Generative AI is built on large language models (LLMs) which in turn are trained on large and diverse pools of data. These LLMS are consequential; many analysts believe that these technologies could transform society, undermine trust and longstanding democratic norms, and lead to major changes in how we conduct and replicate research. For example, researchers find growing evidence that content creators are less willing to share their data which could, over time, undermine openness and approaches to analyzing and replicating data that underpin science and knowledge generation.

Policymakers must find ways to encourage the development of LLMs while protecting individuals, groups, and local firms from potential harm. One such tool is data governance because there is no AI without large pools of data. But there are many types of data, strategies to govern data, and gaps in the governance of data. Moreover, no one yet knows how to ensure that the data that underpins AI, data built on troves of knowledge accumulated over centuries, is as accurate, complete, and representative as possible.

Audience & Format

The conference will consist of four panels designed to examine the current strategy for obtaining data, what we can learn about data governance through open- and closed-source models, and the long-term implications of the reliance on LLMs for society as a whole, and to discuss the gaps in data governance and how we can involve more people in the discussion over data governance. Each panel will begin with an overview presented by experts from business, civil society, and government. A moderator will then engage with the audience and the panel as they explore key questions related to data and LLMs.