Roskomnadzor, Russia’s internet watchdog, is creating a neural network called Oculus that will use artificial intelligence (AI) to scan websites for banned information.
The automatic scanner will analyze images, videos, chats, and URLs on forums, social media, websites, and even messenger channels to detect material that should be taken down or redacted.
Information targeted by Oculus includes misinformation that discredits official state and army sources, homosexuality propaganda and instructions on manufacturing weapons or drugs. The system will also look for expressions of disrespect for the state, calls for mass protests, and signs of terrorism and extremism.
Read More: Meta AI Releases Implicitron For Representations In PyTorch3D
Oculus’ real-time scanning capacity will be about 2.3 images per second or 200,000 images per day. For this, Eksikyushn RDC LLC will utilize 48 servers with solid GPUs. Oculus will be integrated onto a network of monitoring systems, the Unified Analysis Module, which is currently under development. The aim is to give the government a grip on controlling information flow.
According to Kommersant, Oculus will cost $965,000 (Russia 57.7 million rubles) and must be completed by December 2022. However, experts in the field suggest that the amount is unlikely to cover the cost of achieving such an aspiring project. Roskomnadzor will likely have to allocate more funds along the way.
Introducing Oculus will result in users either giving up on taking part in online discussions or using anonymization tools like the Tor network, a VPN. People may also resort to chatting apps that do not require PII when registering or logging user data.