Mapping artificial intelligence bias: a network-based framework for analysis and mitigation

In the evolving era of generative artificial intelligence (AI), bias in AI systems remains a significant concern, leading to unfair outcomes across applications. Addressing this issue requires understanding how biases are introduced and propagated through AI models. In this paper, we present a network-based framework to map, analyze, and mitigate biases in AI systems. Our approach includes a systematic review of literature to identify key sources of bias, categorized into three areas: Data, Design, and human-AI teaming. using a network-based methodology, we illustrate the interconnected nature of bias propagation in AI systems. This framework highlights critical points where bias can be introduced and offers strategies for mitigating them. We demonstrate its effectiveness through use cases in healthcare and recommender systems, underscoring the importance of integrating ethical considerations throughout the AI development lifecycle and the necessity of continuous bias monitoring. The framework provides actionable insights for developers, practitioners, and decision-makers, guiding them toward more equitable AI systems. Our paper advocates for a holistic, collaborative approach to mitigating bias, contributing to inclusive technological ecosystems.