Open-source investigations are often characterised by the exhaustive, long, manual processes of pouring over online information.
We were working on a project just like this, trying to trace the story of a Russian military unit.
In October 2022, Russia sent men from the elite 155th Marine Brigade—the ‘Black Berets’— to take the town of Vuhledar in Ukraine. Ramaz Gorgadze, a young aspiring TikTok rapper, was among them. Before the assault, he sent a last message to his mother, and then went missing.
Days later, a rare protest letter from the Black Berets emerged online. The men claimed that ‘inept’ commanders threw them into an ‘incomprehensible offensive’ and called them ‘meat’. The Russian authorities denied any unnecessary loss of life.
But who was right? What happened to Ramaz and his comrades of the 155?
Using only publicly available footage and testimonies, BBC Eye’s 28 minute film showed how glaring tactical mistakes caused one of the highest concentrations of losses for a single Russian unit since the beginning of the war.
Our analysis of over 300 videos, combined with semantic searches on Telegram groups, showed that the soldiers were using human-wave tactics to take Vuhledar and were severely lacking critical equipment such as thermal imaging. We showed the materials to military experts who confirmed the findings with one explaining on camera to the audience how this led to the unit’s failings in Vuhledar.
This included matching people’s death records with images from the training ground to confirm that they were part of this unit, as well as pouring over hundreds of social media messages to identify people who were speaking out about this or reporting deaths and the conditions of the soldiers.
To cope with the volume of images, videos, and social media posts, we developed internal tools for searching local file systems using facial recognition to track soldiers between videos and images, and created pipelines of analysis using chatGPT which analysed the comments, looking for claims about the conditions of the soldiers, and reports of their deaths.
This project is notable for its advanced integration of traditional AI with generative AI tools. By combining facial recognition technology with ChatGPT, the project created a powerful pipeline for analyzing vast amounts of footage and social media data. This innovative approach not only streamlined the investigative process but also uncovered detailed insights into the military events. This seamless blend of AI technologies exemplifies cutting-edge innovation.
