Indeed, LJ.
There is a
potential paradox with RPS, particularly as there is increased automation and a move towards 'supervised autonomy'. Because the aircraft must operate safely and within rules and procedures 100% of the time, on-board systems will typically default to the safe option. Here's a quote from an in-hourse journal:
Accordingly, and until artificial intelligence is a proven technology, autonomy will be used to assist potentially lethal missions, rather than to execute them. As Moshkina and Arkin showed, the public expect a higher standard of behaviour from robots than the humans who would otherwise do that job. Whereas humans are fallible, and will make errors of judgement that could result in legally disproportionate collateral damage, autonomous combat systems will not have that ‘error of judgement’ leeway. Even though unmanned aircraft will do the dull, dirty and dangerous and deep jobs without fault[1], the requirement to discriminate against exacting and binary standards of yes/no, rather than ‘maybe’ could result in autonomous unmanned aircraft completing fewer ‘lethal’ missions than their human counterparts as an ‘artificial conscience’, decision software and learned AI behaviour will default towards safe or low-risk options, such as withholding the release of weapons when a human operator may prosecute an attack albeit with some un-quantifiable reservations. This, of course, is on the assumption that there is a clear computational definition of a ‘civilian’ that is compliant with Additional Protocol I.[2]
[1] For a discussion of these factors, see the UK MOD's JDN 2/11 pp 3-43-6.
[2]Art 50(1) of Additional Protocol I, 1977 of the 4th Geneva Convention essentially defines a civilian in the negative sense, ie ‘as any person not a combatant’. Malcolm Shaw International Law 5th Edn CUP 2003 p 1061.