How AI can help libraries – Ines Vodopivec

“Artificial Intelligence is here. It’s not going to go away. We can hate it, we can refuse it, we can have doubts. But it’s here – and maybe we can use it for our own purposes and our own workflows in our institutions.”

Dr Ines Vodopivec is convinced that AI (Artificial Intelligence) 1 isn’t a novelty bolted onto library systems; it’s already reshaping users’ expectations and our definitions of data, and the workflows that connect the two.

Ines began her career in a Franciscan monastery in Slovenia, cataloguing 16th Century manuscripts, and has since then moved between librarianship, restoration and digitisation. Today, she is a digital‑heritage leader and Secretary General of AI4LAM whose career spans deputy directorship at Slovenia’s National and University Library, roles with UNESCO’s Memory of the World and the Europeana Network Association board.

In early February 2026, on a whirlwind tour of GLAM 2 institutions in the Asia-Pacific region, she stopped by State Library Victoria to deliver a presentation that explored practical case studies for AI implementation in libraries across Europe and the US, including the National Library of Norway, Standford University Libraries and Bibliothèque National de France.

Alongside the promise of AI, she weighs ethics, access and the realities of staff skills, arguing for AI literacy and cross‑sector collaboration so libraries can remain trusted, user‑centred infrastructures.

View Ines's presentation slides here, plus further reading and case studies.

Transcript

0:00:00 - 0:02:42
Ana Tiquia
Hi everyone. Thank you so much for joining us this afternoon. I’m excited to have our special guest here with us. And welcome to our, I think, first Creative technologist talk for 2026, which is very exciting. Before we start, I'd just like to acknowledge that we gathered here today on the traditional lands of Victorian Aboriginal clans, and I'd like to acknowledge all their cultural practices and knowledge systems.We recognise at State Library Victoria that our state collection holds traditional knowledge, traditional cultural knowledge belonging to Indigenous communities in Victoria and also around the entire country. And we support communities to protect the integrity of this information gathered from their ancestors in the colonial period. I'd also like to pay my respects to elders, past and present, who have handed down this knowledge and systems of practice for successive generations, for millennia.And I'd also like to pay, and acknowledge and pay respects to any First Nations, staff and colleagues and friends who might be joining us today. My name is Anna Tiquia. I'm Head of Digital Strategy, Research and Insights at State Library Victoria. And I'm extremely excited to be able to present to you a very special guest Dr Ines Vodopivec.Ines said I didn't need to pronounce her last name, but I'm going to give it a try. Because I think it's a beautiful one. You said it means the “drinker of water”. And names are important and they tell us a lot about where we come from and our stories. So, it's really wonderful to have Ines with us at the library today.And Ines is a leading voice in digital heritage, AI and the evolving role and future of libraries. She has a really extensive and very impressive CV, and I'll let her talk more about what she's done and her areas of specialisation. But amongst many other things, her leadership in the library space includes serving as Deputy Director of the National Library of Slovenia, and she's also Vice Dean at Nova University in LjubjlanaShe's also an active member of Unesco’s Memory of the World National Committee. In addition to that, she's here today speaking as Secretary General of AI4LAM, and AI4LAM, which many of you will be familiar with, is AI for libraries, archives and museums. It's involved in global cooperation around the sharing of knowledge and practices and how they're evolving in the libraries, archives and museum spaces.I'd like you to join me in welcoming Ines to State Library Victoria. And thank you so much for coming in, talking to us today.0:02:42 - 0:43:16
Ines Vodopivec
Thank you so much for the introduction. And thank you also for giving me this opportunity. It's amazing. What a crowd! I will start with AI but before I do that, I would like to share a bit of a background of what I have been doing in the last 20 years. One is the official roles that I hold and I have, but the other is actually the research background.And where I got my experiences from. So I am not an AI developer. I am actually a book historian. A cuneicologist. So I started in a monastery, in a Franciscan monastery in Slovenia, working in the 16th century collection, cataloging. So I have all the licenses for cataloging. I know the different materials, restoration practices, etc. But since then I worked in a lot of digitisation projects.That means that, for example, if you want to think about medieval reading culture, or any other aspect of cultural heritage materials, of course, typically today you digitise it, you make it available, you do OCR, HTR whatever technology you use so that you provide data for the final users, for the researchers, for students, etc. So all of these projects that you see here are something connected with that.So we did Digital Encyclopedia of Natural and Cultural Heritage of Slovenia in 2010. We were working on an impact project that's [unintelligible] reader, if you know, that was a huge project. All Europe was cooperating, 2010/2011. We did some Unesco Memory of the World co-financed projects for Incunabula, for digitisation of Incunabula all over Europe. We did, reading in the Middle Ages projects with some partners from Europe.And, maybe I didn't mention before, but I am also on the management board of Europeana. I don't know how much you know about Europeana. You know, you know. Okay, so it's a huge network that we are building, and working together as a community also besides that. And that was something