基于ChatGPT构建多维度AI审计框架  

Constructing a Multi-dimensional AI audit framework based on ChatGPT

作　　者：程皖川 CHENG Wanchuan(School of Business,Chengdu University,Longquanyi District,Chengdu Sichuan 610106,China)

机构地区：[1]成都大学商学院,四川成都610106

出　　处：《四川文理学院学报》2025年第2期147-153,共7页Sichuan University of Arts and Science Journal

摘　　要：围绕ChatGPT的基本情况及其在多维度AI审计框架中的应用,阐述了ChatGPT基于深度学习和自然语言处理技术,通过Transformer架构实现高效预训练,并引入“人类反馈强化学习”机制和指令微调技术,在多个领域展现卓越性能.分析了其高度拟人化、智能化升级和通用性强化三大突出特性.探讨了基于ChatGPT构建多维度AI审计框架的方法和意义,包括配置层对因子类和非因子类参数的调整以打造符合审计需求的智能模型,以及运行层的传达、实施和矫正三个子层的功能和实现方式.总结了该框架在提升审计自动化水平、实现审计联动和解决信息孤岛等方面的应用价值,同时指出AI审计在数据安全和道德规范方面面临的挑战,强调未来技术进步和规范完善的重要性.This paper explores the fundamental principles of ChatGPT and its application in building a multi-dimen-sional AI audit framework.It how ChatGPT leverages deep learning and natural language processing technologies to achieve efficient pre-training through the Transformer architecture.By integratingthe“Human Feedback Reinforcement Iearning”mechanism and instruction fine-tuning techniques,ChatGPT demonstrates exceptional performance across di-verse domains.Three distinctive characteristics are analyzed:its highly anthropomorphic interaction capabilities,intelligent performance upgrades,and strengthened generalizability.The paper focuses on methodologies and implications for estab-lishing a ChatGPT-powered multi-dimensional AI audit framework which includes:(1)Configuration Layer:Adjusting fac-tor-based and non-factor-based parameters to develop intelligent models tailored for audit requirements;(2)Operational Layer:Designing functional implementations across three sub-layers-communication task interpretation,execution pro-cess automation,and rectification error correction.The framework demonstrates significant value in enhancing audit auto-mation,enabling cross-system audit synergy,and resolving information silos.Meanwhile,challenges of AI audit in regard-ing data security and ethical compliance in AI auditing are highlighted,emphasizing the necessity for technological ad-vancements and regulatory refinement.Future directions stress the importance of balancing innovation with standardized governance in AI audit ecosystems.

关 键 词：ChatGPT 多维度 AI审计框架 

分 类 号：F239.1[经济管理—会计学]