Technology-driven causal inference prospects and challenges

The relationship between technology and causal inference is multifaceted. It can be broadly divided into two aspects: the role of causal inference in technological practice, such as engineering reasoning, and how technology enables or enhances our ability to infer causal relations, for example, through machine-driven inference. We focus on the latter aspect. Technology serves as a base medium for most causal inferences, ranging from storing data on a hard drive to drawing graphs on a piece of paper. In some cases, these technological media are not central to causal inference, since one can draw the same causal inferences using different technological media. However, there are certain instrumental technological applications that are fundamental for enabling or enhancing certain causal inferences. These applications will be the focus of our discussion. This chapter aims to assess successes, challenges, and opportunities for machine-driven causal inference. We will begin with a brief introduction to the topic, after which we will discuss the limitations of current approaches, with a focus on the thorny issue of transferability (i.e., extrapolating or generalizing). We then propose a direction to move forward in handling this challenge that combines insights on counterfactual causal reasoning and explanation with the idea that the content and relevance of (counterfactual) explanation-seeking questions relate to the roles (to be) performed by stakeholders in a machine learning ecosystem. In this chapter, we will be concerned with methodological and epistemological issues concerning technology-driven causal inference rather than metaphysical ones.