Artificial intelligence could catch antitrust laws flat-footed, CMA’s Currie says

3 February 2017. By Matthew Newman.

Competition law may struggle to tackle antitrust issues arising from the use of artificial intelligence in pricing algorithms by online retailers, according to the UK's top competition official, David Currie.

Speaking at a conference* in London, Currie said that competition authorities may not have the right tools to confront pricing algorithms that react quickly to changes in competitors' prices.

"The rise of the algorithmic economy raises difficult questions for competition policy," said Currie, the chairman of the UK's Competition and Markets Authority.

"I [previously] suggested that the competition tools at our disposal can tackle the competition issues we face," he said. "Perhaps this is one area that it's not true."

Anticompetitive behavior may arise when the prices of several online retailers align through the use of pricing algorithms — which rely on information about potential buyers and competitors' prices to maximize the seller's profits. The algorithms may also lead to higher prices, Currie said, by aligning prices based on the highest price.

The situation can be made worse when artificial intelligence automatically leads to price coordination, the official said. "Machine learning means that the algorithms may themselves learn that coordination is the best way to maximize longer-term business objectives," he said.

In other words, artificial intelligence may choose to adopt anticompetitive behavior without the consent or, even, the knowledge of company managers.

But that does not mean competition law is totally powerless when facing algorithms, Currie told the conference, saying that authorities have been able to pursue antitrust violations when individuals abuse pricing algorithms to form a cartel.

This was the case in 2015 when the US Justice Department fined an online retailer that conspired to fix the prices of posters sold online.

But when it comes to artificial intelligence, Currie said the problem for antitrust authorities is that there may not be a "human agent" who planned the coordination.

"Does that represent a breach of competition law?" he said. "Does the law stretch to cover sins of omission as well as sins of commission?"

The regulator asked the conference whether the retailers should be pursued for failing to "build in sufficient constraints on algorithmic behavior" so that the algorithm doesn't adopt anticompetitive outcomes.

Another scenario is that the constraints are built in, but they are "inadequately designed" and the algorithm ultimately learns a way to coordinate prices, he said.

"How far can the concept of human agency be stretched to cover these sorts of issues?" the chairman said.

Currie said that the message from these challenges for competition authorities was that regulators needed to keep up with fast-moving technology.

* "Innovation Economics for Antitrust Lawyers," Concurrences Review and King's College, London, Feb. 3, 2017.

	Eliot Gao