Posted on 27 November 2020
Price-setting algorithms can lead to noncompetitive prices, which the law cannot stop, researchers say in Science
Artificial Intelligence (AI) behind online platforms steers anything from the news we read to the music we listen to. It also determines the prices we pay for online purchases. As pricing decisions are increasingly delegated to algorithms, consumers risk to pay higher prices.
In traditional markets, collusion is mainly detected by observing employees of different firms communicating with the intent of agreeing on price-setting. But algorithms determining prices autonomously opens a loophole lawyers, economists and public officials worry about.
Prof Giacomo Calzolari
Researchers including the European University Institute’s Professor of Economics Giacomo Calzolari discuss the problem in scientific journal Science. They used simulations to study algorithms based on artificial intelligence systems that learn autonomously through active experimentation. Even relatively simple algorithms tend to converge towards high prices, supported by reward and punishment schemes typical of cartels between companies.
“Antitrust law and its enforcement currently rely on communications between firms, rather than on collusive pricing rules or higher prices. That leaves us empty handed if collusion is caused by algorithms without human intervention,” said Prof Giacomo Calzolari.
As firms’ pricing algorithms can be audited and tested, Calzolari and his co-authors propose that antitrust policy should “shift its focus from communications and intention (with humans) to rules of conduct (with algorithms)”.
The researchers invite the scientific community to experiment “in vitro” to identify factors that can counter collusive strategies.
“One can draw a comparison with clinical trials for new medicines. Given the potential side effects of self-learning algorithms, they would need to be tested before being released on the market,” added Prof Calzolari. Algorithms tending towards collusion would be put on a blacklist.
In their paper, the researchers emphasise that computer scientists, economists and legal scholars would need to join forces to protects consumers. “Economists and computer scientists are crucial in the process, but in the end legal scholars should use the knowledge in the context of prosecution and prevention of collusion,” concluded Calzolari.
Prof Calzolari co-authored the Science paper with researchers of the University of Bologna (Italy) and the University of Pennsylvania (USA).