As the future of warfare pivots towards artificial intelligence, Ukraine is sitting on a valuable resource: millions of hours of footage from drones which can be used to train AI models to make decisions on the battlefield.
AI has been deployed by both sides on the battlefield during Russia's invasion of Ukraine to identify targets, scanning images far quicker than a human can.
Oleksandr Dmitriev, founder of OCHI, a non-profit Ukrainian digital system that centralises and analyses video feeds from over 15,000 drone crews working on the front lines, told Reuters his system had collected two million hours, or 228 years, of battlefield video from drones since 2022.
That will provide vital data for AI to learn from.
"This is food for the AI: If you want to teach an AI, you give it 2 million hours (of video), it will become something supernatural," he said.
Drone footage can be used to train AI models in combat tactics and spotting targets, experts say. (EPA PHOTO)
According to Dmitriev, the footage can be used to train AI models in combat tactics, spotting targets and assessing the effectiveness of weapons systems.
"It is essentially experience which can be turned into mathematics," he said, adding that an AI program can study the trajectories and angles at which weapons are most effective.
The system was originally made in 2022 to give military commanders an overview of their areas of the battlefield by showing them drone footage from all nearby crews side by side on one screen.
After the system was rolled out, the team running it realised that video being sent back by drones could prove useful as a record of the war – so they began to store it.
On average, Dmitriev said five or six terabytes of new data were added every day from the fighting.
Dmitriev said he was talking with representatives from some of Ukraine's foreign allies that had expressed interest in his OCHI system, but declined to provide details.
Samuel Bendett, adjunct senior fellow at the US-based Centre for a New American Security, said such a vast pool of data would be extremely valuable in teaching AI systems to identify what exactly they are seeing, and what steps they should take.
"Humans can do this intuitively, but machines cannot, and they have to be trained on what is or isn't a road, or a natural obstacle, or an ambush," he said.
Kateryna Bondar, a fellow at Wadhwani AI centre at the Centre for Strategic and International Studies, said the size of the data set and the image quality were important, as AI models learned to recognise targets based on shapes and colours.
Bondar said that the dataset was valuable in the context of training to fight Russia.
However, she said US officials and drone makers prefer a dataset that trains AI systems to operate in the Pacific against a potential Chinese adversary.
"(They want) systems ready and able to fight against China because that's the main priority for the US right now, rather than getting a lot of footage of Ukrainian fields and forests."
Ukraine also has another system, called Avengers, developed by its defence ministry, which centralises and collects video from drones and CCTV.
The ministry declined to provide information about this system.
However, it has previously said that Avengers spots 12,000 Russian pieces of equipment a week using AI identification tools.
Thousands of drones are already using AI systems to fly themselves into targets without human piloting, and Ukraine is using AI technologies to help demine its territory.
Ukrainian companies are developing drone swarms, where a computer system will be able to execute commands for an interlinked cloud of dozens of drones.
Russia has also touted its use of battlefield AI, most notably for target recognition in Lancet strike drones, which have proved lethal against Ukrainian armoured vehicles.