The Higher School of Economics (HSE) has unveiled an AI model designed to detect collusion in state auctions. The system successfully identified cartels in over 90% of cases, reports Kommersant.
The current antitrust control system faces challenges due to the scale of the public procurement market. In 2023, 2.4 million contracts were signed, amounting to 10.6 trillion rubles.
Konstantin Efimov, a graduate student at HSE, developed a neural network capable of predicting signs of collusion in auctions with 91% accuracy. The model was trained using data from 89,000 auctions of state-owned companies between 2016 and 2020, along with over 1,100 decisions by the FAS on cartel cases.
The model can be applied immediately after an auction concludes, relying primarily on participant behavior during the initial bids.
The integration of automated procurement analysis could simplify the detection of violations. Although auction participants might attempt to alter their strategies to deceive the algorithm, this issue can be addressed by adding additional parameters, the author noted.
Media outlets have highlighted potential risks for honest businesses in automating cartel detection. The stability of the algorithm’s performance depends on the quality of the data used during training.
The FAS has expressed support for the development of information technologies.
“The implementation of such tools is intended to identify risk scenarios that indicate possible signs of the conclusion and implementation of anti-competitive agreements, which does not imply automatic decision-making on violations,” the agency noted.
The regulator emphasized that conclusions regarding violations of antitrust laws are based on a combination of facts and evidence.
Back in September, Russia was suspected of using AI to influence elections in the United States.