Abstract
The understanding of bias in AI is currently undergoing a revolution. Often assumed to be errors or flaws, biases are increasingly recognized as integral to AI systems and sometimes preferable to less biased alternatives. In this paper we review the reasons for this changed understanding and provide new guidance on three questions: First, how should we think about and measure biases in AI systems, consistent with the new understanding? Second, what kinds of bias in an AI system should we accept or even amplify, and why? And, third, what kinds should we attempt to minimize or eliminate, and why? In answer to the first question, we argue that biases are "violations of a symmetry standard" (following Kelly). Per this definition, many biases in AI systems are benign. This raises the question of how to identify biases that are problematic or undesirable when they occur. To address this question, we distinguish three main ways that asymmetries in AI systems can be problematic or undesirable-erroneous representation, unfair treatment, and violation of process ideals-and highlight places in the pipeline of AI development and application where bias of these types can occur.