Australian Defence Force invests $5 million in ‘killer robots’ research
The Australian Defence Force has invested more than $5 million in researching the possibilities of artificially intelligent weaponry in an effort to design ethical killing machines.
If lethal AI weapons were used by armed forces, it would fundamentally shift the decision to kill from the hands of soldiers into the those of designers and engineers.
According to UNSW Canberra, a partner in the six-year project, it was the largest ever investment in AI ethics.
Lead researcher Dr Jai Galliot said it would investigate the current values of people who, in the future, could be deciding when a machine kills.
He said it would be up to designers to guide the military on using killer AI.
“These technical designers do need to realise that in some scenarios these weapons will be deployed and the sense of ethics and legality is going to come from them,” he said.
“The accountability is shifting in a pretty significant way.”
Armies already use limited “automatic” weaponry such as the phalanx gun on Australian Navy destroyers, which automatically detects and fires upon incoming hostile objects.
Dr Galliot said that in the same way car manufacturers were training self-driving cars to make ethical choices on the road, the army was considering how drones and other weapons could make decisions on the battlefield.
“They still need to be able to do the things that your average car might do: identify people and moving objects, try and distinguish between things, and make decisions,” he said.
“But taking lethal action in a military setting does complicate things a little bit — we need to be able to figure out, ‘Is that person a moral threat to us?’ Is that person holding a gun? Does that person have the intention to kill or harm Australians?'”
Killer weaponry could ‘make war more ethical’
Despite the challenges still to work through, Galliot said AI weapons could in fact make war more ethical.
He said drones could be taught not to shoot at “protected symbols” such as the red cross sign, or not to shoot at children, by being trained not to target people below a certain height.
“The ideal is to achieve an equal or better morality as we would find in human decision-making,” Dr Galliot said.
“So part of it is understanding how human military members go about making their decisions, and where they maybe go wrong sometimes.”
He said the aim was to “avert or minimise” scenarios WikiLeaks has dubbed “collateral murder”.
Dr Galliot used the example of the 1999 Grdelica bombing in Serbia, when NATO aircraft fatally fired missiles at a railway bridge at the exact moment a passenger train was crossing.
The death toll was estimated to be as high as 60 people — the worst rail disaster in Serbian history.
“We can foresee a scenario in which an artificial intelligence-enabled system in a similar scenario would automatically identify, ‘Oh, a train is going past,’ and therefore immediately abort that action.”
The research has been financed in part from the Defence Department’s $730 million technology fund to develop future intelligence, space and automation technologies.
Humans to stay involved in ‘decision-making loop’
The wide-ranging project will investigate public perceptions of AI weaponry, how the technology could be used to enhance military compliance with the law, and the values of the people who would be designing this killing technology.
But Dr Galliot said however the technology progresses, there would always be a human involved in the decision-making “loop”.
“People tend to anthropomorphise robots and think about RoboCop and Terminator-type scenarios … we’re still a very long way from that kind of circumstance,” he said.
Rather, he said the “ethical conundrums” would come when teaching drones and other weaponry how to interpret data coming through their sensors.
“The more you’re getting into civilian situations … the more discerning you have to be. And it may be that engineers just need to say no in some cases,” he said.
“It’s all about how far you want to go, and what the benefit is.”