Scariest of all is a scenario in which a computer figures out both the advantages of collusion and how to make it happen. Here, the situation might resemble what happened with AlphaGo, the computer program developed to play the board game Go. The program’s success was mostly due to machine learning. The computer played countless games against itself and figured out what worked best. The end result is a black box: We don’t really know how the computer is making decisions, only that it works. Because successful collusion leads to higher profits, it would make sense that computers—left to their own devices—would figure this out. Antitrust authorities would have no way to punish this type of collusion under existing laws.
Virtual Competition The Promise and Perils of the Algorithm-Driven Economy Ariel Ezrachi and Maurice E. Stucke Harvard University Press, 2016. 364 pp.
Science 04 Nov 2016:
Vol. 354, Issue 6312, pp. 560