Roskomnadzor to Test AI-Based Video Lie Detection System
The National Center for Cognitive Developments at ITMO University is developing an artificial intelligence (AI) service called Expert for the Main Radio Frequency Center (GRChTs), which operates under Roskomnadzor. This service will be able to analyze video recordings and detect lies and manipulations within them, according to sources and representatives from ITMO, Roskomnadzor, and GRChTs, as reported by Kommersant.
The Expert service is also designed to quickly identify “deepfakes”—that is, altered elements in original videos used to create fake footage featuring statements by politicians and celebrities.
One source revealed that a prototype of the service was presented to Roskomnadzor during a meeting of the GRChTs expert council on AI in July. The head of Roskomnadzor, Andrey Lipov, who attended the meeting, expressed interest in the solution and requested it be provided to his agency for testing.
How the Expert System Works
The Expert system analyzes video or audio content by evaluating confidence, levels of internal and external aggression, congruence, and contradictions. It also compares spoken words with scientific articles and statements from other experts.
Oleg Basov, head of the “Cognitive Nonverbal Communication” research lab at ITMO’s National Center for Cognitive Developments, explained that English was chosen as the primary language because it is the most widely used: “Adapting the system to extract text from Russian speech is planned for the near future.” The system underwent preliminary testing ahead of the U.S. Senate elections, according to Basov: “We analyzed YouTube interviews of six candidates for signs of contradictions, aggression, and self-confidence.” He noted that the system detected hidden aggression in the statements of three candidates and identified one as lying.
Challenges and International Context
Currently, over 90% of Russian AI developments are based on open-source foreign libraries and datasets, according to a top manager at a specialized IT company: “That’s why ITMO’s development is adapted for English. Retraining it for Russian will require a large amount of data and significant resources.”
Similar projects have been developed in the West. In 2017, the University of Maryland and Dartmouth College in the U.S. created a neural network for lie detection, trained on video recordings of court proceedings. The American Innocence Project has also used AI for lie detection in court, utilizing Amazon Mechanical Turk for analysis.