A report co-written with Kylie Message on a panel discussion between University Researchers and GLAM research and discipline experts at the NMA in July 2023. See also online here: https://hrc.cass.anu.edu.au/news/event-report-ai-and-beyond-exploring-shared-work-stem-and-glam-research-institutions
Cultural Conversations: Ideas, Practice, Networking session 5
AI and beyond: Exploring the shared work STEM and GLAM research & institutions
ChatGPT says ‘AI can revolutionize the museum experience by offering innovative ways to engage visitors, streamline operations, and preserve cultural heritage’ but is also notes potential drawbacks: museums ‘must strike a balance between leveraging AI technologies and preserving the unique qualities that make museums special, such as human interaction, cultural context, and inclusivity’. When it comes to Galleries, Libraries, Archives and Museums (GLAM), technology has always been an intimate partner.
The fifth Cultural Conversations event, hosted by the National Museum of Australia, was focused on Exploring the shared work of STEM and GLAM. The topic invited speakers to respond to the current moment of flux and change in regards to technology and innovation for those of us working with collections as curators, archivists, researchers, conservators, and more. The arrival of publicly accessible tools like ChatGPT and DALL-E from OpenAI have propelled terms like (AI) Artificial INtelligence, (ML) Machine Learning and LLMs (Large Language Models) into our daily conversations.
They have also created excitement and anxiety across a range of professions from educators worried about students outsourcing their essay writing to ChatGPT, screen writers and actors going on strike to protest plans to replace them with AI developed content and artists protesting the use of images of their art for deep learning without consent. At the same time these emerging technologies offer exciting possibilities as tools that can help us to organise, discover and work with cultural collections in new ways. Our panel, chaired by Craig Middleton, offered contrasting points of view and looked at the issues from the perspectives of archivist, archaeologist and conservator.
Our first speaker Tamara Osicka, a long time Sound Archivist at the National Film and Sound Archive told us about the ways she was using the AI tool Curio tool to open up the sound and moving image archives at NFSA. It offered an overview of her talk delivered at the 2023 AMAGA (Australian Museum and Galleries Association) Conference. The tool is a trial (ending in 2024) and is being used to help staff process, organise and enrich the huge amounts of data that are being created as they digitise their collections. Some examples of the uses include using AI to ‘read’ the titles and other text data on digitised film and moving image files, which can then be added to as searchable text. Tamara commented that the AI tools allows for a different type of discovery that is often more serendipitous than traditional approaches (i.e. putting a text keyword in a box). It can search for images of people that might not be flagged in traditional metadata based on facial recognition. It is also being used to handle some of the more sensitive parts of the collection, like the sex and erotica collection, where staff might come across sensitive or confronting images. The AI tool curio could be used to flag these beforehand and offer a safer browsing and search experience to staff and other end users. While the tool is useful and has huge amounts of potential, Tamara flagged that all the work still requires checking by human curators, it’s a tool that supports, rather than replaces their work.
The next speaker Cate Frieman, Associate Professor in Archaeology at ANU, gave the flipside view. Archaeologists aren’t that worried about AI takeover as the work is complex, physical, and site dependent. Cate also asked for collections and curatorial staff to consider what is being lost in the drive to digitise collections. While digital photos of objects and catalogue records made available online offers convenient and broader access to people who are remote from collections, it is often a poor substitute for examining the physical object. Often only one photo is taken, they don’t capture details and they hide elements that require physical examination. Cate’s experience in Europe is that while basic records are made available, full literature attached to objects and collections, such as notes, reports and other ‘grey literature’ is usually still only available to people who make a visit to the physical museum. But, because there is a perception that digitisation has allowed access remotely, access on site is getting harder with resourcing not provided for this type of access (i.e. no staff available to open collections). Another issue Cate shared is that creating masses of image data through photography, is not necessarily an improvement over older techniques of capturing objects. Methods such as line drawing and sketching, traditionally a key approach for anyone working on a site, require a closer observation that improves understanding of an object.
The last speaker was Asti Sherring, who started at the National Museum of Australia last year as the Manager, Changeable and Digital Collections. She talked from her experience approaching digital conservation and digitization as a conservator. She developed her experience working with time-based media (eg film, objects, and installations that have a durational aspect). Although these are typically digital, from a conservator point of view it is impossible to separate the digital from the material. Digital images, film and data have a material presence, the machines that store data, the screens that we view it on, and so on. The challenge for conservators is to interrogate what it means to preserve an experiential or time-based thing in an authentic way for a museum. It means considering what the intangible and sensory conditions of that work of art or social history object are, not just what files the collection has or what hardware is used to play them. Asti started out at the Art Gallery of NSW but reflected that museum objects that fall under these categories are often harder than art, because typically artists’ set conditions for viewing and they are aware of the inherent nature of change. But objects collected for social history collections often were not originally destined for collections or display. Bringing them into the world of the museum changes their meaning and fits them within a museum ontology.
Question time offered a lively debate over the pros and cons of digitisation and the advantages and threats of new tools driven by emerging technologies like AI. The panel acknowledged the need for discipline, curatorial and researcher expertise to put the ‘findings’ or ‘work’ of AI tools into context. Views included an acceptance that these technologies are ubiquitous and already driving apps and programs we use every day and to approach them with a ‘cautious optimism’. But there was also acknowledgement of the anxiety many people working with collections and cultural heritage feel about the pace of change and the difficulty in understanding exactly how AI-driven tools are working (often called the black box of AI) and what threats they might pose. There was concern about projects and approaches that were jettisoning the step of human checking not being sure what is driving the results (hence the need for human checking).
There was broad discussion about what kind of safety nets could or should be put in place to protect users (be it staff members or the public) from coming across potentially confronting material without warning. It is suggested that some of the new scholarship from public history/humanities areas might offer frameworks for considering the preparation required by users for emotionally demanding work. See, for example, this article which, examines the ‘emotional labor in conducting oral history research, entanglements of responses and responsibilities, and ways of practicing an ethics of care’, as points that may be relevant for those training users for working with or accessing sensitive collections or materials’, and this one, which considers conventions of support that may be available to oral history researchers. Although oral history work is not the same as archival work, in the new environment of mass information, some similar issues around protection may arise.
It remains clear that digitisation, digital-driven tools and now the use of AI and Machine Learning in collections and collections-based research is a core challenge and there is value in creating spaces to have these honest conversations about it and its effects on the work we do.
Leave a Reply