AS THE coalition of armed forces led by the US continues its strike on the
Taliban, America鈥檚 Air Force is gearing up to test a new weapon that could
change the way wars are fought鈥攂oth tactically and politically.
Last-minute checks are under way for the first flight tests of the Unmanned
Combat Air Vehicle (UCAV) in December. The aim of this $120 million
programme is to make fully armed, autonomous combat planes capable of carrying
out offensive missions.
But the prospect of a future in which planes choose their own targets brings
with it grave concerns of reliability and accuracy. Can a machine ever decide
what is or is not a legitimate target, distinguishing friend from foe and
civilians from the military? 鈥淚t would have to be able to tell if a truckload of
nuns was about to cross a bridge before blowing it up,鈥� says Andrew Brookes, air
defence analyst at the International Institute for Strategic Studies in
London.
Advertisement
If the machines can鈥檛 make such distinctions, their use could breach the
Geneva Convention, an international agreement outlawing the indiscriminate
firing of weapons.
The initial aim is to develop an uncrewed plane capable of flying ahead of,
or with, piloted aircraft to suppress enemy air defences. But New
快猫短视频 has learned that the long-term goal is to make an uncrewed
aircraft with the same capabilities as existing fighter planes. And it doesn鈥檛
stop there: a complete robotic military attack force is planned. Both the US
Navy and Army have similar projects in development to make armed autonomous
helicopters and ground vehicles, including tanks.
While human operators will be able to issue orders, change operational
instructions and ultimately control all of these machines from afar, the
automatons will be designed to defend themselves if communications break down.
鈥淲e are concerned,鈥� says Joost Hiltermann, head of the arms division at Human
Rights Watch in Washington DC. 鈥淓specially if the ability to pull the trigger
occurs increasingly further away from the human.鈥�
How it will all work is far from clear. The UCAV programme explicitly states
that no lethal action can be taken without prior human consent. But, according
to the UCAV specifications, 鈥済iven prior consent the UCAV operational system
should be capable of autonomous self-defence actions and engagement of pop-up
迟丑谤别补迟蝉鈥�.
Precisely what a 鈥減op-up鈥� threat is, the US Defense Advanced Research
Projects Agency (DARPA), which funded the project, refuses to say. It will only
state that this is a tactical matter that would ultimately rest with the US Air
Force. Deciding what is or isn鈥檛 a target for action is usually the job of
military intelligence. But a fully autonomous UCAV, cut off from its operators,
will at some stage have to make this decision itself.
Some argue that this shouldn鈥檛 be a problem because legitimate targets, such
as enemy planes and anti-aircraft batteries, are easy to spot. Most emit
well-documented electronic signals and signatures, says Kevin McLaughlin, a
former US Air Force commander at the Center for Strategic and International
Studies in Washington DC.
However, mistakes do occur. An investigation is under way to determine
whether the Russian airliner that crashed into the Black Sea last week was
accidentally shot down by a radar-homing Ukrainian missile during a military
exercise. And unarmed crewless aircraft have received control signals intended
for other planes, crashing as a result.
To avoid civilian and 鈥渇riendly fire鈥� casualties, the machines need to know
the context of any engagement. 鈥淭here has to be a very complex set of rules of
engagement criteria,鈥� says McLaughlin. Like pilots, UCAVs would have to abide by
them.
But no matter how sophisticated the technology, its decision-making
capabilities will be crude compared to a human鈥檚, says John Lynn, a military
historian at the University of Illinois, Urbana Champaign. Indeed, the
state-of-the-art software is a long way off from such subtle reasoning, says
James Hendler, an artificial intelligence expert at the University of Maryland
and former chief scientist of information systems at DARPA. 鈥淵ou would have to
make them observe much stricter criteria than pilots,鈥� he says.
Finding 鈥渓egitimate鈥� targets in a ground war is even more difficult. Unlike
tanks and anti-aircraft radar, troops and bunkers don鈥檛 give off easily
identifiable electronic signatures. 鈥淚t would make the problem more acute,鈥� says
Hiltermann. Distinguishing between an armed and unarmed person, or a soldier and
a civilian, could prove impossible.
Even if the technology finally meets these requirements, it could mean a more
鈥渢rigger-happy鈥� approach. Once the risk of troops returning in body bags has
been removed, it could make governments more prone to take military action.
