AI Models to Be Trained on the Work of Soviet-Era Russian Scientists
Russia plans to digitize its Soviet scientific heritage and use it to train domestic AI models. This could significantly expand researchers’ capabilities and help advance Russian science.
The country’s state database currently contains about 1.5 million articles and other scientific materials from the Soviet era, but archives hold more than 2 billion. Of those, 600 million are expected to be added to the central database soon.
“We are digitizing the collections held by the state system of scientific and technical information. Our main task is to digitize all of our Soviet scientific heritage. The goal is to use this material to train our sovereign AI models,” Russian Deputy Prime Minister Dmitry Chernyshenko said.
AI Services for Science
The quality and volume of training data are among the key factors driving modern AI development, alongside computing power and algorithms. Russia is now set to gain a unique body of scientific knowledge that could support AI services for research, engineering, education and industry. This work could strengthen the position of Russian scientists and expand the range of problems they can solve with AI.
Once the materials have been digitized, scientists, engineers, teachers and students will have easier access to scientific literature and records. The new AI models could also underpin intelligent search engines for scientific information, tools for analyzing publications and systems that automate parts of the research process.
Russia is therefore building foundational infrastructure for domestic AI development. In the future, scientific AI assistants could help researchers identify connections across studies, analyze archival publications and develop new hypotheses. Optical character recognition (OCR), semantic search and document analysis could turn archives into machine-readable knowledge bases.
Digitization Reaches the Archives
In 2022, the Federal Archival Agency began a digital transformation initiative, creating a platform for storing electronic documents and starting to convert archival materials into digital formats. Plans also called for an information system that would enable remote access to archival documents and use AI to help users find relevant records. This marked the beginning of efforts to prepare Russia’s archival collections for AI applications.
In 2023, Yandex launched Poisk po arkhivam (“Search the Archives”), a service that uses neural networks to recognize historical documents and convert handwritten materials into machine-readable text. The service continues to evolve, while the digitization of additional documents is expanding access to historical information.
Libraries and historical newspapers have also been undergoing digitization. In 2025, Yandex partnered with the Izvestia newspaper and the National Electronic Library to launch a digital archive of the publication. Neural networks were used to process more than 30,000 issues published between 1917 and 2024. These projects show how AI has become an essential tool for working with large collections of library and archival material.
A Foundation for the Next Level
Digitizing the Soviet scientific heritage marks a shift from traditional archival storage to building a national knowledge base for artificial intelligence. The project’s major challenge is to make that knowledge available in a format suitable for training AI algorithms.
Russian scientists are already making extensive use of AI, but the systems’ knowledge remains limited. Expanding the shared pool of training data is essential to improving the quality of AI models. Over time, these databases could become the foundation for specialized Russian AI systems for scientific research, helping take the country’s scientific capabilities to a new level.
