| 兰天,袁鸣,何彦青,徐红姣,张 蕊,许德山.AIGC 深化应用背景下主要国家和地区的 “幻觉”治理研究[J].全球科技经济瞭望,2025,(10):33~46 |
| AIGC 深化应用背景下主要国家和地区的 “幻觉”治理研究 |
| AIGC Practices and Hallucination Mitigation Governance in Major Countries and Regions |
| |
| DOI：10.3772/j.issn.1009-8623.2025.10.005 |
| 中文关键词: AIGC 实践；生成式人工智能；大模型幻觉；AI 治理 |
| 英文关键词: AIGC practices; generative AI; LLM hallucination; AI governance |
| 基金项目:中信所青年基金项目“面向领域知识动态更新的大语言模型时效性评测方法与知识优化策略研究”（QN2025-16）；中信所 面上基金项目“基于情报事实一致性增强的多文档综述生成方法研究”（MS2025-07）。 |
| 作者 | 单位 | | 兰天 | | | 袁鸣 | | | 何彦青 | | | 徐红姣 | | | 张 蕊 | | | 许德山 | |
|
| 摘要点击次数: 9 |
| 全文下载次数: 8 |
| 中文摘要: |
| 从实践案例、政策法规与治理方法角度，对人工智能技术发展较快的美国、欧盟、日本和中国 等世界主要国家和地区使用人工智能生成内容（Artificial Intelligence Generated Content，AIGC）文 本生成技术在关键领域的实践应用及相关政策进行深入分析，系统归纳了识别、溯源、引导、缓解、 过滤、预防、标记和水印等环节的 AIGC“幻觉”治理方法，认为治理趋势正在从事后纠错转向事前 预防，建议构建“技术为基 + 多元协同”的综合治理方案，并具体提出加大核心技术研发投入，构建适 配监管框架，完善可信度评估体系，加强主体技术素养，深化国际规则协同等启示。 |
| 英文摘要: |
| From the perspectives of practical cases, policies and regulations, and governance methodologies,
this paper deeply analyzes the practical applications and relevant policies of Artificial Intelligence Generated
Content (AIGC) text generation technology in key sectors across major countries and regions, including the United
States, the European Union, Japan, and China. It systematically summarizes AIGC hallucination governance
methods across stages such as identification, traceability, guidance, mitigation, filtering, prevention, labeling,
and watermarking. It suggests that the governance trend is shifting from ex post correction to ex ante prevention,
recommends constructing a comprehensive governance solution based on “technology as the foundation + multi-stakeholder collaboration”, and specifically proposes key implications including increasing investment in core
technology R&D, constructing adaptive regulatory frameworks, improving credibility assessment systems,
enhancing technical literacy among relevant subjects, and deepening international regulatory coordination. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
