Understanding AI Collusion and Compliance

Artificial Intelligence and Competition Policy and Antitrust

Article Snapshot

Author(s)

Justin Johnson and Daniel Sokol

Source

Chapter in Cambridge Handbook on Compliance, D. Daniel Sokol & Benjamin van Rooij, eds., 2021

Summary

Artificial intelligence (AI) allows firms to adopt new types of anti-competitive behavior, but may also aid in the detection of such behavior. AI collusion could include non-price elements, such as product reviews and ratings.

Policy Relevance

Firms and competition authorities can use AI to prevent collusive behavior.

Main Points

  • Collusion is said to have occurred when prices are higher than they would be if the players interacted with one another only in the short run, rather than interacting on a long-run basis.
     
  • Some observers posit that easy price monitoring combined with the ability to rapidly change prices will foster collusion; AI might simply speed up this process.
     
  • Firms and regulators should consider the possibility of different types of AI collusion.
     
    • Algorithms directed to maximize profits might learn to collude without the direct involvement of humans.
       
    • Humans could intentionally design algorithms to collude.
       
  • AI makes it easier for vendors, intermediaries, and other disruptive players to quickly enter and exit markets, challenging collusive attempts to raise prices.
     
  • Humans could choose an algorithm to facilitate collusion in a traditional hub-and-spoke conspiracy; alternately, humans could design AI to support collusion without any conspiracy, programming in behaviors that punishes rivals for lowering prices.
     
  • AI could help conspirators trust one another by reducing the number of employees in any firm who need to be involved in the collusion and who might report it.
     
  • “Screens” are mechanisms that identify anticompetitive behavior using data such as prices and market share; some look for improbable events, while others compare the questionable behavior to that of a control group.
     
  • AI screens could help detect the price effects of collusion; AI could also help identify firms that use fake reviews to manipulate their own ratings or those of rivals.
     
    • In a successful cartel, prices increase and the number of fake negative reviews should fall.
       
    • An unexplained fall in average ratings could mean that the cartel is punishing rivals.
       
  • AI screens should be used only when the amount and quality of data is high.
     
  • Procurement auctions involving government or large commercial buyers would be a good place to begin using AI collusion screens.
     

Get The Article

Find the full article online

Search for Full Article

Share