six to 12 months.
“As a defence industry, these processes have since become slower and more expensive,” he said. “That needs to dramatically change, and autonomous systems present an opportunity to do that.”
Military applications
A key role for autonomous systems is force multiplication. That means for every fighter aircraft, there will be multiple Ghost Bats, Arnott said.
The Ghost Shark’s job is to work more independently as well as together with other autonomous and crewed systems, where the goal is to have a wider presence in the ocean, particularly due to the size of Australia’s coastline.
According to the Hon Pat Conroy MP, Minister for Defence Industry and Minister for International Development and the Pacific, the Ghost Shark will become Mission Zero (0) for the Advanced Strategic Capabilities Accelerator (ASCA), and will be used to conduct persistent intelligence, surveillance, reconnaissance and strike.
With the first prototype of the Ghost Shark delivered one year early, and the first production variant expected by the end of 2025, it could be deployed as early as 2026.
During missions, the autonomous systems will need to communicate with each other, perform collaborative actions, and communicate these back to the human – whether on another plane or vessel or back at headquarters.
That’s where engineering, and regulations come into play – determining what’s permissible, how much oversight is required, and the ability of humans to override actions.
“This provides the authority to the robot on its scope and that it can operate within these parameters, but no further – which will change depending on the mission,” Arnott said.
What’s technically possible and permissible differs in times of peace and conflict.
“More risk may be [considered acceptable] during conflict than peacetime, which adds an extra level of complexity,” Arnott said.
Explore the frontiers of air and space operations at the 21st Australian International Aerospace Congress.