不一致的“开放”:人工智能开源话语的社会化生产与可信知识的“审计剧场”Inconsistent “Openness”: The Social Production of AI Open-Source Discourse and the “Audit Theater” of Trustworthy Knowledge
方师师,王易鑫
摘要(Abstract)：
开源与闭源是软件技术迭代的核心路径，但在人工智能大语言模型领域呈现出定义、话语与实践的“不一致”问题。基于科学知识社会学(SSK)的技术-社会互构视角，本文将人工智能开源“知识”界定为技术事实的社会建构产物，综合运用模型审计、内容分析与社会实验方法，依托“技术事实建构-话语权力争夺-公众参与协商”三维分析框架展开实证研究。研究发现，在技术层面，存在全栈开源与形式化开源的漏洞风险认知分野；在话语层面，科技企业、技术社群、政府部门与媒体机构形成差异化博弈格局，其中科技企业与技术社群争夺因果解释权，政府部门与媒体机构承担中介平衡角色；在实践层面，公众因知识隔阂难以有效参与治理，但“审计剧场”可有效提升其风险认知水平与参与意愿。本文由此提出，对人工智能开源知识可构建从技术合规转向社会协商的可信知识体系，以此实现技术信任的社会共建。
关键词(KeyWords)： 大语言模型;开源话语;科学知识社会学;审计剧场;可信知识
基金项目(Foundation):
作者(Author): 方师师,王易鑫
参考文献(References)：
- 大卫·布鲁尔(2001):《知识和社会意象》(艾彦译),北京：东方出版社。(原书Bloor,D.[1976].Knowledge and social imagery.Chicago:University of Chicago Press.) 戴宇辰(2021):传播研究的“物质性”取径：对若干核心议题的澄清，《福建师范大学学报(哲学社会科学版)》,第5期，142-152,171页。 傅宏宇、贾开、彭靖芷(2026):人工智能开源的价值、风险与生态治理研究，《电子政务》,第2期，58-69页。 国家统计局(2025年1月17日):王萍萍：人口总量降幅收窄人口素质持续提升，获取自https://www.stats.gov.cn/xxgk/jd/sjjd2020/202501/t20250117_1958337.html 李智超，李鹿嘉(2024):人工智能公众风险感知的类型与演化——基于全球视角的分析，《电子科技大学学报(社科版)》,第26卷第6期，21-33页。 谭小荷(2023):AI想象的展褶：技术乐观主义与无能主体的共生——以普通人对ChatGPT的社会技术想象为中心，《新闻界》,第11期，52-65,96页。 Abbas,F.,Chesterman,S.& Taeihagh,A.(2025).Building trust in the generative AI era:A systematic review of global regulatory frameworks to combat the risks of mis-,dis-,and mal-information.AI & Society,1-38.doi:10.1007/s00146-025-02698-9. AI World.(2025).Open-Source AI:Year in Review 2025.Hugging Face Spaces.Retrieved From https://huggingface.co/spaces/aiworld-eu/Open-Source-AI-Year-in-Review-2025 Aydarova,E.(2022).Neoliberal fictions for the audit theater:University educators and administrators navigating accountability regimes.Anthropology and Humanism,47(1),133-146.doi:10.1111/anhu.12345. Barnes,B.& Bloor,D.(1982).Relativism,rationalism,and the sociology of knowledge.In Hollis,M.& Lukes,S.(Eds.),Rationality and Relativism (pp.21-47).Oxford:Basil Blackwell. Batlle-Roca,R.,Ibáňez-Martínez,L.,Serra,X.,Gómez,E.& Rocamora,M.(2025).MusGO:A community-driven framework for assessing openness in music-generative AI.In Proceedings of the 26th International Society for Music Information Retrieval Conference (pp.727-736).Daejeon:ISMIR.doi:10.5281/zenodo.17811474. Bengio,Y.,Mindermann,S.,Privitera,D.,Besiroglu,T.,Bommasani,R.,Casper,S.,… & Zhang,Y.Q.(2024).International scientific report on the safety of advanced ai (interim report),1-91.arXiv preprint arXiv:2412.05282. Brynjolfsson,E.,Li,D.& Raymond,L.(2025).Generative AI at work.The Quarterly Journal of Economics,140(2),889-942.doi:10.1093/qje/qjae044. Don-Yehiya,S.,Burtenshaw,B.,Fernandez Astudillo,R.,Osborne,C.,Jaiswal,M.,Kuo,T.S.,Zhao,W.T.,Shenfeld,I.,Peng,A.D.,Yurochkin,M.,Kasirzadeh,A.,Huang,Y.S.B.,Hashimoto,T.,Jernite,Y.,Vila-Suero,D.,Abend,O.,Ding,J.,Hooker,S.,Kirk,H.R.& Choshen,L.(2025).The future of open human feedback.Nature Machine Intelligence,7,825-835.doi:10.1038/s42256-025-01038-2. Eiras,F.,Petrov,A.,Vidgen,B.,de Witt,C.S.,Pizzati,F.,Elkins,K.,Mukhopadhyay,S.,Bibi,A.,Csaba,B.,Steibel,F.,Barez,F.,Smith,G.,Guadagni,G.,Chun,J.,Cabot,J.,Imperial,J.M.,Nolazco-Flores,J.A.,Landay,L.,Jackson,M.T.,Rottger,P.,Torr,P.,Darrell,T.,Lee Y.S.& Foerster,J.N.(2024).Position:Near to mid-term risks and opportunities of open-source generative AI.In Proceedings of the 41st International Conference on Machine Learning (pp.12348-12370).Vienna:PMLR. Executive Office of the President.