Procedural and Structural Justice Through Causal Understanding, Component Decoupling, and Relation Characterization

The proposed research aims to better understand, precisely characterize, and effectively mitigate bias and discrimination by developing and leveraging appropriate ideas and methods in causal learning and reasoning. This project focuses on achieving procedural and structural justice in criminal and juvenile justice systems, and broader aspects of our daily lives as well. Recent developments in the algorithmic fairness literature have proposed various technical notions and approaches to empirically quantify violations of justice and fairness. However, the important issue of the disguised procedural violation of justice is often overlooked. The proposed research regards causal learning and reasoning as essential and necessary for principled and effective analyses towards procedural and structural justice. To this end, this proposal presents research plans to investigate three different but highly related tasks, including (1) a fairness flowchart as a high-level roadmap, encompassing implicit assumptions and expected outcomes of different types of fairness inquiries, clarifying their distinctive emphases of justice semantics, (2) a series of technical approaches focusing on the procedural fairness and structural justice, in both static settings and in the long run, and for practical scenarios involving tabular data or language processing, and (3) a causality-guided framework to debias outputs of large language models (LLMs), and furthermore, to evaluate bias metrics themselves in the language processing context. The goal of proposed project is to reveal and clearly present the challenges and opportunities in this important, promising, and potentially fruitful research direction. The proposed research aims to provide scientific solution and perspective to the impartial administration of criminal and juvenile justice that necessitates transparent, fair, and just structural foundations, and also to broader aspects of fairness and justice considerations. CA/NCF