Aiding Humans in Financial Fraud Decision Making: Toward an XAI-Visualization Framework

TL;DR

The paper addresses the challenge of AML investigations requiring final human decisions and information overload by introducing an XAI-VIS framework that fuses heterogeneous data and employs weakly supervised AI to align decision criteria $C_{1},...,C_{N}$ with label functions $L_{1},...,L_{N}$. It structures decision support into four stages—intelligence, design, choice, and review—mediated by a VA system that integrates input data, structured verification, and simulations, while a generative/discriminative/transformer AI loop yields and refines labels $Y_{P}$, $Y_{1},...,Y_{S}$ and $L_{T}$. The contributions include a unified, human-centered AML decision framework, explicit coupling of AI explanations with human criteria, and an iterative multimodal AI loop enabling GT feedback to improve models, thereby reducing bias and labor in AML investigations. This approach has practical significance for transparent, accountable, and scalable AI-assisted AML decision making in regulated environments.

Abstract

AI prevails in financial fraud detection and decision making. Yet, due to concerns about biased automated decision making or profiling, regulations mandate that final decisions are made by humans. Financial fraud investigators face the challenge of manually synthesizing vast amounts of unstructured information, including AI alerts, transaction histories, social media insights, and governmental laws. Current Visual Analytics (VA) systems primarily support isolated aspects of this process, such as explaining binary AI alerts and visualizing transaction patterns, thus adding yet another layer of information to the overall complexity. In this work, we propose a framework where the VA system supports decision makers throughout all stages of financial fraud investigation, including data collection, information synthesis, and human criteria iteration. We illustrate how VA can claim a central role in AI-aided decision making, ensuring that human judgment remains in control while minimizing potential biases and labor-intensive tasks.