I participated at an interesting panel at an antitrust conference in Brussels today. I was about big data and whether it will be a problem for competition policy. This is something that has been widely discussed but there is little resolution on the issue.
When it comes to the potential problems that arise from big data, the focus is usually on two things. First, data is an asset and so are naturally concerned if such assets become concentrated in terms of ownership. Second, that concentration may be at the level of the individual (e.g., data about their behaviour) and that can make it difficult for individuals to switch between suppliers. Both of these may give rise to anti-competitive effects. However, we are really ahead of the game in these concerns. We do not really know which data is significant. And we do not really know which data is replicable (i.e., where substitute services can easily be provided).
In reality, a better place to focus at the moment is on the technologies that will make the data valuable for it is not of value on its own. You need technologies to clean it up and make it useful. And the uses are well known. You can use it to make better products (e.g., suggested Gmail responses, Siri etc). You can use it to allow better value capture through price discrimination (with the usual trade-offs that entails). But what has concerned myself (in discussion with my colleagues Ajay Agrawal and Avi Goldfarb), is that these technologies may be used to reduce competition per se.
How might this occur? Stepping back, as we have already written, machines can use data in dynamic settings and generate better predictions. For instance, machines could observe pricing outcomes and the profits of their firm and use it to find optimal pricing. The machines can also use pricing data to predict other pricing — a key step towards playing a game.
AIs have already learned to play complex games against themselves and humans (most recently, Go). But what if they learned to play pricing games? Imagine some large online retailers deploying pricing AI that learn to play pricing games (repeated ones) and engage in tacit collusion but in environments where this would be computationally impossible for humans. In fact, to make matters worse, imagine that the environments are so complex that no one can observe how they are doing it and the AI itself — as is common these days — can provide no “explanation.” The AI has just been told to maximise long-run profits and there is nothing wrong with that. But the key is that prices may be coordinated even though there is no agreement or even intention from a human. This would make it very difficult to prosecute under current laws.
We aren’t the only ones to have wondered about this. In their recent book, Virtual Competition, Ariel Ezrachi and Maurice Stucke raise similar concerns. I see such developments as inevitable (maybe in just a few years). Others at the conference were sceptical that machine learning would work with such a general objective and would need more specificity so the human complicity would be more clearly present. I guess we will see.
What might we do about it if the current law isn’t up to it? One of the best ways to counter price collusion is to provide consumers with the means of encouraging competitive entry that undercuts prices. The problem is that if consumers aren’t away of lower prices, that entry effect is muted. In this regard, the best defense against a bad AI is to have a good AI. That is, an AI who operates on behalf of consumers. We have seen this recently with adversarial machine learning whereby AI can be used to uncover the behaviour and predictions of other AI. The problem is that to do this consumers have to have access to the same information as competitors. That is, if a competitor allows other firms to access key data (through an API), they should be required to have such access open to all. It is asymmetry of access we should fear. I wouldn’t be as concerned if a firm just kept pricing and other data to themselves.
Admittedly, this is all speculative stuff. But that is the nature of these things. I am not in favour of pre-empting regulatory and competition battles when technology is uncertain. However, we can still be concerned about asymmetries that seem to favour firms over consumers and the area of data may be one place where we can pre-empt somewhat.