(Thread) The EU Parliament has just adopted a resolution for a common EU position on the military applications of AI.

Here are some highlights.
Full text: https://www.europarl.europa.eu/doceo/document/TA-9-2021-0009_EN.html

The resolution "Insists on the need for an EU-wide strategy against LAWS and a ban on so-called ‘killer robots'"

AI "must be subject to meaningful human control"

MHC is referenced multiple times throughout as a key pillar of responsible AI.

The resolution "Points to the clear risks involved in decisions made by humans if they rely solely on the data, profiles and recommendations generated by machines."

i.e: human control needs to be more than just trusting whatever the machine says or does.

"Considers that LAWS should be used only as a last resort."

"autonomous decision-making should not absolve humans from responsibility...people must always have ultimate responsibility for decision-making processes so that the human responsible for the decision can be identified."

Amy rules for military AI must be accompanied by mechanisms to ensure "that the regulatory framework does not become outdated as a result of technological developments."

Urges the EU to put itself at the forefront of UN efforts to align the use of military with relevant law—and stresses that any framework for military AI "must never breach, or permit breaches of, the dictates of public conscience and humanity as stated in the Martens clause"

Calls on the EU to lead in global negotiations on "an AI arms control regime."

Calls on negotiations to update "all existing treaty instruments on arms control, disarmament and non-proliferation so as to take into account AI-enabled systems used in warfare."

To be legally compliant, military AI depends on the "comprehensive situational understanding of the human operator, the predictability, reliability, and resilience of the AI-enabled system..."

"...the human operator’s ability to detect possible changes in circumstances and operational environment, and their ability to intervene in or discontinue an attack."

Minimum requirements for military AI:
-"able to distinguish between combatants, non-combatants"
-"recognise when a combatant surrenders or is hors de combat"
-"not have indiscriminate effects, not cause unnecessary human suffering"

Minimum requirements for military AI (cont.)
-"not be biased or trained on intentionally incomplete data"
-"comply with the principles of international humanitarian law, proportionality in the use of force and precaution before intervention."

Calls for lethal autonomous weapons to be subject to the Arms Trade Treaty of 2 April 2013.
https://www.un.org/disarmament/convarms/arms-trade-treaty-2/