Reconsidering AI-driven Accounting Information Systems

Abstract

This study theoretically reexamines the role of AI-driven accounting information systems (AIS) within ERP environments. Traditionally, AIS has been regarded as suitable for automation due to its numerical processing and rule-based characteristics, leading to expectations that AI could automate accounting judgments. However, accounting judgment is inherently contextual and discretionary, as it is influenced by institutional constraints, managerial decision-making, and relationships with stakeholders. Based on this characteristic, this study reconceptualizes AI-driven AIS not as mechanisms that automate accounting judgment but as information infrastructures that structure judgment conditions and support decision-making. In particular, the study focuses on foundation models (FM) and multimodal foundation models (MFM) and examines their potential to analyze both structured ERP data and unstructured data such as documents and images in an integrated manner. The findings suggest that AI does not replace accounting judgment but can function as a complementary analytical layer that visualizes the decision environment through comparative analysis and deviation detection. At the same time, AI-driven AIS introduces new challenges related to explainability, training data governance, and system governance. This study contributes theoretically by repositioning AI-driven AIS as a governance-oriented decision-support infrastructure.