What is a cat? The answer may seem intuitive, but giving a clear definition is more complex than you think. Will you define it by the number of ears, fangs or legs? The presence of fur or whiskers? It may be easy to distinguish between a cat and a panther, but will that be thanks to your definition or is that because of what you have learnt? Today, thanks to advancements in artificial intelligence, computers can do it better and faster than you can.
From Google recommendations to self-driving cars, the AI (Artificial intelligence) revolution has and will impact many aspect of our lives. But what about competition policy? Should the authorities worry about this evolution?
Several competition experts have given this some thought. In 2016, two lawyers, Stucke and Grunes, published the first book on competition policy and big data. This was followed by two OECD round table discussions. In 2017, the head of the federal trade commission, Maureen Ohlhausen, gave a speech on the interaction between anti-trust law and algorithmic pricing. The most challenging issue faced by competition authorities is when algorithms facilitate collusion or in the more futuristic example when AIs decide to collude. This raises a number of issues, with liability and detection being the two most pertinent.
To determine the effects on competition, it is important to first distinguish between AIs and algorithms. An algorithm is a structured method. A step-by-step instruction guide. Computers run algorithms, since they are very good at following clear instructions, and doing exactly what they are told. Artificial Intelligence is a research field in Computer Science. It investigates how to construct specific algorithms that behave in a way that can be deemed intelligent in some way or other. Currently, we speak about AI when the algorithm has a deep learning component. Deep learning is a statistical method with networks, that can learn from unstructured or unlabeled data. It allows making prediction but does not reveal the statistical relationship established between the data. The program relies on a multiplicity of small elements (node and neural) taking a number as value, and the general picture is almost never interpretable. In that sense, deep learning is a black box, as opposed to other statistical methods that estimate the parameters of an explicit model that are easier to interpret.
Coming from these definitions, there are two ways for an algorithm to affect collusive behavior. First, the algorithm can provide more information to the firms, helping them to collude. For instance, an algorithm can screen websites of competitors and immediately detect a change in price. This may increase reaction speed to a potential deviation from a collusive agreement, and therefore help sustain the agreement. For such algorithm, it may not be pertinent to speak about AI, but more about statistical algorithm. In that way, the introduction of deep learning method does not change the mechanisms and thus the liable parties of the collusive agreement.
However, AI may change collusive practice in another respect. Indeed, the algorithm can directly choose your strategic variables like price or quantity. For instance, a company can sell price recommendations to firms competing on the second hand car market. By setting an algorithm that maximizes the joint profit, the price selling company can make the second hand car market collusive. Obviously, this effect is not new but the utilization of deep learning can make the algorithm unreadable even for an expert. In that case, the responsibility of the firms externalizing their pricing decision is questionable. Did they know that the algorithm would implement collusion?
If these practices are not already here yet, the quick progress of AI may make it arrive sooner than we think. Competition authority and academics should be aware.
Rossouw van Jaarsveld and Thomas Fagart
OECD (2017), Algorithms and Collusion: Competition Policy in the Digital Age www.oecd.org/competition/algorithms-collusion-competition-policy-in-the-digital-age.htm
OECD (2016), Big Data: Bringing Competition Policy to the Digital Era https://one.oecd.org/document/DAF/COMP/M(2016)2/ANN4/FINAL/en/pdf
Maureen K. Ohlhausen, (2017), Should We Fear The Things That Go Beep In the Night? Some Initial Thoughts on the Intersection of Antitrust Law and Algorithmic Pricing
Stucke, M. E., &Grunes, A. P. (2016). Big data and competition policy. Oxford UniversityPress.