Particapatory Harm Auditing of Generative-AI for Use in Cultural Heritage

As a society, we are in a period of technological development where AI is becoming a ubiquitous and sometimes even hidden tool within our daily activities. In the context of cultural heritage, for example, there is enormous potential for AI, especially generative AI, to streamline time consuming or resource intensive tasks, allowing professionals the time to engage with activities that require a human hand. However, there are still numerous debates surrounding the ethics and practicalities of implementing such tools. One of the key driving forces behind these debates is that the decisions that under-pin AI creation and usage are often solely made by a limited demographic, primarily consisting of AI experts, and, equally, these decisions are often opaque to the users. In this paper, we will explore the initial findings from the project which focus on auditing generative AI within cultural heritage, with a focus on metadata creation and enrichment. Through various workshops with cultural heritage professionals, we have taken the first steps to providing tools for use by subject matter experts to approach a generative AI system, at various stages of the development lifecycle, and effectively assess the degree to which it can be safely and effectively implemented within their workflows. Our co-design methodologies have resulted in a clearer understanding of user personas and journeys for the use of generative AI in a cultural heritage context, as well as clarifying ideas around the acceptability of generated responses for use within cultural heritage when set against factors such as potential for harm and quality of output. These are foundational tools and concepts for delivering a full workbench of auditing tools, as well as for enabling a more nuanced understanding of the AI needs of the cultural heritage sector.