Does facial recognition tech in Ukraine’s war bring killer robots nearer?

Clearview AI is offering its controversial tech to Ukraine for identifying enemy soldiers – while autonomous killing machines are on the rise

Darian Meacham and Martin Gak on OpenDemocracy

Technology that can recognise the faces of enemy fighters is the latest thing to be deployed to the war theatre of Ukraine. This military use of artificial intelligence has all the markings of a further dystopian turn to what is already a brutal conflict.

The US company Clearview AI has offered the Ukrainian government free use of its controversial facial recognition technology. It offered to uncover infiltrators – including Russian military personnel – combat misinformation, identify the dead and reunite refugees with their families.

To date, media reports and statements from Ukrainian government officials have claimed that the use of Clearview’s tools has been limited to identifying dead Russian soldiers in order to inform their families as a courtesy. The Ukrainian military is also reportedly using Clearview to identify its own casualties.

This contribution to the Ukrainian war effort should also afford the company a baptism of fire for its most important product. Battlefield deployment will offer the company the ultimate stress test and yield valuable data, instantly turning Clearview AI into a defence contractor – potentially a major one – and the tool into military technology.

If the technology can be used to identify live as well as dead enemy soldiers, it could also be incorporated into systems that use automated decision-making to direct lethal force. This is not a remote possibility. Last year, the UN reported that an autonomous drone had killed people in Libya in 2020, and there are unconfirmed reports of autonomous weapons already being used in the Ukrainian theatre.

Our concern is that hope that Ukraine will emerge victorious from what is a murderous war of aggression may cloud vision and judgement concerning the dangerous precedent set by the battlefield testing and refinement of facial-recognition technology, which could in the near future be integrated into autonomous killing machines.

To be clear, this use is outside the remit of Clearview’s current support for the Ukrainian military; and to our knowledge Clearview has never expressed any intention for its technology to be used in such a manner. Nonetheless, we think there is real reason for concern when it comes to military and civilian use of privately owned facial-recognition technologies.

Clearview insists that its tool should complement and not replace human decision-making. A good sentiment but a quaint one

The promise of facial recognition in law enforcement and on the battlefield is to increase precision, lifting the proverbial fog of war with automated precise targeting, improving the efficiency of lethal force while sparing the lives of the ‘innocent’.

But these systems bring their own problems. Misrecognition is an obvious one, and it remains a serious concern, including when identifying dead or wounded soldiers. Just as serious, though, is that lifting one fog makes another roll in. We worry that for the sake of efficiency, battlefield decisions with lethal consequences are likely to be increasingly ‘blackboxed’ – taken by a machine whose working and decisions are opaque even to its operator. If autonomous weapons systems incorporated privately owned technologies and databases, these decisions would inevitably be made, in part, by proprietary algorithms owned by the company.

Clearview rightly insists that its tool should complement and not replace human decision-making. The company’s CEO also said in a statement shared with openDemocracy that everyone who has access to its technology “is trained on how to use it safely and responsibly”. A good sentiment but a quaint one. Prudence and safeguards such as this are bound to be quickly abandoned in the heat of battle.

Clearview’s systems are already used by police and private security operations – they are common in US police departments, for instance. Criticism of such use has largely focused on bias and possible misidentification of targets, as well as over-reliance on the algorithm to make identifications – but the risk also runs the other way.