(November 1,2023).Executive Order 14110:Safe,Secure,and Trustworthy Development and Use of Artificial Intelligence.Federal Register,88(210),75191-75226.Retrieved from https://www.Federal Register.gov/documents/2023/11/01/2023-24283/safe-secure-and-trust worthy-development-and-use-of-artificial-intelligence Ferrer,X.,van Nuenen,T.,Such,J.M.,Coté,M.& Criado,N.(2021).Bias and discrimination in AI:A cross-disciplinary perspective.IEEE Technology and Society Magazine,40(2),72-80.doi:10.1109/MTS.2021.3056293. Gunning,D.& Aha,D.W.(2019).DARPA's explainable artificial intelligence (XAI) program.AI Magazine,40(2),44-58.doi:10.1609/aimag.v40i2.2850 Hall,P.,Mundahl,O.& Park,S.(2025).The pitfalls of “security by obscurity” and what they mean for transparent AI.In Proceedings of the 39th AAAI Conference on Artificial Intelligence (pp.28042-28051).Philadelphia:AAAI Press.doi:10.1609/aaai.v39i27.35022. Ho,A.,Besiroglu,T.,Erdil,E.,Owen,D.,Rahman,R.,Guo,Z.C.,Atkinson,D.,Thompson,N.& Sevilla,J.(2024).Algorithmic progress in language models.In Proceedings of the 38th International Conference on Neural Information Processing Systems (pp.1856).Vancouver:Curran Associates Inc. Hoffmann,M.,Nagle,F.& Zhou,Y.U.(2024).The value of open source software.Boston:Harvard Business School.doi:10.2139/ssrn.4693148. Jeon,K.W.,Han,H.J.& Lee,K.W.(2025).Evaluating the openness of impactful AI models with a focus on LLMs.The Transactions of the Korea Information Processing Society,14(6),468-479.doi:10.3745/TKIPS.2025.14.6.468. Kathikar,A.,Nair,A.,Lazarine,B.,Sachdeva,A.& Samtani,S.(2023).Assessing the vulnerabilities of the open-source artificial intelligence (AI) landscape:A large-scale analysis of the hugging face platform.In Proceedings of 2023 IEEE International Conference on Intelligence and Security Informatics (pp.1-6).Charlotte:IEEE.doi:10.1109/ISI58743.2023.10297271. Liesenfeld,A.& Dingemanse,M.(2024).Rethinking open source generative AI:Open-washing and the EU AI act.In Proceedings of 2024 ACM Conference on Fairness,Accountability,and Transparency (pp.1774-1787).Rio de Janeiro:Association for Computing Machinery.doi:10.1145/3630106.3659005. Loizos,C.(August 28,2025).Anthropic users face a new choice-opt out or share your chats for AI training.TechCrunch.Retrieved from https://techcrunch.com/2025/08/28/anthropic-users-face-a-new-choice-opt-out-or-share-your-data-for-ai-training/ Maffulli,S.(October 18,2024).Defining open source AI.Open source initiative.Retrieved from https://opensource.org/wp-content/uploads/2024/10/osi_town hall_21.pdf Mehrotra,A.,Zampetakis,M.,Kassianik,P.,Nelson,B.,Anderson,H.,Singer,Y.& Karbasi,A.(2024).Tree of attacks:Jailbreaking black-box LLMs automatically.In Proceedings of the 38th International Conference on Neural Information Processing Systems (pp.1952).Vancouver:Curran Associates Inc. Mitchell,M.,Attanasio,G.,Baldini,I.,Clinciu,M.,Clive,J.,Delobelle,P.,Dey,M.,Hamilton,S.,Dill,T.,Doughman,J.,Dutt,R.,Ghosh,A.,Forde,J.Z.,Holtermann,C.,Kaffee,L.A.,Laud,T.,Lauscher,A.,Lopez-Davila,R.L.,Masoud,M.,Nangia,N.,Ovalle,A.,Pistilli,G.,Radev,D.,Savoldi,B.,Raheja,V.,Qin,J.,Ploeger,E.,Subramonian,A.,Dhole,K.,Sun,K.,Djanibekov,A.,Mansurov,J.,Yin,K.,Cueva,E.V.,Mukherjee,S.,Huang,J.,Shen,X.D.,Gala,J.,Al-Ali,H.,Djanibekov,T.,Mukhituly,N.,Nie,S.R.,Sharma,S.,Stanczak,K.,Szczechla,E.,Torrent,T.T.,Tunuguntla,D.,Viridiano,M.,Van Der Wal,O.,Yakefu,A.,Névéol,A.,Zhang,M.K.,Zink,S.