Artificial intelligence against collusion: What (not) to expect?

An information system designed to detect collusion, primarily at auctions, using algorithmic methods called artificial intelligence (AI) is currently being created in Russia. The article is devoted to the possibilities and limitations of using AI to identify public procurement where participants engage in illegal agreements. It describes the evidence used in cases of bid rigging cartels in the court decisions regarding claims to annul decisions of the Federal Antimonopoly Service of the Russian Federation during 2015—2018. The article summarizes the results of the development of algorithms predicting collusion, depending on a certain set of parameters based on ex-post data on detected cartels both in Russia and abroad. Three conclusions are justified. First, no algorithm or system of algorithms currently provides a sufficient level of reliability. Second, the hypothetical AI training sample for identifying collusion in Russian public procurement is biased towards a specific procurement method (electronic auction) and therefore does not help predicting collusion in purchases conducted in another way. Third, the signs of cartels in Russian public procurement are largely the result of flaws in the procedures used.