快猫短视频

Artificial intelligence may soon be deciding who lives or dies. The US Army wants to build smart, cannon-fired missiles that will use AI to select their targets, out of reach of human oversight. The project has raised concerns that the missiles will be a form of lethal autonomous weapon 鈥� a technology many people are campaigning to ban.

The US Army鈥檚 project is called Cannon-Delivered Area Effects Munition (C-DAEM). Companies will bid for the contract to build the weapon, with the requirements stating it should be able to hit 鈥渕oving and imprecisely located armoured targets鈥� whose 鈥渆xact position has high uncertainty鈥�. Unlike laser-guided weapons, which hit a target highlighted by a human operator, C-DAEM will find targets for itself.

A parallel project will aim to develop algorithms for the weapons. These will be similar to face-recognition algorithms, but will use infrared cameras instead of traditional ones to identify targets, such as tanks. Each missile will contain a chip like those found in smartphones for running the algorithms.

The weapons will have a range of up to 60 kilometres, taking more than a minute to arrive, and will be able to search an area of more than 28 square kilometres for their targets. They will have a method for slowing down, such as a parachute or small wings, which they will use while scanning and classifying objects below. Their algorithms will have to discriminate between friend and foe in areas that may also contain civilians.

The weapons will autonomously hunt for targets, deciding when they have found one and attacking without human intervention, says Mark Gubrud at the University of North Carolina. They are effectively killer robots, he says.

Lethal autonomous weapons

鈥淚t moves us from the current situation, in which a human operator needs to be reasonably sure that the target is legitimate, to one in which the human operator need only have a vague intuition that somewhere in a 10-square-mile area there might be a bad person,鈥� says Stuart Russell at the University of California, Berkeley.

The new weapon is meant to replace much-criticised cluster warheads, which scatter dozens of grenades over a wide area. These are effective against armoured vehicles, but have a high dud rate, meaning that dangerous unexploded grenades remain after an attack.

Smart missiles that recognise and avoid civilians ought to be safer than indiscriminate bombs, but Russell says it isn鈥檛 plausible that the new weapons would be used in the same way.

鈥淚t seems likely that less care would be taken in target selection by the operator and attacks would take place with less reliable intelligence,鈥� he says.

Many campaigners demand that such weapons are banned, including the Campaign to Stop Killer Robots, a coalition of NGOs that want to see the use of force controlled entirely by humans.

While some would see this type of system as an autonomous weapon, it technically complies with the Pentagon鈥檚 rules, which say that a human operator must select each target or group of targets.

鈥淭he word 鈥榮elect鈥�, as it is used in the US policy, is very slippery,鈥� says Gubrud. 鈥淎 human can 鈥榮elect鈥� a target before it has even been seen.鈥� Select may simply mean providing a description of the target and its general area, he says.

The US Army already has a missile that uses infrared sensors to autonomously locate and attack tanks. But it can only search an area 200 metres across, meaning that the human operator is more certain of the target before firing it.

Several contractors are competing for the C-DAEM project. Prototype demonstrations are due to be held in 2021. Developing the software for an autonomous round is scheduled to take around three years.

A spokesperson for the US Army said: 鈥淭his is not an autonomous weapon, nor is it intended to be. We seek an advanced capability for a round 鈥� once fired 鈥� to continue pursuing a target despite the types of interference that might cause it to pursue something else. This would improve our capabilities to avoid collateral damage.鈥