Market Power

Market PowerNew technologies often have implications that are pro-competitive: for instance, when the internet created online shopping opportunities to bring new degrees of competition to retail markets.Similarly, in another century, railways opened up greater competition and trade between regions.However, particularly in digital and communication technologies, markets have evolved to be dominated by a few larger firms with considerable market power.Anticipating that with respect to artificial intelligence has brought about a regulatory interest in the potential anti-competitive effects of this new technology.This chapter's purpose is to examine models that may shed some light on the nature of market power (if any) that might emerge in the development and from the adoption of AI. 1 As has been the theme in this book, the focus is on AI prediction.In this regard, concerns about market power often relate to the inputs into those predictions-specifically, training and input data.While many models of information, data, and AI in competition tend to consider all three of these as somewhat equivalent, a closer reading suggests that some are more applicable to one characterization than another.Hence, the organization of this chapter will divide models into ones that talk about market power in training data, input data, and predictions.This does define a bright line between those things that may obscure some commonalities.However, the goal here is to provide an entry point for those wishing to examine competition issues and policy outcomes that are accompanying the emergence and growth of AI technologies. Market Power in Training DataAI prediction algorithms generate predictions by being fed input data.The prediction algorithms themselves are created using training data.The training data and input data are often distinct, with the training data being generated in the past, whereas the input data is often to be generated in the future.Sometimes,