As explored through Europeana Pro news earlier this year, AI has huge potential for the cultural heritage sector. It offers the possibility to generate extensive amounts of data to enrich heritage collections, making them easier to explore and interconnected. Beginning in December 2019, the EuropeanaTech AI in relation to GLAMs Task Force aimed to carry out a horizon scanning exercise to investigate the role and impact of this technology in the sector, and gain a greater understanding of projects which use it.
Surveying the sector
In September 2020, the Task Force surveyed professionals working in galleries, libraries, archives and museums (GLAMs), research institutions and the wider industry (including technology suppliers and the creative industries). The survey received 56 responses and the results offer valuable insights into the use of AI in the cultural heritage sector.
Almost all the respondents (91.8%) were interested in at least one AI topic, and more than half of them (54%) had expertise in this area. Several shared that they were already working on AI-related projects which were mostly aimed at digitisation and discoverability. However, many people also reported challenges in working with AI, particularly in relation to the skills and expertise which projects required of staff and a lack of appropriately annotated training data.
Insights from interviews
In addition to the survey, the Task Force interviewed eight cultural heritage professionals from various European institutions. Their responses add rich perspectives on the approaches that cultural heritage institutions are taking towards AI, and are included as case studies in the report.
Everyone who was interviewed agreed that AI has great potential for cultural heritage, and wanted to investigate using it further. It was clear from many responses that AI will play an increasingly large and valuable role throughout the activities of cultural organisations, especially with regard to access, metadata extraction and enrichment. However, as in the survey, many people referred to challenges when working with AI, stressing the need for cross-departmental collaboration, the challenging lack of data with suitable annotations and the complexities of integrating AI into existing infrastructure. They also expressed concerns regarding ethics and how best to demonstrate and communicate the value of applying AI.
Reflections on the report: challenges and opportunities
From the results of this Task Force, it is clear to see that AI and Machine Learning (ML) projects are already being carried out in GLAMs and have been for several years, even if they have not always been visible. We hope that the results of this Task Force will be one way to share and promote this work!
While we have seen obvious interest in and enthusiasm for AI across the sector, there are also obstacles. People reported that it can take a long time before the results of an AI strategy or research become apparent, which means that it can sometimes be a challenge to convince organisations of the necessity of the work. The report gives insights into the main challenges faced by GLAMs, ranging from access to training data, ethical considerations, and issues related to scaling up projects to institution-wide implementations. AI is not seen as a 'quick technology fix' for the humanities, and precisely because of this time-consuming aspect, data and knowledge sharing seem essential for the continuation and success of these projects.
Future plans
The EuropeanaTech community will continue to support knowledge exchange on AI, and seek further collaboration with other initiatives to increase its impact. The Task Force and the EuropeanaTech Steering Group have outlined action points for this future work which focuses on the themes knowledge exchange, data sharing and strategic input:
We want to achieve knowledge exchange through the EuropeanaTech x AI webinar series and by organising other events which focus on specific topics and collaborations with Cultural AI Lab, AI4LAM and The Museums + AI Network. If you have got an idea, please get in contact with Europeana’s Research and Development Manager Antoine Isaac, [email protected].
Following on from the EuropeanaTech challenge for AI/ML datasets, Europeana has begun to publish datasets of potential interest for AI applications under the Europeana community umbrella in Zenodo. The first contribution is a style classification dataset from the V4Design project.
Strategic input will build on the reflections from this final report, including integrating and linking new AI systems into existing systems, publicly sharing models trained on domains specific to cultural heritage collections, implementing AI in a way that is considerate of ethical and legal, and reducing the carbon footprint of training models and processing at scale.
Keep following the work of EuropeanaTech to explore how this progresses, and download and read the report in full!