Authors: Benton N. Taylor and Andrew B. Davies
As any field advances along the three parallel axes of theory, data, and synthesis, progress along any one of these axes can only outpace the others for a limited time before it must wait for progress along the other axes to catch up. We find ourselves at a time in history where major advances in our ability to gather empirical data and handle those data computationally have, in many fields, created an overabundance of information (data) but a scarcity of knowledge (synthesis). Artificial Intelligence (AI) holds immense promise for uncovering patterns in these data that allow us to make inference about the world around us that would be otherwise impossible. However, patterns occur in data for many reasons, some of which can lead to incorrect or misleading inference if the data were collected in a manner inappropriate for the question of interest. The field of experimental design aims to structure data collection in a way that maximizes the ability of the researcher to make appropriate inference from the data collected, given the question or objective of the study1. There are numerous texts devoted to designing effective experiments, many of which are written to provide pertinent advice in a specific field of interest. In this chapter, we cover fundamental concepts of experimental design and provide guidance for how to use and incorporate these concepts into analyses harnessing AI.