ABSTRACT
Artificial intelligence (AI) systems are not, and cannot be, value-free. They reflect human values. Given this, and given the potential of AI to impact and disrupt society in significant ways, it is important that the values that influence AI systems are critically scrutinized. This chapter has two primary aims. The first is to examine some important respects in which AI systems are value-laden. More specifically, the chapter examines the influence of values at four stages in the AI lifecycle: (1) problem framing and operationalization; (2) data; (3) modeling and validation; and (4) deployment. This examination paves the way for a second task, which is to address the question of how AI should be governed—including whether there should be moral or political “guardrails” on AI. There is much disagreement surrounding this question, and this chapter will not attempt to answer it definitively. Rather, it will discuss three recent approaches to AI governance, which shed light on ways that policymakers are attempting to address practical questions about the management of values in AI. This discussion will illustrate how controversial and important these questions are.
Readers may be interested in these Handbook chapters as well: Bennett Holman and T. Y. Branch, “Reflecting on Responses to the New Demarcation Problem”; Stephen John, “Transparency in Science”; Alexander Tolbert, “Algorithmic Abolitionism and The Racial Algorithm.”
