From Bias to Fairness: Challenges in Artificial Intelligence Systems through a Social Systems Lens

Abstract

As artificial intelligence (AI) systems increasingly permeate diverse societal realms, concerns regarding inherent biases within these systems have gained prominence. Such biases can perpetuate and worsen existing issues within social structures, leading to unjust outcomes and undesirable, unethical practices. This study delves into the contexts and classifications of AI biases. It begins by providing an overview of typical forms of biases observed in AI systems, encompassing human, systemic, and computational biases. Furthermore, it outlines methods for mitigating biases in AI systems, including techniques for algorithmic fairness, conscientious data collection practices that prioritize diversity, and involving stakeholders in the design process. Additionally, the paper explores the distinction between fundamental challenges – accountability and responsibility, followed by adaptability and flexibility. Drawing on insights from social systems theory, it investigates the emergence and propagation of biases within what is termed the "genetic form" of the social system as represented in advocacy organizations. This analysis underscores the interrelation of legal professionals, legal actions, and judicial procedures from an ontological standpoint. The inquiry posed is whether AI can serve as an advocate. Lastly, a systemic sociological approach is proposed for constructing a framework of meaning, juxtaposed with AI methodologies.