& Talat,Z.(2025,April).SHADES:Towards a multilingual assessment of stereotypes in large language models.In Proceedings of 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics:Human Language Technologies (Volume 1:Long Papers) (pp.11995-12041).Albuquerque:Association for Computational Linguistics.doi:10.18653/v1/2025.naacl-long.600. Mozilla.(2023,October 31).Joint statement on AI safety and openness.Retrieved from https://open.mozilla.org/ai-safety/ Open Source Initiative.(March 22,2007).The open source definition.Retrieved from https://opensource.org/osd Open Source Initiative.(2024).Open source AI definition (OSAID) version 1.0.Retrieved From https://opensource.org/ai/open-source-ai-definition Organisation for Economic Co-Operation and Development.(August 14,2025).AI openness:A primer for policymakers.Retrieved from https://www.oecd.org/en/publications/ai-openness_02f73362-en.html OWASP.(n.d.).OWASP Risk Assessment Framework.Retrieved from https://owasp.org/www-project-risk-assessment-framework/ Payne,K.(2024).The geopolitics of AI.The RUSI Journal,169(5),54-55.doi:10.1080/03071847.2024.2413274. Raymond,E.S.(1999).The cathedral and the bazaar:Musings on Linux and open source by an accidental revolutionary.Sebastopol:O'Reilly & Associates,Inc. Rivera,A.,Abhari,K.& Xiao,B.(2025).Responsible AI design:The authenticity,control,transparency theory.Journal of the Association for Information Systems,26(5),1337-1389.doi:10.17705/1jais.00948. Roberts,H.,Hine,E.,Taddeo,M.& Floridi,L.(2024).Global AI governance:Barriers and pathways forward.International Affairs,100(3),1275-1286.doi:10.1093/ia/iiae073. Sanderson,K.(2023).GPT-4 is here:What scientists think.Nature,615(7954),773.doi:10.1038/d41586-023-00816-5. Sapkota,R.,Raza,S.& Karkee,M.(2025).Comprehensive analysis of transparency and accessibility of ChatGPT,DeepSeek,and other SoTA large language models,1-23.arXiv preprint arXiv:2502.18505. Solaiman,I.(2023).The gradient of generative AI release:Methods and considerations.In Proceedings of 2023 ACM Conference on Fairness,Accountability,and Transparency (pp.111-122).Chicago:Association for Computing Machinery.doi:10.1145/3593013.3593981. Vaughan-Nichols,S.(August 5,2025).OpenAI returns to its open-source roots with new open-weight AI models,and it's a big deal.ZDNET.Retrieved from https://www.zdnet.com/article/openai-returns-to-its-open-source-roots-with-new-open-weight-ai-models-and-its-a-big-deal/ White,M.,Haddad,I.,Osborne,C.,Liu,X.Y.Y.,Abdelmonsef,A.,Varghese,S.& Le Hors,A.(2024).The model openness framework:Promoting completeness and openness for reproducibility,transparency,and usability in artificial intelligence,1-28.arXiv preprint arXiv:2403.13784.
- ① 本文涉及三个相关的概念，开源人工智能、 开源大模型与开源大语言模型。其中开源人工智能(Open-Source Artificial Intelligence)所指范围最广，是上位概念。主要指源代码、模型结构、训练框架、数据集、推理工具等任何与人工智能相关的组件向公众开放使用、修改、分发的人工智能体系。开源大模型(Open-Source Large Models)范围中等，是开源人工智能的重要子类，特指参数规模巨大、预训练而成的通用人工智能模型，且其权重/代码开源。不限于语言，可包括文本、图像、语音、视频等多模态。开源大语言模型(Open-Source Large Language Models)范围最窄，是开源大模型更具体的子类。专指面向自然语言处理、基于Transformer 架构的开源大模型。典型的如Llama 2、阿里云Qwen等。因此在本文中，如果是指“人工智能领域的开源整体生态”时，我们采用“开源人工智能“的表述；如果是在实证环节，鉴于本文的研究，我们采用”开源大语言模型“来指代具体的研究对象。