Russian Neural Networks Could Become Smarter: Soviet Scientists' Works May Help

In the USSR, a state system for scientific and technical information (GSNTI) collected nearly all Soviet scientific output, from physics to biology, amassing over two billion items by 1980. These data have remained largely unused for decades. Now, Russia is digitizing these archives to train artificial intelligence, as current data for training neural networks is insufficient. The digitization process involves converting old paper and microfilm documents, including scientific reports, drawings, and formulas, into machine-readable formats. As of mid-August, 483 million items have been confirmed for digitization, which will be combined with modern research data from the GosTech platform dating back to 1998. The goal is to make Russian AI models specialized experts capable of providing evidence-based consultations to engineers and scientists, and to achieve technological independence by relying on internal archives inaccessible to other countries. Deputy Prime Minister Dmitry Chernyshenko has called for updating GSNTI regulations to give algorithms legal access to this knowledge, turning forgotten archives into a driver of future technological breakthroughs.