According to a UN report, released in June, in March of 2020 an autonomous UAV killed a human target in Lybia without a man-in-the-loop operator involved.

The UAV in question – according to the report- was a Turkish-built STM Kargu-2 small quad-rotor kamikaze drone (loitering munition) armed probably with an embedded electronically fuzed explosive and some kind of shraphnel-generating material (maybe the casing itself).

The UAV, that is classified as a lethal autonomous weapons system (LAWS) was allegedly attacked a retreating group of Khalifa Haftar’s troops on the battlefields of Lybia, torn apart by warring factions backed by regional and superpowers.

The alleged attack was based on the capabilities of “the lethal autonomous weapons systems” that “were programmed to attack targets without requiring data connectivity between the operator and the munition: in effect, a true ‘fire, forget and find’ capability.”

At the time of the publication of the report a number of experts expressed their opinions that on the one hand the aforementioned attack might be everything, but a first, citing the widespread use of loitering munitions during the past decade or so and on the other hand they warned that there is still no exact evidence that the attack really was an autonomous kill.

Yet another panel of experts warned that if found true this event equals to the dawning of a new and nasty era, where the responsibility for taking a life can’t be established any longer.

Counter-AI Collective analysis:

The use of UAVs is sometimes portrayed as something groundbreaking and high-tech, yet the concept is hardly new. The first combat use of an UAV happened in 1981 by the then-South African Defence Forces, during Operation Protea. It came just months before Israel used the same UAV, the IAI Scout against a range of targets in the Beqaa Valley in the 1982 Lebanon War. While these UAVs weren’t autonomous, the operating pronciples had been set, including the use of sophisticated communication networks to include E-2 Hawkeye AWACS aircraft in a Command & Control function and the IAI Mastiff dedicated reconnaissance UAV.

During the last four decades a host of countries developed and fielded UAVs in every imaginable sizes and forms, but the USA and Israel took a firm lead in this field – until recently. Both countries pioneered the use of UAVs thanks to their respective doctrines, with the USA pursuing a way to eliminate so-called High Value Assets (generally purported terrorist leaders) and Israel to maintain a 24/7 situational awareness and reaction capability in what they always considered a powder keg waiting to explode anytime, and time after time.

Other countries lagged behind not because of their technical inabiloties, but rather the lack of use according to their standing doctrines. This changed with the introduction of the opaque warfare into Eurasia, in Ukraine and in Syria, only to spread into Lybia, Nagorno-Karabakh and who knows wherever else. These latter events changed the perception of what UAVs could bring to the table in both Beijing, Moscow, Tehran and Ankara and a speedy race started to implement a complex diplomatic, technological and military tool that is based on UAVs. As of now it became a top priority for both countries.

Turkey for example became one of the leading nations to produce a vast array of UAVs in every size and use and also it made sure that these newfound capabilities are supporting their foreign policy ambitions, inasmuch as the “Bayraktar Diplomacy” became synonymous with Turkey’s grand schemes in Syria, Lybia and Nagorno-Karabakh, the latter campaign becoming so visibly showcasing their grandeur, that it marked Turkey’s return to a regional power status. And according to the UN report the alleged first autonomous UAV strike happened during Turkey’s campaign to shore up the internationally-recognized Government of National Accord.

Kargu-2 UAVs of the Azerbaijani side during the 2021 Nagorno-Karabakh War

What made it possible, however was a combination of certain technological advances and the rethinking of past technologies from a half century ago.

What technological advances are prevalent in the proliferation of autonomous or semi-autonomous UAVs are mostly have to do with miniaturization of batteries, sensors and microcomputers and the advent of simple and open source real-time operating systems.The novel capability, cited by the manufacturer, that is “machine learning” is something hard to comprehemd, since this is a kamikaze drone, so it might has to do something with the producer’s mainframes learning on the datasets collected from the actual UAVs via some kind of a communications link, but not the UAV itself.

But the “real-time image processing” property might be something way simpler, because of the lack of sophisticated sensory equipment or the necessary computing power. What we have here is – according to the producer – is actually a facial recognition system, that makes it possible for the kaikaze drone to identify a given human target and then proceed to eliminate that human target. It takes a software to detect a human shape, then detect the position of the head and then compare the visual stills with the stored original facial model for a positive match. It really sounds like very 21st century, but it ain’t necesarily so. The first mass-production weapon that introduced such a capability is more than 50 years old, even if the targets were objects back then, not individual humans.

Electro-optical sensors such as used on the AGM-62 Walleye 50 years ago and on the Kargu-2 depend on both light and optical contrast for target searching and identification. This obviates their use at night and in significantly adverse weather or visual conditions where the line of sight to a target is obscured. The requirement for visual contrast between the target and its immediate surroundings imposes problems for this kind of sensor. A target sometimes indistinguishable from its own shadow. This makes it difficult to reliably designate/identify the actual target, rather than its shadow, for a true hit. The the low-light conditions at dawn and dusk often provide insufficient light for the required degree of optical contrast. Low-light optics to see through daytime haze and at dawn and dusk, however, permits its use in some of the conditions in which other optical systems are limited. The guidance phase of such a weapon is usually a contrast seeking targeting. Contrast seekers are subject to problems when the contrast spot changes. For this reason, infrared cameras could be used, which allows for a more sure hit. A more subtle problem is that the contrast seeker, unlike most seeker systems, loses accuracy as it approaches the target. This is due to the image of the tracking point growing as it approaches. What might have been a single pixel on the screen when the missile was launched from 1km away might extend over dozens of pixels by the time it has reached 20 meters from the target. At that point the tracking logic no longer works as naturally, any area within the gated range will now return a positive signal, causing the seeker to track back and forth within the area. The Kargu-2 is having a 10x digital zoom, meaning that in case of a non-moving or walking target the ammunition used might be enough to be lethal several metres away from the explosion. It is even possible that to help overcome this, Kargu-2 can apply any required level of image processing desired to address this concern.

Summing it up the best defences here is to use camouflage for the individual trooper to break up the contrast of the face: one can say that not much is changing on the battlefield. So, the way to survival is (partly) to do the basics right.