AI wars

Consider the lessons of the Ukraine war so far. First, the impact of drones goes far beyond legacy weapons. Drones have indeed rendered tanks and armoured personnel carriers extremely vulnerable, so Russian ground assaults now frequently use troops on foot, motorcycles, or all-terrain vehicles. But this hasn’t helped, because drones are terrifyingly effective against people as well. Casualties are as high as ever — but now, drones inflict over 70% of casualties on both sides.

Drones are also effective against almost everything else. Ukraine has used drones to destroy Russian targets as varied as weapons factories, moving trains, ammunition stores, oil refineries, ships, and ports. It could be worse; in fact, Ukraine has shown great restraint, considering Russia’s barbaric conduct. Airport terminals, train stations at rush hour, athletic and concert stadiums, pharmaceutical factories, hospitals, schools, nursing homes — all are equally vulnerable.

Two additional sobering lessons from Ukraine concern how drone warfare depends on its industrial base. First, speed and responsiveness are critical. Drone technology, weapons, and tactics now evolve at a blinding pace. A new drone will be useful for only two to six months. The other side develops countermeasures, requiring the development of new products, against which new countermeasures are developed, and so on.

At first, the drones used in Ukraine were crude weapons, radio-controlled by a pilot who needed to be nearby. As drones became more sophisticated and lethal, jamming was used to block their radio signals, which led to frequency changes and then frequency hopping, which was then countered by multi-frequency jammers, which then engendered drones that attack jamming equipment. Then Russia developed drones controlled via fibre-optic cable — impervious to jamming. Ukraine tries to track the cable to its source and kill the pilots, with drones. Now Ukraine has fibre-optic drones too.

Guidance is ever more sophisticated, so that drones can evade radar by flying very low or using stealth technology. But drone detection and tracking systems have also advanced, employing networks of cell phones and microphones connected to audio analysis software, as well as using Lidar, radar, and cameras.

In this ferocious environment, falling even a month behind is fatal. Normal defence industry procedures are totally inadequate, and most US drones and drone producers have proved to be hopelessly slow, expensive, and unusable. In response, however, Ukraine’s drone industry and military developed a revolutionary model of weapons research and development, production, and deployment, based on direct, continuous communication between frontline units and drone producers. Ukraine’s military command and Digital Transformation Ministry have even developed a points-based system that publishes continuously updated rankings of military units’ performance based on verified drone kills.

Here, Ukraine benefited from having a strong start-up ecosystem, which supports a weapons industry with hundreds of companies capable of designing, producing, and fielding a new weapon in a matter of weeks. This year, Ukraine will produce more than 4 million drones, most of them models that did not exist even a year ago. Unfortunately, Russia has adapted as well, also relying heavily on private start-ups.

Drone warfare in Ukraine provides yet another lesson for the United States and Europe: the need to address Chinese dominance of the global drone industry.

Ukraine evolved its own drone industry because the US and NATO had almost none of their own, much less one with the speed and flexibility required, and because China has gradually tightened supplies to Ukraine in favour of Russia. Some 80% of the electronics used in Russian drones are sourced from China. While Ukraine was initially highly dependent on China, it has reduced its reliance to perhaps 20%, most of that obtained covertly.

Yet US and European defence R&D and procurement remain slow and uncompetitive, which cripples their ability to defend against drones, as well as their ability to use them. Although few people realise it, the US and NATO now desperately need Ukraine for its drone expertise. Ukraine is now the only country that could possibly match Chinese and/or Russian technology and reaction time in a war. Without Ukraine, and without modernising their own forces, NATO and the US would suffer horrific casualties in a war with Russia or China, and might even lose.

Moreover, AI will change everything. Ukraine’s 1 June operation used 117 drones, each controlled by a skilled operator, and reports suggest that something like half were defeated by Russian defences — jamming, mainly — because the drones needed to be in radio communication with their controllers. Had they been autonomous, there could have been a thousand of them. And Ukraine evolved its own drone industry because the US and NATO had almost none of their own, much less one with the speed and flexibility required Five years from now, it will be terrifyingly easy to launch pre-emptive strikes on conventional targets.

AI also increases the lethality and precision of drones used against people. Chinese researchers have already demonstrated drone swarms navigating through a forest and then re-forming as a swarm after passing through. This is not just about warfare; it also works for terrorist attacks. 

True, the required AI functionality still demands far more computing power and memory than can be put into a small drone. Nor is it cheap. Nvidia chips, for example, cost up to $50,000 each, so even one powerful AI processor would make most drones prohibitively expensive.

But that’s changing fast, driven by the goal of putting serious AI capability into every phone. When that happens, those same capabilities will be available to every drone weapon. And with the sole, vital exception of AI processors, the entire supply chain for both phones and drone weapons is dominated by China.

Stuart Russell, an AI specialist at the University of California, Berkeley, has long argued for an arms-control treaty to prevent the spread of small, mass produced, AI-controlled drone weapons. He even underwrote the production of a short film, Slaughterbots, which dramatises the risks these drones could pose in the wrong hands. At a dinner years ago, he told me that it would soon be easy to target individuals using facial recognition or, say, everyone wearing a cross, a yarmulke, or any other religious or political symbol.

Since any meaningful treaty is unlikely in the current geopolitical environment, we must prepare for a world that probably will contain such weapons. But the Western defence establishment increasingly looks like the typical “legacy” company that has been caught off guard by technological disruption. In markets, legacy resistance can be costly, but the costs are purely monetary. In warfare, they can and will be deadly.