英国第二大图书馆:牛津大学博德利图书馆藏书被曝用于 OpenAI 模型训练UK's second largest library: Oxford University's Bodleian Library's collection exposed as being used for OpenAI model training.


AI free token/AI免费积分
英国第二大图书馆:牛津大学博德利图书馆藏书被曝用于 OpenAI 模型训练

IT之家 9 月 26 日消息,据英媒卫报消息,正值各大科技公司在学术机构中四处搜寻全新语料之际,牛津大学已允许 ChatGPT 开发商 OpenAI 利用其博德利图书馆(Bodleian Library)的历史典籍来训练其人工智能模型。

IT之家注:博德利图书馆是英国牛津大学主要的研究图书馆,是英国规模仅次于大英图书馆的第二大图书馆。

一份内部文件表明,由 OpenAI 完成数字化的博德利藏书已被用来“构建 OpenAI 的训练集”。牛津大学于 2025 年 3 月宣布与该公司达成合作,使用 OpenAI 的软件对该校这座享誉世界的图书馆藏书开展数字化处理。

校方当时表示,此举将让广大学生和研究人员更便捷地查阅这些文献资料。不过,当时的合作公告并未提及这些资料会被用于训练 OpenAI 的模型。

OpenAI 的一位发言人表示,公司很荣幸可以促成“当今的 AI 模型为后世守护人类的历史文化瑰宝”。

该发言人还补充道:“当下全球已有超过十亿人在日常生活中使用这项工艺,因而让它充分反映不同的文化、历史和视角至关关键。”

一份依据《信息自由法》调取的牛津大学会议纪要表明,涵盖博德利管理委员会成员在内的教职员工均表达了顾虑。他们不仅担忧与 OpenAI 合作可能会引发声誉风险,还指出这项合作涉及高能耗工艺,可能会对大学自身作出的环保承诺产生负面影响。

在一个名为“NextGenAI”的项目下,OpenAI 已与涵盖波士顿公共图书馆、加州理工学院、麻省理工学院以及密歇根大学在内的多家美国研究型图书馆签署了类似协议。牛津大学则是该项目中唯一的英国成员。

截至 2025 年 6 月,博德利图书馆已向 OpenAI 给出了 12.5 万份历史学位论文的扫描影印页,其中涵盖了 19 至 20 世纪欧美高校的博士论文。其他已扫描的文献还涵盖一份罕见的 16 世纪“宽边民谣 / 街头民谣(Broadside Ballads)”珍藏集,共计 1 万首,记录了曾流传于都铎王朝街头巷尾的歌词与乐谱。

另外,教职人员还商讨了后续的数字化计划,涉及 18 世纪爱尔兰国家档案、爱尔兰小说家玛利亚 · 埃奇沃思(Maria Edgeworth)的私人信函,以及多萝西 · 霍奇金(Dorothy Hodgkin)关于青霉素研究的实验笔记。

校方否认了向公众和学生隐瞒机器学习相关用途的指责。发言人表示,数字化确实是学校的核心诉求,但教职员工始终明确承认,该项目同时会为模型给出训练数据。

牛津大学发言人表示:“借助与 OpenAI 合作完成数字化的文献资料规模有限,均已超出版权保护期;另外,OpenAI 对这些资料的使用权是非排他性的。”

博德利图书馆同样保留了自主发布这些数字化资源的权利。在将来几个月内,图书馆将逐步在网络上公开给出这批资料,正如笔者对待其他数字化合作成果一样。

事实上,该项目将协助博德利图书馆惠及更广泛的公众群体,让那些原本难以接触到这些文献的人可以从容查阅。

本文来源:https://www.ithome.com/1/007/426.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

IT Home, September 26 — According to the British media outlet The Guardian, as major technology companies search academic institutions for new corpora, the University of Oxford has allowed ChatGPT developer OpenAI to use historical texts from its Bodleian Library to train OpenAI's artificial intelligence models.

IT Home note: The Bodleian Library is the main research library of the University of Oxford in the United Kingdom and is the second largest library in the United Kingdom, behind only the British Library.

An internal document indicates that the Bodleian collection digitized by OpenAI has been used to "build OpenAI's training set." The University of Oxford announced in March 2025 that it had reached a partnership with the company, using OpenAI's software to digitize the holdings of its world-renowned library.

At the time, the university said that the move would make it easier for a wide range of students and researchers to access these materials. However, the announcement at the time did not mention that the materials would be used to train OpenAI's models.

An OpenAI spokesperson said the company was honored to help enable "today's AI models to preserve humanity's historical and cultural treasures for future generations."

The spokesperson added: "More than one billion people around the world now use this technology in their daily lives, so it is essential that it fully reflects different cultures, histories, and perspectives."

Oxford University meeting minutes obtained under the Freedom of Information Act showed that staff, including members of the Bodleian Council, expressed concerns. They worried not only that the partnership with OpenAI could create reputational risks, but also pointed out that the partnership involves energy-intensive processes and could have a negative impact on the university's own environmental commitments.

Under a project named "NextGenAI," OpenAI has signed similar agreements with a number of U.S. research libraries, including the Boston Public Library, the California Institute of Technology, the Massachusetts Institute of Technology, and the University of Michigan. The University of Oxford is the project's only UK member.

As of June 2025, the Bodleian Library had provided OpenAI with scanned facsimile pages of 125,000 historical dissertations, including doctoral theses from European and American universities from the 19th to 20th centuries. Other scanned materials include a rare 16th-century "Broadside Ballads / street ballads" collection totaling 10,000 pieces, which record lyrics and musical scores once circulated on the streets of the Tudor era.

In addition, staff discussed further digitization plans involving 18th-century Irish national archives, the private letters of Irish novelist Maria Edgeworth, and the laboratory notes of Dorothy Hodgkin concerning penicillin research.

The university denied accusations that it had hidden machine-learning-related uses from the public and students. A spokesperson said that digitization was indeed the university's core objective, but staff had consistently and explicitly acknowledged that the project would also provide training data for models.

An Oxford University spokesperson said: "The volume of materials digitized through the partnership with OpenAI is limited, and all of it is already beyond its copyright protection period; in addition, OpenAI's right to use these materials is non-exclusive."

The Bodleian Library also retains the right to independently publish these digitized resources. Over the coming months, the library will gradually make this batch of materials available online, just as it has done with other digitization partnership outputs.

In fact, the project will help the Bodleian Library benefit a broader public, enabling people who would otherwise have had difficulty accessing these materials to consult them with ease.

UK's second largest library: Oxford University's Bodleian Library's collection exposed as being used for OpenAI model training.

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry