Recently, Russian military lorries in occupied Ukraine have been seen with a striking new colour scheme of vivid black-and-white stripes. Its purpose is to frustrate the machine-vision systems that are fitted to the Ukrainian drones looking for prey. They want to fool the drones into thinking that they are not lorries.

Machine vision is based on pattern-matching and is trained by exposing it to images, some of which contain lorries (or tanks, or aircraft) and some of which do not. Over millions of exposures, the computer deduces rules that allow it to identify the things its trainers want to teach it about. Because zebra-striped trucks are unlikely to appear in the training data, an AI that encounters one in the real world may not realise what it is looking at.

Computers often rely on very specific features to recognise an object. This can mean that small variations can cause bizarre conclusions. Testing is therefore important. One test, developed at MIT in 2017, featured a plastic turtle. Misled by the pattern on its shell, a computer confidently classified it as a rifle.

In the Ukrainian conflict, as part of their effort to confuse, parked Russian aircraft have been seen with rows of old tyres on their wings to confuse the image-matching software on drones. These tactics are likely to become more common.

At the moment most battlefield drones are flown by remote human operators, unlikely to be fooled by such disguises. However, as the drones proliferate and technology improves, AI will take on more of the work. And so there will be an arms race between increasingly sophisticated machine vision systems versus even cleverer methods for fooling them, with direct human involvement a rarity.

There is an attempt by the United Nations at the moment to incorporate laws of war into international law. This has been a long-term campaign led mainly by international lawyers. One of the earliest successes in that field was the treaty against laying land-mines. Princess Diana became well-know for her involvement with a charity promoting their clearance.

Why particularly land-mines? Well it’s because, once laid, whoever treads on one will be blown up. Except for the initial laying of the mine, there is no human agency involved in the chain of events. The mine is autonomous, although not in the sense of artificial intelligence guided autonomy such as that used by the Ukrainians.

The UN is however attempting to say that all automated warfare systems should be unlawful because there is no human agency involved in the final decision to attack a particular target. The AI decides based on its training data.

In light of Russian aggression against Ukraine, however, those hitherto peace-loving countries, Estonia, Latvia, Lithuania, Finland and Poland last year decided to withdraw from the land-mine treaty. Neither the US nor Russia have signed up to the treaty and Ukraine has ignored its provisions in its war with Russia. As has Iran with its mines in the Straights of Hormuz.

So then, when the chips are down, the idea of international law as a real constraint on a Nation’s actions is a bit over-blown.

The Bayeux tapestry has just arrived in London ready for viewing in the land where it was created. Legend has it that it was embroidered by William’s wife Matilda and her band of ladies-in-waiting in England and then taken over to France, presumably to Falaise, William’s stronghold.

However, it only became noticed in 1729 when it started to be displayed in the Cathedral in Bayeux. Why it turned up in Bayeux nobody quite knows, although another theory of its origin says that it was actually commissioned by William’s half-brother Odo – regent of England when William had other lands to conquer. And Odo’s other role was as Bishop of Bayeux.

What I find interesting is its depiction of the way war could then be justified and how warfare was carried on. As we have seen before, In the mediaeval Christian environment, moral justification had to be found for attacking someone-else. In this instance it was the alleged failure by Harold to abide by his oath that William should replace Edward the confessor upon his death.

Harold in effect replied, like the late queen, “recollections may vary” and so the battle of Hastings took place.

As we all know, legend has it that the death of Harold came about as a result of an arrow in the eye, although the tapestry seems to show someone-else as the unlucky victim.

The other lesson to draw, however, is that there were no rules of war in those days. Arrows are not accurate weapons. The average archer was neither a Robin Hood nor a William Tell. Arrows were fired in batches and rained down on the opposition. Once released by the archer, they flew wherever they went and killed anyone in their way. No human agency was involved.

But we know that the history of warfare is one in which there is an effort to keep your men away from harm. Hence the idea of long-range weapons, whether arrows or spears, rather than clubs or swords. And of course, these days bullets and missiles.

Bullets are mainly shot from sub-machine guns and so are not usually intended to kill a particular target but to spray out a hail of bullets to kill everyone who is in front of you.

Modern missiles, even those without AI, are supposed to be able, with the use of GPS, to find their targets accurately. The human agency is that of defining which target it is to be. Unfortunately, inherent inaccuracies mean that the effect of that agency diminishes with the distance between the site of the take-off and the target itself.

The human agency is in any event a rather dubious safeguard. As we saw at the beginning of the US – Iran war, a school was demolished by a missile and so many school-children killed because US intelligence told them that the building was used by enemy forces.

Intelligence is not all it should be: it is human, with all its errors. And shoddy intelligence is often relied upon as an excuse for not having to pause and so try harder to avoid the reckless killing of civilians – something already contrary to international law.

Which all rather explains the difficulty for the UN in defining what they actually mean by ‘autonomous weapon systems’. As we have seen, once launched, an arrow is an autonomous weapon system. The human agency is simply that of being on the battle-field and firing in the general direction of the enemy.

There seems to me to be very little difference between that and asking AI trained missiles to seek out pre-defined targets and kill them. If the training given to the missile is good enough, it may even mean that fewer unintended deaths will be caused than the inaccurate ‘point, shoot and hope’ type of weapons we have at the moment.

There is of course the sci-fi version of the modern killing machine incorporated in so many video games. Landmines are so last year. Instead, it would be sufficient to have autonomous drones on the ground which, when they sense the presence of people, will literally rise up and kill them. But until those weapons start autonomously manufacturing their replacements, the human agency would still be there, although hanging on by its finger nails.

But I think it’s this sort of thing that is behind UN thinking. It is trying to stop autonomous weapon systems roaming across the face of the earth killing at random. That, however, is not what anyone is trying to produce at the moment. It would be exceedingly wasteful: every weapon has to be manufactured at a significant cost.

The Ukrainians are showing that attacking sensitive logistical targets is a much better use of scarce resources. Cutting off supply lines and damaging oil refineries is a far better policy than simply attacking the civilian population – as the Russians to their shame are doing.

14 July 2026

Paul Buckingham