that, of course, as a librarian and art historian, none of us can, escape.We were working on the cataloging databases. We were working on the, EU resources, on repositories, on Open Science. I'm sure you also work with Open Science here as well as we do in Europe. Worked on all different sorts of connecting different systems in one huge, big connected database. Through these years, through these 20 years, of course, everything was kind of connected with digitisation, accessibility and reuse.We still think a lot about reuse, but in later, years, we got artificial intelligence, which is here. It's not going to go away. So we can hate it. We can refuse it. We can, have doubts, but it's here. And we maybe, maybe also can use it for our own purposes and our own workflows in our institutions.I'm not addressing here the generic AI, what we use, for example, to improve our emails, to maybe, make a nice letter to, to the boss, which we don't want to hurt or something like that. So I'm not talking about that, that you can use in any case. I'm talking about the adaptive systems for libraries, archives and museums.But typically today I have more of the library examples. But I also I think 1 or 2 are also more museum, appropriate, but nevertheless maybe just for your knowledge or maybe some insights. So AI is here and of course is changing and it's really, really changing whole systems, how we work and how we operate and how we think also.So it's changing the users and their perspectives on how to use data and access information. It changes how we work with data and how we provide data. And of course, it's also changing the institutional workflows, how we organise our work. And of course, it also changes what the meta or data is or what is it that we should, as cultural heritage institutions, provide to final users. We’ll come back on that.So this is kind of a schedule for today. Okay. So we first going to go into the human perspective. So us as librarians, creators of data, enablers of data access and users’ perspectives of how they perceive us and how they perceive the data to be given to them. Employers, we, employees.Sorry. I more I was also the employer. I'm also an employer, but I'm also an employee. We have have a huge fear of change. That's not something that is only connected with AI. I'm sure many of you, as I did, started with the card catalogs. Then we went to DOS systems. Is it right? Yeah. Okay. It was a change.It was sometimes a rebellion among employees to do that. Then we went from DOS to windows kind of systems. It's true. Yeah. And we know today we have open repositories. Okay. These were all steps of changes. The only thing that is different now with AI is that it's happening too fast. So all of those changes, we had some years to adapt.You can think about what metadata schema you will use in your libraries. If Mark 21 is Ada system, whatever system you use, so you had some time to think about it and to make a strategy to implement it, then you have some years to educate your staff to be able to work. You have some licenses for cataloging, etc., etc. but now the change is shifting in months, not in years.And we as human beings are not able to perceive that change as a positive one. And that is also at the beginning of our fears of what will happen, not only with our jobs, but also for example, if the technology is used, what will be the results? What will be the output? Actually, will it work the way we need to work?Will the output be, the way we want it, etc. etc.. So this is the first thing that we are dealing with as employees, that is skills and knowledge gaps. Of course, most of us, because we are information specialists, we know something about how to work with computers, how to work with data, maybe more than in some other professions, but still, of course, AI is completely different than system than it was before.And because we know and it's a lot of also in the media, the hallucinations etc., etc., So we have this gap of how to deal with a technology that is not already in place in our workflows, and it's not already a part of our systems that we already know. Next thing that is happening, the profiles are changing, so new generations are coming.They are completely into tech. They don't think about card catalogs. They never saw a metadata on the card catalog – how it should be written, how we know it should be written. Then, of course, also the quality of services. We're also thinking about, what will be the quality of our services, because we still need to be the trustful institution that provides information to the final users.Will the quality be still the same as it was before? So then we have users. Okay. On the one side is [unintelligible] on the other side users. So they are completely different than they were before. Why? They are used to ChatGPT, the use of generic AI. They're used to typing a few words in a Google prompt, and they get all the results immediately.That is true today, but if we think back in any IFLA conference that I remember from before, it was always ‘how can we access our final users if they're not coming back to the library?’ Everybody's going to Google. That was even before AI. That's nothing different than it was 20 years ago. Now., the only difference is that every generation is going to Google not only maybe the young ones.So also, for example, students or some other that needed to come to the library to have reliable resources for their research. They, of course go first to some generic tools. So access to information is different. Then search information also. So they search and they want to have the answer right away. They don't want to spend hours searching, browsing on the shelves, for the materials that they can use for their own purposes.They expect information differently. If before a librarian could sit behind the counter and say, yes, yes, you have to search for the shelf number in the catalog, and then you go to the third floor and then I don't know in which, which part. And then you find that book you are searching for. Well, they don't expect that anymore.And they don't want to do that anymore. They want to have the information as an abstract from all the resources you have in digital, library, in the repository, in any other collection that you have. And that is a huge change that they expect from us. And of course, they also process and use information differently. So the generations that are coming with the educational systems that are now implemented already on the university level, the new users are also more knowledgeable of how to use technology, and they also are aware if they use AI tools that they, for example, have to write it, for example, in the thesis or something likethat. So they understand that the new technology is something that they need to also provide information, that they use it. So it's a bit of a changing process in that direction. Nobody, cited in the thesis that they're using the library catalog – or did they? Only for special collections, maybe sometimes, but typically for the catalogs, no.So it's a huge change in the user perspective. Institutions are also changing. What is changing? We know we have gaps and needs. We know that if we want to serve the purpose that we have, if we want to serve the society for the one reason that we were established, and also for the reason that we are financed at the end, we have to address the changing environment.So what we do is we educate and train our employees. We did that also before. I was into, for example, in training of Excel youth. I'm sure most of you were here also on the Excel youth trainings, etc. etc. so we do that. So as libraries we understand that education is important. We also train users. In some libraries like the University of Virginia,They also have the training programs for use of AI with library resources. We seek for new skills and we recruit new profiles. I have here AI librarian. I already saw in few cases that libraries are seeking for AI librarian as a job position. So it can actually apply and be an AI librarian employed in a library like it was a data scientist or any other profiles that emerged in the last maybe ten years or so.We are also addressing ethical issues. And for this part, there's no common ground, not on national level, typically, but not even on the regional or global ethical issues are of huge concern. Also, when it comes to multilingual use of metadata, we are all speaking English, but we started also with a bit of acknowledgments and Slovenian smaller language groups.Slovenia is less than 2 million of us are living together. So small language groups, we have issues developing large language models because we don't have enough of data to feed it in. So of course, whenever we are talking about ethical issues is always a case of shall we use modern material? Because if you use historical material, the output is completely different.So there's a lot of, I will say, layers of those ethical issues that are in place. So the workflows are also changing. I started with metadata and I will be addressing metadata further on today through the talk, because I think this is one of the most important, I'll say, workflows that we have in libraries and typically, I'll say at least 30% of personnel in each library is somehow related to the metadata creation.I don't know what's your percentage here, but a national library is typically around 30%. And of course we are, and we can be, a common voice to the, I'll say big players, but then I'll also say on the governmental policy level, providing guidance but also providing support. And also speaking about the use of data on the more I'll say, trustful way on the national level in any country, no matter where, on which side of the world.Okay. But as I said, education is really important. So education and training is one of the most important things that we have to address. And we have to think about in our institutions. If we want our personnel to be able to work with the final users in the decades to come. But we know that because we are librarians and we always do that.And that's nothing new. Okay. The data. Typically librarians, when we talk about data it’s metadata. It's catalog. And it's descriptive metadata. There is the – how do you say, the first one. So the, author and... I don't find the word now. So you have metadata that is typically in the catalogs.But now of course this is completely different. So you don't have only catalog as metadata material. You have digitally born material which can be also full text search. It can be when it's if it's digitised and you have OCR or HTR you have full text search and you have the data that is a full object.You have harvested web. I don't know if you do that. IIPC you know. So harvested web is one of the biggest databases that we can use. But typically in libraries when we have catalogs and we send our users to find information in our databases, we never combine both. So the results that are coming from the catalog or the repositories are never also combined with the results from the harvested web.So that's a huge part of the data that, we kind of we have it, but it’s kind of neglected. But we have it somewhere stored and we don't use it as a proper or as well as we could. So the data, the thinking about the data is changing. Here in the center is a picture of, I will say, a future library.There are a few of them, also. I think, Library of Congress has something similar and Danish Library and British Library also. So this is the library without the librarian. This is just a robot and there's no shelving system anymore. So the book can be, each time in different boxes because it's a GPS driven and the robot finds the book.So, okay, we can say it's wild because the users cannot go inside. There's, too low temperature and too little oxygen to work in those storages. But on the other side, the book can never be lost. Why? Because if you have, users among your shelves, they're putting the books on the on the wrong shelf. It