In Short:
Mumsnet, a popular UK parenting forum, discovered AI companies were using its vast data without permission. After initial talks with OpenAI about potential licensing, the partnership fell through when OpenAI deemed Mumsnet’s 6 billion-word dataset too small and publicly accessible. Mumsnet plans to take legal action, frustrated as the content is primarily female-driven and unique.
Mumsnet, a well-known UK-based parenting forum, has become a digital repository of discussions related to various aspects of raising children, with an impressive archive exceeding six billion words contributed by its active user base. Topics range from the mundane, such as dirty diapers, to more controversial issues, including gender dynamics in parenting. The platform has solicited extensive user engagement, reflecting the diverse experiences and concerns of modern motherhood.
Data Concerns and Licensing Deals
This spring, Mumsnet became aware that artificial intelligence companies were scraping its vast trove of data without consent. Consequently, the company sought to negotiate licensing agreements with major players in the AI space, notably OpenAI. Initial discussions with OpenAI indicated a willingness to explore collaboration, leading to anticipation from Mumsnet’s leadership. However, these talks ultimately fell through.
Failed Negotiations
In July, Mumsnet publicly announced its decision to pursue legal action.
During preliminary discussions, a strategic partnership lead from OpenAI expressed interest in datasets containing over one billion words, which excited Mumsnet’s leadership. As Justine Roberts, founder and CEO of Mumsnet, recounted, multiple exchanges occurred, requiring non-disclosure agreements and detailed information exchanges. However, a month later, OpenAI communicated that it was no longer interested in a partnership, deeming Mumsnet’s dataset insufficiently large to merit a licensing arrangement.
OpenAI’s Position
OpenAI’s rationale was disclosed in communications reviewed by WIRED, indicating that they favor datasets capturing a broad spectrum of human experience rather than those already available to the public. In a statement, OpenAI spokesperson Kayla Wood clarified that the organization pursues partnerships focused on large-scale datasets reflecting the diversity of human society.
Roberts expressed her frustration with the outcome, noting that the uniqueness of Mumsnet lies in its predominantly female-generated content, which comprises approximately 90 percent of the discussions. This high-quality conversational data is viewed as an underutilized resource in AI training.
OpenAI’s Strategic Partnerships
Despite its declined partnership with Mumsnet, OpenAI has successfully established various data-licensing agreements over the past year with media organizations and platforms like Vox Media, the Atlantic, Time, and Condé Nast, as well as platforms with user-generated content like Reddit. The specifics of these agreements have not been disclosed, leaving the scale of the datasets involved uncertain.
When asked about their criteria for commercial licensing, OpenAI refrained from providing specific details but underscored that collaborations with publishers are geared towards enhancing the visibility of their content within OpenAI’s offerings.
