A new Pentagon project envisages automated, coordinated attacks by swarms of many types of drones that operate in the air, on the ground and in the water. The idea is raising concerns about whether human oversight of such a 鈥渟warm of swarms鈥� would be possible.
The (AMASS) is a project from US defence research agency . Most details are classified, but according to US government contract documentation, it will enable multiple swarms of small aerial, ground and underwater drones to work together to knock out enemy defences. The drones will be equipped with a mix of weapons and tools such as jammers of radar, GPS and communications.
AMASS wouldn鈥檛 require general human assistance, because the swarms could coordinate across an entire area of operation such as a country. But there would be people somewhere overseeing and able to step in if necessary.
Advertisement
The project involves creating a control system to enable thousands of units, including drones, submarines and robot tanks to communicate, exchange information and coordinate their actions autonomously. DARPA has given suppliers until 10 February to bid for the $78 million contract.聽The idea is to perform a series of experiments with real and virtual drone swarms, increasing their size and the complexity of missions until thousands, of several different types, are working together.
at the Center for Strategic and International Studies think tank in Washington DC says the Pentagon has already experimented with swarms of dozens to hundreds of drones, but larger swarms with land, air and sea components introduce more complexity and may make communication more of an issue.
鈥淚f the drones are operating in a region where most or all communications are jammed, that makes oversight more difficult,鈥� he says.
Small, low-cost drones have proved effective in the Ukraine conflict, where they have destroyed tanks, swamped Ukrainian air defences and damaged the power grid.
These have been individually controlled, though: in a swarm, one operator controls many drones, allowing far more to be deployed with minimal staff. A key question is just how much human oversight is possible with AMASS, says at the Schar School of Policy and Government at George Mason University in Virginia.
鈥淎s the swarm grows in size, it鈥檒l become virtually impossible for humans to manage the decisions,鈥� he says. 鈥淎utonomy and AI will be needed to make those decisions, with all the brittleness that entails.鈥�
That raises the prospect of the drones being able to use lethal force without direct human oversight, but it isn鈥檛 clear how AMASS will handle this. The US has recently updated its , which seek and attack their own targets, and it still requires that systems have a level of human control.
The central technical challenge 鈥渋s to design human-in-the-loop planning and establish criteria to bound the autonomous operations鈥�, a DARPA spokesperson told New 杏吧原创, so drones don鈥檛 make big decisions themselves. 鈥淜ey to this will be developing software that fails into safe modes regardless of communications status and the ability of humans to intercede and set limits.鈥�
鈥淚n theory, AMASS could be entirely non-lethal, carrying out jamming or other non-kinetic attacks in support of other platforms that actually destroy the defences,鈥� says Kallenborn. 鈥淚 think that鈥檚 unlikely though.鈥�
Allen is also dubious that AMASS could carry out its mission without using lethal force. But he notes that the Pentagon鈥檚 policy on autonomous weapons, known as 3000.09, only applies to operational systems, whereas AMASS is still at an early stage. 鈥淭hey wouldn鈥檛 have to go through the 3000.09 senior review process unless and until the programme moves into formal development,鈥� he says.
Kallenborn says DARPA is known for ambitious and potentially world-changing projects, and thinks AMASS falls into that category, but notes there are risks. 鈥淎 massive drone swarm prone to errors would be a terrifying thing; a new weapon of mass destruction,鈥� he says.