will be very difficult to track it back.So this is a system that is working perfectly, actually. And it's also all digitised data. The National Library of Norway could do that. Why? Because they digitise everything. And I'll show you later on how they did it, also with the help of AI. So they have digitised everything. Everything. They are lacking the material now. So now they're digitising the police archives.National police archives, all the material for museums, and they're all storing this and similar storage, which are, completely automated. Okay. So library of the future, We have to understand why we are addressing, data and why the data is important for us. So if we had before catalogs, repositories digitised, digitally born today.And I'll start speaking about the global players because I think we have to think about our role in the society in the future. Today, this is a very small part of all the data that is interesting for the training of models, for use of models, because you have internet as a huge database, which can be already because it's open, it's already harvested.This is how our models for the language, for example, were trained, they harvested the web. They didn't even touch your collections. They're not even interested in your collections because it's too small and it's not accessible. It's somewhere they're stored at the back. So they're not even thinking about that. They already trained the models on the on the everyday data they can access on the internet.And that's a complete change of where we are and how important we are in the global system of information sharing. This is something that we really have to think about. That's why I'm asking you. We have open science, medical research is open. Chemistry, open. Space, open. nanotechnology, biotechnology. Everything is open. Everything is open. And we are closing our cultural heritage.We don't want to share our cultural heritage. We're closing it down. And we are only the gatekeepers of the ones that can access our cultural heritage. So sometimes I understand I am coming from the... and that's why I started with that. I'm a cuneiocologist. I'm in love with smell of old books. It's amazing. But if we want to stay in the game, we will have to play the game.So here is something that we will need to think about how and what to do. To be there, to be present, to be visible. I'm not sure how is it with you with the AI factories, but in Europe we have this now, all-European infrastructural, movement. And there are a lot of AI factories established, also quantum computers.These are huge, very capable systems that can process a lot of data. GLAM institutions, so galleries, libraries, archives and museums, typically we are all cooperating one way or another as data providers in research projects. I'm sure you're also facing that, because we are typically always invited to provide kind of data in research projects.Our data can be used for AI training models. It can be used in AI factories, which is already happening, for example, Poznań’s supercomputer in Poland, they established a network to cover the whole the national system of science and culture. And they have a hub, I mean, it's a house, where you have storages of, quantum computers and HPCs together to support all the science and all the culture on the national level.That's, I think, an interesting case. I don't have it here. Maybe next time I’ll insert it. But it's something that keeps us working with the big tech. We have our data storage also, which can also provide data. So if you are harvesting web, you have long term preservation storage access, you have also I'm sure you use IIIF II'm sure you use a lot of protocols.So you are already working with and you have some data that can be shared and that can be used and that can be implemented into all workflows so that we can serve the final users. But a final user, of course, although we are typically, serving more individual users or some research group, some research project, typically the final user now can also change.And I'll showcase that to you later on. I’m sure you know that, so I will not repeat it. But the one thing that time I still keep this slide is because of course, what data you give into the model is the information that you get out, it's just a calculation. It's nothing else. Even if they have enhanced workflows in 2025.And there's some, thinking of that we have now, systems that are rational, that are thinking rational. It's not true. The computer is not rational. It will never cry. That's for sure. It will. It will never smile. I mean, you can implement the workflow that it will act as it does, but it's not doing on its own.So you have systems that combine workflows, and it seems like it would be rational because it is built in a way that it combines information that it calculates from different workflows, which was created by human at the back. But it's not. It's actually just calculation. So it's still all about the data that we provide into the model. For that reason, we can be a really an important part of the infrastructures that are building on the data of today.We can be the ones that can, be important for science, for industry, tourism, culture, education, whichever level of national infrastructures that you can think. But we have to have a voice. We have to be visible. We have to stand up. Because even if we are just sitting at the back and, trying to neglect the technological development, we will never get there.So the way to success is collaboration. And here I will talk about a little bit about the artificial intelligence for libraries, archives and museums and how it all started and why I am standing here today. We started in 2018. There were a few institutions working together and established a community. The community started with the Fantastic Futures event, and I'm sure some of you were in Canberra two years ago in 2024.But now we have more than 1500 members globally. And we established also as a, sustainable organisation in