AI and Discrimination: Sources of Algorithmic Biases

Abstract

In this editorial, we define discrimination in the context of AI algorithms by focusing on understanding the biases arising throughout the lifecycle of building algorithms: input data for training, the process of algorithm development, and algorithm execution and usage. We draw insights from a few empirical studies to illustrate biases codified in algorithms that could result in harmful outcomes. We call on information systems scholars to prioritize scholarship in the area of algorithmic discrimination that can help generate new knowledge systems that would help safeguard against widespread and unaccountable harm.