User:Susannaanas/tmig
AvoinGLAM invited GLAM professionals, researchers, wikimedians and other open knowledge advocates to explore the emergent area of AI together during two crispy chilly but sunny days in Helsinki.
The event piggybacked on one hand on the Wikimedia Hackathon arranged in Tallinn during the weekend before, and brought a number of wikimedians to Helsinki for an extended hacking experience. On the other hand, it was a hackathon the Finnish GLAM sector was interested in exploring AI practices together.
Osma Suominen from the National Library of Finland (Annif & FintoAI) presented 5 important points in working with AI in the library context.
- Use AI to make the world better
- Use the smallest AI that works
- Don't depend on corporate AI
- Evaluate & create data sets
- Be open and transparent
Creating and Using a National Linked Open Data Infrastructure for Cultural Heritage Applications and Digital Humanities Research: Lessons Learned
editEero Hyvönen, Director Aalto University Semantic Computing Research Group (SeCo) Dept. of Computer Science; Professor University of Helsinki, Helsinki Centre for Digital Humanities (HELDIG), presented the Sampos, 21 linked data repositories created over 20 years, ranging from data on museums, students, war events, books, law and more.
Explainable AI is needed for humanities research, and symbolic LOD is a way to go!
Fiona Romeo, from the Culture & Heritage team at the Wikimedia Foundation, discussed the organisation's role in supporting Wikimedia projects. She emphasised partnerships with museums to improve the representation of cultural heritage online. When it comes to AI, Wikimedia has guiding principles that include equity, transparency and human-led augmentation.
We can prioritize human understanding and contribution of knowledge back to the world – sustainably, equitably, and transparently – as a key goal of generative AI systems, not as an afterthought. Selena Deckelmann, Wikimedia Foundation statement: Wikipedia’s value in the age of generative AI
How can knowledge commons communities use their leverage to advocate for the needs of other parts of the truth infrastructure of the web (e.g. journalism), on which knowledge commons communities often rely? How can knowledge commons communities partner with institutions that create knowledge in different ways for shared goals? Bellagio research agenda.
Three examples of 'human-led, AI-assisted' cultural heritage work included AI-assisted design for International Museum Day activities, HTR for under-represented languages with the m:Wikisource Loves Manuscripts project, and 'suggested alt text for accessibility' to improve image accessibility on the projects.
Johan Jönsson presented the latest AI experiments of the m:Future Audiences team at the Wikimedia Foundation, where he works as Movement Communications Manager. The team explores changes in online behaviour and technology in relation to Wikimedia projects. They focus on experimentation rather than product development, and aim to quickly and cost-effectively gain insights into issues such as the impact of AI on Wikimedia.
The first experiments involved integrating Wikipedia into ChatGPT, but resulted in limited audience expansion. The ongoing "Citation Needed" Chrome extension experiment aims to assess Wikipedia's role in verifying online content. Future experiments include simplifying the process of adding factual information from external sites to Wikipedia. Throughout these efforts, the focus remains on using AI tools as complements to human-curated information, rather than as stand-alone sources of information.
George Oates, Director of the Flickr Foundation, presented the exploration of AI image generation using Flickr images inspired by the iconic 1955 exhibition The Family of Man at MoMA, organised by Edward Steichen. Students from Sacramento State College curated a publication called A Flickr of Humanity from Flickr images, re-imagining themes from the original exhibition with contemporary imagery.
The team set out to re-create the original exhibition photographs using AI.
- Using the original photographs as the source, they wrote captions, or fed the images to a caption generation program.
- Plugged each type of captions into an image generator (Microsoft Bing Image Creator)
- Generated different options with two iterations
- Compared the two captions as prompts
As a conclusion, this explorations of synthetic image generation underscored the crucial role of human intervention and prompted to critically evaluate the scope and scale of projects. A Generated Family Of Man is available at the Internet Archive as well as the publication of The Family of Man of 1955.