December last year with almost now we are at 50 organisations joining with a memorandum of understanding. The institutions are all of the big ones. So from Stanford, Harvard, Rijksmuseum, I don't know, you have them all here.Also smaller ones, because we think that we need to, if we want to make a change, we need to go in the field so that every institution and I can give you my example. So I said, Slovenian, population is small. But also National Library is small because it cannot be big. Okay. So we have to perform all the duties of the national Library.But we have only 130 people altogether. Okay. But you still have to have legal deposit. You have to digitise. We have restoration. All the collections, exhibitions, educational final use – all, everything. We have to cover everything. But we are 130, okay. And we also our finances are also way smaller than of any other, for example, of these institutions.But we could still join because the system is built in a way that everybody needs to go into the process of, I would say, digital transformation. Digital transformation, because and I'll speak about this now, AI is just another digital tool. So what is more, the community is connected with Europeana and the network association where it's also 6000 people.So if you're not, you can be it's free. Join follow our our communities, follow work of copyright communities and other communities and, see what's happening in Europeana. And we are also issuing blueprints, policy papers, etc., to stitch into policymakers and others who are thinking about strategic implementation of AI and cultural heritage.What is more, we are also connected with archival networks. And, I don't know if, you know, Time Machine maybe, but it's also a huge community that's working with the data. So the community is, I'll say, three fold. You have GLAM representatives, you have university representatives, and you have developers. And in all different communities, they work together. And that's the most important part of the structure, because only with cross-sectoral collaboration, we can actually if we all exchange knowledge and ideas, we can make progress and go a step further, because otherwise we are just sitting in our silos and we are not reaching out of the knowledge that we have.And with collaboration, we can actually, because we all have similar issues and also opportunities in digital environment, we can all reach further. What is also interesting is and we did some, analysis of the network is that a lot of heads of collections, directors, managers, etc.. So the ones that need to, to develop the strategies are a part of our networks because they need to have the idea of where we are going in which direction and how to lead organisations to develop further.Okay. So the power of change is in collaboration as such. And the power of change is also in the trustful solutions of AI. And this is so trustful information, trustful solutions, trustful implementation is something that we can provide different than and I'll say, big players, global players because they are using them. Of course, the information that is everywhere accessible.But we have the curated information, we have collections that have been built for decades. Even more. So, we have the information that can be really trustful for the final user. That's why we are trying to develop the skills and educate professionals working in the cultural heritage sector. But we are also trying to accelerate developing and testing of tools.And we have some, I'll say a few – this is for member organisations. This is maybe for later – we have a few workgroups where professionals join together to work on particular subjects and topics. Okay. One is an evaluation group, where we are evaluating different AI tools to use for our own purposes. We have metadata working group also with people working together on the metadata creation.We have speech to text working group for extraction of speech from the for example, videos and etc. etc. and we also have a teaching a learning working group where we provide webinars and materials for the employees, of cultural heritage institutions. These are the, I say, always the old working groups which are established from the 2018 and are already, I would say, old ones, but we also have new ones that are emerging now.Some of them, like cybersecurity, are really interesting for today's work of organisations. So cybersecurity, AI literacy, for example, infrastructures. So how to build protocols to access our collections and similar. So you're all invited to join our working groups. You can also follow us on Slack, or on other channels that we have. And, well, just use the knowledge that you can then implement in your own work.I will showcase something now, which I also think is important. And that is that when we speak about digital humanities, and I'm sure you do that every day, there's no digital humanities anymore because you cannot do research if in humanities don't use digital tools. So there is no digital humanities. It's only humanities. Similar goes with artificial intelligence.There's no artificial intelligence. That's an umbrella term that covers all different, topics of new technology. We have vector analysis, we have 3D AI reconstruction, for example. We have data management and etc. So we have many other technologies that we can use in our workflows. And AI is just kind of covering the overall subject.And the problem with that is that AI is also in media every day for, generic AI, which is completely different than what we are speaking about here. Now. So what we do in our network is that, organisations, of course, use AI for internal purposes. And external purposes as well as they develop their own AI tools and services, and they use the ones on the market adopted and adapted to their own systems and train it on their own material.This is already happening now. This is a workflow that was created by the National Gallery of Art in Washington, DC. So this is their what they do also, I