Australia will invest billions of dollars in autonomous weapons systems over the next decade while simultaneously taking part in international negotiations to restrict and control their use. Autonomous weapons systems – particularly drones – have been deployed in Gaza, Sudan, the Red Sea and in the US-Iran conflict, and with deadly effect during the Ukraine-Russia war in recent years.
Rapidly adopted by defence departments around the world, autonomous drones can offer a strategic edge in asymmetric warfare. Different nations have widely varying policies on the ethical use of autonomous weapons systems and Australia is continuing to refine its official position on the ethical use of these weapons.
Most prominently, Australia will invest $1.7bn in the Ghost Shark autonomous underwater drone and more than $2bn in the Ghost Bat autonomous flying drone: both have strike capabilities. In a trial late last year, a Ghost Bat shot down a fighter-class target drone with a missile.
Sometimes referred to as “meaningful human control”, it is thought to prevent weapons systems making the decisions on what to target.
The fear is AI will run amuck, with “malfunctioning robot armies equipped with the potential to autonomously decide to destroy cities”, says an Australian Army Research Centre journal article.
Yet according to Deane-Peter Baker, professor of ethics at the University of NSW and director of the Military Ethics Research Lab and Innovation Network (MERLIN), a “human in the loop” is not the final answer to the ethical deployment of lethal autonomous weapons.
“Enabling this requires a reliable communications link to the weapon system,” he says, “but if you can achieve that, you might just as well use a remotely piloted system.”
There is no real understanding, Baker says, on whether a “human in the loop” would have any meaningful impact on outcomes, or whether the results would be affected by operators’ so-called “automation bias” – the tendency for humans to favour suggestions made by automated decision-making systems.
“It’s misguided to talk about ‘ethical’ autonomous weapons,” he adds. “It is the commander or operator of the system who decides what is ethically appropriate.
“The question is whether or not they have a tool that is able to operate in accordance with what they decide.”
Autonomous weapons systems powered by artificial intelligence are mostly designed not to learn from their operational engagements, he adds, because designers recognise this would be far too risky.
According to Glen Schafer, formerly the chief executive of the since disbanded Defence Co-operative Research Centre on Trusted Autonomous Systems and now chief executive of Liminal Systems, a company developing multi-robotic software control, there is a “level of concern” in the Australian Defence Force regarding the ethical use of lethal autonomous weapons.
“There is, and needs to be, a tension between effectiveness of systems and the ethical overlay that we as a nation want to adopt in how they are deployed and for what purpose,” he says.
“If you deploy weapons with no consideration for collateral damage, and with full lethality, you’re soon breaching treaties and laws of armed conflict.”
Amid growing calls for regulation, international talks concerning a legally binding instrument on autonomous weapons are underway in Switzerland. The agreement would follow on from instruments such as the landmine ban treaty and a convention on certain conventional weapons.
Australia, Russia and the US, among other nations, currently oppose a new legally binding instrument of this kind, saying existing international humanitarian laws and agreements are sufficient to control the use of autonomous weapons.
Even so, Australia is continuing to consult with other nations involved in the Switzerland discussions.
Autonomous weapons systems can be held accountable, Schafer says. It’s not possible to put a machine in the dock, but systems can be “discoverable”, with an unbroken chain of causality and records of every data point and every decision.
“If that can be unpacked,” he says, “at least we have the ability to still prove compliance with things like conventions.”