Kaj Arnö from the MariaDB Foundation, which oversees the development and maintenance of the MariaDB open source relational database management system, presented their approach to AI. Together with Robert Silén, he coordinates the work of Projekt Fredrika, which focuses on the representation of the Swedish in the Finnish Wikipedia. Their three goals in using AI to improve the content of Wikipedia:
- Increasing the amount of high-quality information.
- Identifying missing neutral point of view.
- Spreading quality information between languages.
AI can serve as an infinitely eager young junior research assistant but the junior assistant must not be allowed to update Wikipedia directly.
Connor Benedict, Open Culture Coordinator at Creative Commons, discussed the intersection of AI and open culture, highlighting the opportunities and challenges it presents. The discussion covers various aspects such as privacy, labor concerns, and the risk of technology misuse. Copyright issues related to AI-generated content are also explored, alongside calls for contributions to improve AI software within the Open Culture Movement. The presentation emphasizes the potential benefits of AI in areas like automated descriptions, translation, and combatting illicit trafficking. It concludes with an invitation to a webinar on AI and open culture, with gratitude expressed to the participants.
Alicja Peszkowska, from Open Future Foundation, talked about the importance of democratizing AI development and usage and emphasized the need to counteract the concentration of power and promote open access to technology. The answer(s) may lie in
- Democratizing AI development
- Democratizing the use of AI tools
- Redistributing the value generated by AI
- Commons-based governance of AI systems
Hack projects
editWorking on the hack projects started on the first day with pitches of the prepared projects and introductions of all participants. That transitioned into people convening around ideas that interested them, and they could continue planning in the saunas and pools in the sunny and chilly Helsinki evening.
These were the projects that finally presented their outcomes on Tuesday:
- Dreaming of workflow for digital humanities Sarianna Silvonen, Julia Matveeva, Saga Jacksen, Ari Häyrinen
- Governance of a collaborative visual name authority Bart, Alicja
- Generate alt-texts for historical images Osma Suominen, Mona Lehtinen, Harri Hihnala, Julia Isotalo, Lu Chen, Vertti Luostarinen
- Vector database of Senates Justice Department Juho Inkinen, Atte Föhr, Mikko Lipsanen, Ilkka Jokipii
- Tagging Short Fiction with AI Writing Atlas (Jennifer 8. Lee & Kevin Payravi)
- Designing Human Scale Preference Signals Connor, George, Christos, Sophea, Julia (see slides)
- AI for Wikimedia Commons Ferdinando Traversa, Jen S., TuukkaH, Tohaomg, Piracalamina
- Using open clip model for searching Journalistic photo archive photos Kimmo Virtanen (see slides)
- Question: How to make promoting multilinguality easier in a digital service like Finna? Susanna Eklund & Niko Partanen, National Library of Finland (see slides)
- Truths or truisms: How to evaluate AI texts? Heikki Kastemaa (see slides)
- Future conversations Johan, Future Audiences
Further ideas from the Ideas page:
- Authorship of political artists Liisi Soroush
- Embodied creative process in the context glassblowing Liisi Soroush
- Hot topics in the Finnish local letters of the 1860s TuulaP
- GenAI for Moroccan Arabic Ideophagous
- Designing human-scale permission for machine-scale operations George Oates (merged into Designing Human Scale Preference Signals)
- Preference Signals in Open Culture Connor Benedict (merged into Designing Human Scale Preference Signals)
- History of the Basque Country in 100 objects
- Summary of all knowledge Susanna
Documentation
editThank you!
edit🙌 Our team Sophea, Janne, Tuomas, Tove, Julia, Liisi, Tuukka, Kaisa, Sanna, Kare, Teemu, Ilkka and Maryam as well as the organizers AvoinGLAM, Open Knowledge Finland, National Library of Finland, National Archives of Finland, University of Turku, CSC and KIRAHub. 👏 Speakers. 🎉 All participants giving their time and sharing their knowledge.
🙏 A heartfelt thank you for the support for the event allowing us to share it as food and sauna: Wikimedia Foundation, Creative Commons, Wikimedia Finland and Flickr Foundation!
🧖
.jpg){kind=small}
