The US military has started using AI to guide its air strikes, according to Secretary of the Air Force Frank Kendall.
Speaking at the Air Force Association鈥檚 Air, Space & Cyber Conference in National Harbor, Maryland, on 20 September, Kendall said that the US Air Force (USAF) had recently 鈥渄eployed AI algorithms for the first time to a live operational kill chain鈥�. He didn鈥檛 give details of the strike, such as whether it was by a drone or piloted aircraft, or if there were civilian casualties.
Kendall said this use represented 鈥渕oving from experimentation to real military capability鈥�.
Advertisement
The kill chain is the process of gathering intelligence, analysing and evaluating it and directing weapons to destroy a target. The AI was incorporated into the (DCGS), which combines data from various sources, including MQ-9 Reaper drones.
DCGS collects and analyses over 1200 hours of video per day and the USAF is seeking to automate parts of this laborious process, for example with automated target recognition. Human analysts can also make deadly mistakes. In one recent case, a car in Kabul, Afghanistan, supposedly full of explosives was targeted. The vehicle was actually loaded with containers of water, and the .
With the recent withdrawal of US troops from Afghanistan, President Joe Biden announced the US will rely on more 鈥� operations in which strikes are carried out without forces on the ground to confirm targets. While the new AI can鈥檛 order a strike, the fact that it is identifying possible targets for human approval represents a significant inroad into the decision-making process, and the lack of details on the system makes it difficult for experts to evaluate. The USAF told New 杏吧原创 it wouldn鈥檛 be able to comment before publication.
鈥淥riginally, automated target recognition was about identifying and tracking well-defined military targets,鈥� says , a lecturer in war studies at King鈥檚 College London. 鈥淲e don鈥檛 know what kinds of objects or patterns of activity these systems are now trying to recognise.鈥�
at the United Nations Institute for Disarmament Research in Geneva, Switzerland, says that although expected, use of AI for targeting raises serious policy questions.
鈥淗ow does one test such algorithms and validate that they are reliable enough to be used in life-or-death missions?鈥� asks Holland. 鈥淗ow can operators know when to trust the AI system and when to override it? If an erroneous strike is launched in part on the basis of an erroneous AI system鈥檚 output, does that affect the balance of human responsibility and accountability?鈥�
Holland says these are tricky, but urgent, questions. While there has been much attention given to a possible future in which AI routinely controls 鈥渒iller robots鈥�, the involvement of AI in lethal military operations now is just as important.
McDonald says that the technology eventually becoming accessible more widely, as has happened with drones, is also a concern. 鈥淚鈥檓 much more worried about if or when this sort of technology gets in the hands of militias and non-state actors,鈥� he says.