will go now into the examples. So I'll start with strategy implementation. Before I go further, you have here examples, if any of the examples is interesting to you, please reach out to me.I'll get you in contact with the colleagues from from other organisations so we can exchange. They can share with you etc. This is just to showcase because there's never enough time. Okay. So Library of Congress has an interesting, strategy they already implemented. If you would want to know more about it, please contact me or, through Ana and we can share that information further. Then we have, University of Virginia I already spoke about, AI literacy, which is so i mportant not only for the final users, but also the, the staff of the of the library and staff of also the university.So they developed a few, pillars on which they are educating their personnel, but also the students. Then we have, maybe this is interesting. This National Library of Norway supported automation, mass digitisation. I said a little bit before that they digitised everything. So what it did, they okay, they digitised, for example, newspapers, but it didn't do it by volume and number of each newspaper.They just digitised everything. Like they cut the backs of the, of the newspapers and they just digitised the whole collection. The problem was then that they had the human in the loop sitting behind the computer, searching for first and last page of the of each, number. Can you imagine that after eight hours, after a month, it was awful.So of course there was a lot of mistakes also, because the one point you cannot see anymore, which, the pages are the ones that you have to point. So they developed an AI tool that does that for them. It's 100%. It doesn't make mistakes. So it's perfect. So it's something that can take a lot of time and a lot of manual work, which is done in seconds.So this is for example, a nice showcase that maybe would be interesting to use for you. Then again, National Library of Norway, they are not just, using AI to enhance workflows. They also develop their own AI and they are working on the Norway national model. And they developed 16 language models for Norwegian language. They do that because they believe that the Norwegian language needs to be implemented in Google search, the same as English is.So there are only 3 million. I mean, Norway is, very rich country, but their the population is small, so they want to be visible on and they did this, and they gave them the models to Google for free just to be implemented into the searches. What is more, they cooperated with the government and the government supported, and they are refunding the authors, which, publish the books now and also newspapers and publishers to use the modern language.So the things that are now coming out, they can use for training the material, but the government is reimbursing the author for authorial rights for the use of the material for training. So that's also an interesting case. And what is more, they cooperate at the moment with Norwegian Health and Care Services. That's an agency who is a final user of the data they can provide because they're developing an AI model to support all activities in Norwegian language for their health services.So the final user in their, definition completely changed for a person, individual or a research group to a global player like Google and OpenAI and also other agencies, governmental organisations and other sectors. So they are completely redefining who the users of national library data is. So then, okay, we spoke a lot about the metadata creation. I have here an example from the Belgium National Library.They had a lot of and we all have, I'm sure, a lot of backlog with retro catalogisation first. So they implemented an AI tool so that they scan for select pages of a book and the AI tool extracted the metadata from the scanned pages. First, they started with the retro cataloging for that purpose. And now this is a part of the workflow that they do on the national level.And if you're maybe interested, our colleague there Hannes published also a book if you want to read it. So maybe interesting to know. Then you have Stanford University. They are also, of course, working with AI. What they did is and maybe a different case, they have a huge pictorial collection with no descriptions. So with no alt text, they have only just really, really short metadata and descriptions.And of course they implemented an AI tool that can search and browse the affinity of the pictures, for example, style or something like that. Not only the basic descriptions. I think this was presented in Canberra two years ago. Next thing that is also interesting, I think also this was in Canberra is that they have a Japanese printed collection, with an historic, Japanese language that nobody knows how to read.So what they did, they tested six AI tools to produce the alt text to description of the of the collection so that at least it's searchable in some way for the final users. So this is also one use of AI. Bibliotheque Nationale de France. I'm sure you know, this is the one of the biggest digital libraries, at least in Europe.And they have, 100 million illustrations with no metadata. It's impossible to find anything in that collection, really. So they also implemented an AI tool that is functioning on the iconographical layer of the, illustrations. So you can find and search and browse the, the collection. Then we have a National Gallery of Art in Washington. They have a huge collection.They also have Da Vinci for example, and other important pieces, but they also have an archive of, this is a paper sheet that's similar like we had before. And the card catalogs with the paper sheet. And they developed a tool that extracts the metadata. This metadata, archive was produced through the years and years and years of a not very standardised way, from typewriting to handwriting.So the tool is really extracting the metadata well, and is, implemented in the e-resources. And it also