Much has already been said about the potential applications of artificial intelligence (AI), from healthcare, to self-driving cars and beyond, but with the shift of labour comes an accompanying ambiguity in legal liability. Can AI algorithms commit anti-trust violations for example? If they do, who should be held legally responsible for the anti-trust violations they commit? Should self-driving cars decide which life to prioritize in the case of a crash? Is the car manufacturer responsible for the consequences of these decisions?

It is sometimes easy to overlook less common scenarios when programming an algorithm, as humans think differently to machines. As the level of complexity increases, humans can be unaware what exactly they are telling machines to do in certain situations and this is true too of the data being fed into AI algorithms.

Eventually an increasing body of case law will build up around the deployment of algorithms and the associated legal liability, but for now, there is much that is yet to be decided. This means that companies should plan for potential regulatory risks when assessing their business’ future and particularly when making the decision to deploy AI in their business.

Jon Jacobs, a former State’s Attorney in the Antitrust Division of the U.S. Department of Justice and now a partner with Perkins Coie, was at a conference recently in Taipei, and talked about the implications of AI specifically in reference to antitrust provisions.

In the U.S. the most important law that applies to anti-trust violations is Section 1 of the Sherman Act and Section 5 of the FTC Act.

Under Section 1 of the Sherman Act, in cases where there are horizontal agreements on prices or price formulas in place between two or more persons or companies, this constitutes a per se violation, which is usually prosecuted as a criminal violation. Agreements to exchange pricing information are only unlawful after thorough investigation and normally only result in civil penalties. If the agreement is vertical (between a customer and a supplier) as opposed to horizontal (between one of more competitors) it is almost always treated as a civil violation, rather than a criminal one.

Section 5 of the U.S. FTC Act also prohibits “unfair or deceptive acts or practices in or affecting commerce.”

Taiwan has a very similar set of per se violations to the US, when it comes to price fixing.

Jacobs pointed out five situations wherein the parties run the risk of getting in trouble with the law.

1. Agreeing to Fix Prices and Implementing this via Algorithms

This is clearly a case of an act which is per se unlawful. The first case to deal with this kind of e-commerce activity in the U.S. was a Department of Justice case in 2015. The case involved a man called Topkins who was selling posters through the Amazon Marketplace. Topkins admitted to agreeing with a competitor to “stabilize” price competition between them. He then implemented the price agreement using an algorithm. He pled guilty and served 6-12 months in prison.

2. Using the Same Algorithm as a Competitor

It’s also per se illegal for competitors to agree to use the same algorithm, whereby the algorithm sets the prices to maximize profits for each competitor in cases. This does not necessarily have to be an explicit agreement.

3. An Algorithm Supplier Initiates and Organizes a Price Fixing Agreement

If a supplier sells an algorithm which can set prices whilst giving prospective customers assurances about competitors using the same algorithm, both the supplier of the algorithm and the competitors could be liable. This depends on the evidence available as to the existence, direct or implied, that an agreement was in place.  The strength of evidence would dictate whether it would be only a civil penalty or criminal too. If it can be shown that there was a de facto agreement among the competitors facilitated by the third-party, they would be liable.

Even with no communication between competitors, there is a precedent for legal liability via a third-party. There was a previous case where the U.S. Fair Trade Commission investigated Toys’R’Us with regards to their attempts to pressurize toy suppliers to restrict supplies to their competitors. The company had allegedly gone around the different toy manufacturers attempting to prevent them from selling to low-cost outlets such as Walmart. The toy suppliers in turn sought reassurance that their competitors would also stop selling to other outlets and Toys’R’Us told them they were in contact with other toy manufacturers and had secured promises from them too. The FTC still found this to be an agreement between the competitors, even though they had not been in direct contact with one another. This was only a civil case, but this does not necessarily mean that if evidence was sufficient that a criminal case could not be brought.

4. Competitors Decide to Use the Same Algorithm Independently

Although competitors independently choosing to use the same algorithm is not a per se violation and does not violate Section 1 of the Sherman Act at all, companies should ensure there is no direct or indirect communication between them which might infer a common understanding. However, it’s not likely, said Jacobs, that the FTC would take an action under Section 5 of the Federal Trade Commission Act.

5. Independent Use of Algorithms on the Part of Competitors Leads to Higher Pricing

Algorithms track competitor pricing and adjust accordingly, so coordination can be a likely consequence.

This can lead to “follow the leader” pricing but this is not unlawful without an agreement, which Jacobs compared to gas stations which display their prices and so are always aware of each other’s prices and will tend to follow each other without any agreement in place. Even if it’s not unlawful, programmed limits or human intervention is sometimes necessary for algorithms when seller prices may be tied to each other, without either company being aware. Jacobs pointed to the example of an obscure textbook called The Making of a Fly, which was able to reach a price of US$23.6 million as two sellers had employed algorithms which set the prices they were selling at as a percentage of each other, where one was 0.9983 of the other’s price, and the other was 1.27 times the first seller’s price. This is why it’s necessary to monitor pricing, even if algorithms purport to be able to save labour.

Although Jacobs said that it is unlikely there will be any substantial changes in the law in the near term, he stated that the FTC established the Office of Technology Research and Investigation in 2015, which will provide guidance on “algorithmic transparency”. He also suggested that the Department of Justice and the FTC may be more proactive in blocking mergers in the IT market to prevent an algorithm monopoly. There may also be increased enforcement under Section 5 of the FTC Act, he stated.

More legal precedents will have to emerge before there is a clearer understanding of the legal liability for AI algorithms, particularly if two agreed to fix prices together without human involvement. With the various inputs of machine learning, it certainly suggests that the waters would be quite muddy on this issue.

This goes on to the larger issue of responsibility for the design of software and algorithms where the issues were not necessarily foreseen or directly programmed for. This was seen with the recent self-driving car crash, where a pedestrian was visible to the car’s sensors, but was taken as a false positive. This points to the compromises that have to be made to pursue efficiency and user experience. However, before making these trade-offs, the secondary consequences of them, and the legal risks should be examined with care. No-one wants to be doing time for the crimes of a badly programmed AI algorithm.

Main image of article © ITU/R.Farrell, under a Creative Commons License

Author: Conor Stuart Current Post: Senior Editor, IP Observer Education: MA Taiwanese Literature, National Taiwan University BA Chinese and Spanish, Leeds University, UK Experience: Translator/Editor, Want China Times Editor, Erenlai Magazine

Facebook Follow the IP Observer on our FB Page