Participatory Harm Auditing of Generative AI in Cataloguing and Description

The great variety and volume of data and the various methods for analysing data are an increasingly prominent and resource intensive element to collections management. Consequently, there is an understandable drive to explore the numerous ways in which generative-AI might streamline these activities, allowing professionals the time they need to engage with tasks that necessitate a human hand. There have been a wide array of examples where institutions have employed generative-AI within limited contexts, but to great effect. Widespread application, however, is hindered by ongoing debates about the ethics and practicalities of implementing generative-AI models/tools. Particularly, the opaqueness of how generative-AI models/tools are created, trained and produce content makes it challenging for information professionals to trust the integrity of any arising outputs . In the proposed workshop participants will have the opportunity to understand and apply audit methodologies developed within our research project and provide views towards further co-design. The methodology has been created to support the archives and records community to critically and consistently assess the quality of generative AI outputs for their viability of use in diverse contexts (e.g. cataloguing and description). The provision of such a methodology is a significant step towards where things could be in relation to responsible technological integrations and innovations with cultural heritage data. NB: This workshop is intended to both provide training for participants and to gather data in an anonymous and aggregated form to help support improvement of the presented methodologies. As such, participants will be required to sign consent forms in order to participate.