gives you, similar results to the ones that you already search for. And it's, it's working on the content level. Then you have, for example, one, institute from Switzerland. They also, developed a metadata search.So pictorial visual search, visual analysis without the metadata and, extraction of metadata from the card catalogs. But this is from the archival collection, but it's some of that something similar that we also have in libraries. Then we have a probabilistic indexing of archival material, which is, I don't know if you have that in your collections, but in our collections, we have a lot of, we say the Book of Death and the Book of Marriages, which is typical in some, archbishopcies, everybody that was born or died, it was inscribed so you can, work with a tool and it will give you some results and thenneed to learn on the basis of what you say, which result is right and which is wrong. So we can teach the tool, to work with the final user when you're searching for a specific name or specific topic. Okay, this is maybe just for interest but it's more maybe for the museums. An institute developed 3D detection by AI. You have graffiti on the frescoes, also 3D scanning together with the AI, and then also detection of the motifs on the 3D objects on the museums, which can be also used for the restoration purposes and detection of the motifs on the materials.Speech to text I already mentioned before, we have a working group which is, working on tools for, extraction of text from the movies and, for example, TV shows and other other materials. What is interesting from their work is that the tools that they're working on now are also, giving you the names of the speakers, not only the spoken word, but also who is speaking.So the tool is detecting the speakers also. In this case, I like I won't say the most, but it's most, user oriented. As I said, the users expect information differently. So what in Luxembourg did they implemented the chat bot in their digital library, and the chat bot extracts information from the digital resources that are available in the digital library.And it gives you the abstract like similar like GPT, but it only uses the resources that are in the library. That means that are reliable and trustful. Of course, if you have an old I'll say medical article in your library, which is outdated and it's not valid anymore. Of course, that's that's the information that is also in the library.But the fact that this, cataloged and in the library resources is still, important for the final user to use further. While it gives you the abstract of everything, all the material it also gives you all the, references to the materials further on. And then what is more, it works in, four different languages.So when I was there, we were searching in English, but it searched also the, for example, German resources. It translated the German resources in English, and it was incorporated into the feedback that was given to me when I was searching there with the prompt. So, it really is multilingual and also, I think, very useful for the final user.And of course, it also works on different electronic devices. I will just touch upon Harvard at the end. They have something similar like Collection Explorer, but I would not go into how they did it. But I would invite you to look at YouTube channel of AI4LAM where you have everything, everything open, and you can search and go through the collection.Also Fantastic Futures ‘25, which happened now in December. So everything is open and available online. So with this I would stop there. Too much time. Okay. And I would of course be very happy to discuss with you of any aspect. You can also say, don't agree with me, which is also good. Okay. Thank you, thank you. [applause]0:43:16 - 0:44:08
Ana Tiquia
Thank you so much, Ines. That was both action packed in such a generous presentation. And I think it's so useful. Particularly I know you had so many examples there, but to see some really concrete practical case studies and applications of AI in libraries and archives. We're planning to have a bit of a conversation.I do have some questions to ask you, but I'm really aware that there will probably be a lot of questions, from the floor. And I wanted to kind of invite you into conversation here, because I know we've got a lot of metadata specialists. We have folks from museums and archives, as well as from the State Library. So to start with, we've got a roving mic.And just a heads up as well that we are doing an audio recording of this session as well. But I just wanted to, throw questions open to the floor and see if there was anyone who had a question or, something they wanted to to mention to Ines.0:44:08 - 0:44:51
Audience member (anon)
Hi Ines, thank you. That was a really insightful presentation. I had a question. I feel like so much, the adoption or aversion, for AI is rooted in either fear or concern. And that's adjacent, I guess, to, core AI and catalog, but rooted in some concern with employment, workforce, jobs, etc.. And from your experience in research, how has the adoption of AI in Norway or other places affected, employment or roles?Were there any transferable skills through the organisation? And what's that newer landscape? What does it look like? Thank you.0:44:51 - 0:50:34
Ines Vodopivec
