Ukraine is amassing a significant repository of war data to enhance the training of AI models

As warfare increasingly incorporates artificial intelligence, Ukraine possesses a significant asset: extensive drone footage that can be utilized to train AI models for battlefield decision-making. Both Ukrainian and Russian forces have employed AI technology during the ongoing conflict to rapidly identify targets, processing images at a speed unattainable by human operators.

Oleksandr Dmitriev, the founder of OCHI, a non-profit organization in Ukraine that consolidates and analyzes video feeds from over 15,000 drone teams on the front lines, informed Reuters that his system has amassed 2 million hours—equivalent to 228 years—of drone footage since 2022. This extensive collection offers crucial data for AI learning.

“This is nourishment for AI: Providing it with 2 million hours of video can elevate its capabilities to extraordinary levels,” Dmitriev stated. He emphasized that the footage is instrumental in training AI models for combat strategies, target identification, and evaluating weapon system effectiveness. “It represents experience that can be translated into mathematical models,” he explained, noting that AI can analyze the trajectories and angles that optimize weapon performance.

Initially developed in 2022 to give military leaders a comprehensive view of the battlefield by displaying drone footage from various crews simultaneously, the system’s utility expanded as the team recognized the potential of the recorded video as a historical archive of the conflict. Dmitriev mentioned that, on average, the system receives five to six terabytes of new data daily from ongoing combat activities.

IMAGE QUALITY

Dmitriev mentioned that he has been in discussions with representatives from several of Ukraine’s foreign allies who have shown interest in his OCHI system, although he refrained from sharing specific details. Samuel Bendett, a senior fellow at the U.S.-based Center for a New American Security, emphasized that such a large dataset would be invaluable for training AI systems to accurately identify objects and determine appropriate actions.

“While humans can intuitively make these distinctions, machines require training to recognize what constitutes a road, a natural barrier, or an ambush,” he explained. Kateryna Bondar, a fellow at the Wadhwani AI Centre at the Centre for Strategic and International Studies, highlighted the significance of both the dataset size and image quality, as AI models learn to identify targets based on their shapes and colors.

Additionally, Ukraine has developed another system known as Avengers, created by its defense ministry, which consolidates and gathers video feeds from drones and CCTV. The ministry has not disclosed further information about this system, but it has previously stated that Avengers identifies 12,000 Russian military assets weekly using AI identification technologies.

Numerous drones are already equipped with AI systems that enable autonomous flight towards targets without human intervention, and Ukraine is leveraging AI technologies to assist in demining efforts. Ukrainian firms are also working on drone swarms, where a computer system can command a network of dozens of drones. Meanwhile, Russia has also promoted its use of battlefield AI, particularly for target recognition in Lancet strike drones, which have demonstrated effectiveness against Ukrainian armored vehicles.