IT HAS become a cliché to note that the war in Ukraine is a drone war. But two recent studies shed light on what that means in practice. In mid-February the Royal United Services Institute (RUSI), a think-tank in London, published the latest in a series of papers taking stock of tactical developments in Ukraine over the preceding year. On March 6th the Centre for Strategic and International Studies (CSIS) released another paper looking at Ukraine’s capacity and plans for war specifically involving artificial-intelligence (AI) tools. Together they paint a picture of a battlefield that is increasingly saturated with and dominated by the presence of uncrewed machines.
The first observation is that drones have become the most lethal weapons in Ukraine. “Tactical” drones, those with ranges in the low tens of kilometres, are now responsible for 60-70% of damaged and destroyed Russian systems, says RUSI. A growing proportion of these are equipped with AI guidance, allowing them to lock on to targets in the final phase of flight even if the link between pilot and drone is jammed. The automatic guidance can kick in at distances of 2km or more, depending on conditions, notes CSIS, and can raise the hit rate from 10-20% (for manually piloted drones) to 70-80%. That means that one or two drones can do work that would previously have taken eight or nine. AI can also counter decoys and camouflage that would trick humans.
The second is that Ukraine plans to go much further. Its aim, says CSIS, “is to remove warfighters from direct combat and replace them with autonomous unmanned systems”. That includes not only aerial drones, but also their equivalents on the ground and at sea. Thirty-three ground robotic systems were approved in the first nine months of last year. And in December Ukraine carried out what it claims was the first fully uncrewed operation near Lyptsi, a village north of Kharkiv, in which dozens of remote-controlled robotic ground vehicles fired machine guns and cleared mines.
A third finding is that Ukraine has embraced the idea of software-defined weapons, whose operating code matters more than their physical design. Ukrainian producers make “modules”, typically chips loaded with software that are smaller than a bar of soap and which can slot into a wide range of different platforms, including drones, vehicles or gun turrets, to enable target recognition or other tasks. Advanced capabilities can thus be fitted or retrofitted to cheap and mass-produced hardware. Should it fall into Russian hands, the hardware might be reverse-engineered but the encryption on the chip would take valuable time to unpick.
Ukraine is also attempting to be efficient in its use of AI. Newer models can be trained quickly on relatively small amounts of data. CSIS gives the example of Zvook, Ukraine’s acoustic-detection system for drones, which has won plaudits from NATO leaders. When a novel type of drone or other weapon turns up on the battlefield, the model can be trained to recognise its sound with just a week’s worth of data.
Fourth, there are still limits to all this. Some 60-80% of Ukrainian “first-person view” or FPV strike drones fail to reach their targets, writes RUSI, depending on the pilot’s skill and target location. (For remote-controlled drones, which need a radio signal, swarm attacks are difficult to pull off because the signals tend to interfere with one another.) Of the 20-40% of FPV drones that do get through, a majority fail to destroy armoured vehicles—though they are good at wounding infantry, which helps explain the astronomically high Russian casualty numbers. Roman Kusiv, the medical commander of the eastern and southern front, tells The Economist that more than 50% of injuries are caused by drones, up from 25% at the end of 2023.
These figures must be put in context. Though it is true that drones now inflict more casualties than artillery, notes RUSI, this is in part because Ukraine is short of artillery pieces and explosive charges for shells. Moreover, it is still the interplay between drones and guns that often makes the difference. An FPV drone might immobilise a vehicle, for instance, with shellfire used to kill the infantry who dismount. Drone operations can also be time-consuming and complex. One Ukrainian officer tells RUSI that it took “hours” to halt a tank with an FPV drone, compared with the two minutes needed to knock out three tanks with five precision-guided anti-tank shells after they were spotted by a drone. Long-range strikes are especially convoluted. Ukrainian attacks on Russian oil refineries using the long-range Lyutyi drone involve a 15- to 20-page planning document and “meticulous preparation”, notes CSIS.
Although AI models are performing a growing range of military tasks, humans are still closely involved in the decision to use force. Ukrainian personnel can “override autonomous functions” when needed, notes CSIS. In ground systems, which face a more cluttered and complex environment than aerial ones, autonomy “remains largely unexplored by Ukrainian defence companies”. This means that one of the greatest potential benefits of automated combat—a reduction in human casualties—is some way off. With the pervasive drone threat keeping machinery at least 7km behind the front line, observes RUSI, soldiers have to dig trenches using picks and shovels. Some minefields are still cleared by hand. The grim irony is that AI and robotics have produced a more lethal battlefield for the men unfortunate enough to be deployed to its edge. ■
Stay on top of our defence and international security coverage with The War Room, our weekly subscriber-only newsletter.