Yeah. Thank you. I know that's quite a typical concern that that everybody has. But I think the environment is always changing. And I think in libraries we're not going to change from today to tomorrow, that's for sure. What is going to happen is already happening in few cases is that we can use human power for intellectual work.Let the machine do that dull typing part. But you, for example, if you have a special collection, the descriptions that are more I'll say, more in-depth, connected to some historical background and anything else that the tool cannot extract from the object. That's where our interference is very worthy. Of course, you always have human for the quality check in the loop.You have to have that even if you have a machine that produces, I don't know, 10,000 books in an hour, you still have somebody to check the quality at least, on some sequences. But another thing that is also happening is that the profiles are also developing. So for example, when open science came up, there was, huge initiatives how to, reeducate library professionals to work in open science. It’s the same with AI.So what we need is to develop educational programs and systems for our librarians to be able to work with technology further on. It's not just the transition without any support, we cannot do that. And another thing that maybe I would open now and maybe just for consideration, and I'm sure that we most we don't agree, but I will still open it.Do we still need to produce library catalogs? If we envision, and I will showcase to you a digital environment. So in 5 to 10 years, most of the material will be in digital format. We have digitised material. We have digitally born material. If you are a legal deposit institution, you can always in some cases in some countries, we already, adapted new Legal Deposit Act where you, for example, for depositing a physical book, you also provide the digital copy of it already because it was already in digital form when it went to the printer.So it was already from start in digital form. Okay. You let it print it out and you put it in the bookstores. That's true. But you already have a digital copy somewhere. So an author, publisher, whatever has digital. So you can always get already the digital copy. Why is that important? Not just because of AI, but also because, you will not digitise that material in ten years time.You already have a digital one in your long term storage. So that's kind of a plus to that. And if you have everything in digital, the AI tool will search, will browse the collection, and so, extra hours for cataloging – for what purpose ? Does any of the final users ask you, yes, I want to have a catalog. I want to have the structure in particular, a particular structure thatis so important for us. I know I was working on that also. So you know what how the fields are structured, what is written in which field, which punctuation is at the end of each of the metadata, we already done that. But it's really so important. Are our users really searching for that? Is that all purpose so that we use our lives and eight hours per day to think about which punctuation we are going to put at the end of each metadata field?I'm not sure that we are paid for that from the state. So you can be completely, not agreeing with me. I understand, because I come from that, but it's actually something that we have to think about. So what the machine can do, let it be. But we can use our, also, for example, our education to support the final user, whichever that is, it can be medical society.It can be, a mother with a child. It can be providing literacy to children, to the generations to come. We will have to educate our users on how to deal with the technology to understand why hallucinations are appearing and how to deal with the information in media, how to understand. And I saw that in Singapore last week.It was last week. They have an exhibition in their national library providing the final users the differences between what is published in the news and what is really, for example, the photography from another angle or whatever. So they doing a comparison and they're actually teaching the final users to understand the misleading information that we are getting in media every day.So we’ll have to teach them to understand that what they get from each tool, that it still needs a little bit of rational thinking at the back. I'll give you a case. Not long ago, maybe two months ago, there was a director of a university in Belgium, and she was fired because she used, generic AI to produce her speech.It was, some important speech. That's strange in a way. But why she was actually fired? Because she, in the speech, there were some citations from different people, and the citations were wrong, and she didn't check it. I wouldn't expect from the director of a university that she would not check the information that you get from generic AI.So you have to understand it. But okay, maybe she didn't have time. Whatever. It happens. Okay, but we need to teach our final users to use the tools, right, to understand the output also.0:50:34 - 0:51:20
Ana Tiquia
Ines, just following on from that, thank you. I just want to circle back to how you framed the talk to begin with. You were talking about the fact that you see that AI is here and it's here to stay, and it has to be addressed by libraries and archives and museums. But you also mentioned that you saw the absolute rate and pace of change and the rapidity of that change in terms of the fields of various fields of AI as posing a challenge.We also talked a little bit about, trust and trustworthiness of the importance of being a trustful institution. I think that was the language you used. And I was wondering if you've seen libraries, archives of museums managing and adapting with these pieces of change, but also in ways that might ensure or engender trust?0:51:20 - 0:51:56
Ines Vodopivec
Yeah, of course, attention is everywhere. So I am showcasing, for example, good examples that are happening. But of course, even for example, from my colleagues from Stanford who are really using AI a lot, when it comes to the catalog, they're also frustrated. You know, the tool is not working 100%. It's not, still adapted to the systems that they want it to.So this is not, I'll say, very limited. And globally, it's happening everywhere. And that's why we also have to think about it and to exchange knowledge and step together and, try to make a change for our own purposes so that we can benefit from the technology.0:51:57 - 0:52:00
Ana Tiquia
I think we have scope for maybe just one last question.0:52:00 - 0:52:49
Jeremy, Victorian Indigenous Research Centre
Thank you. And that's. Hi, I'm Jeremy, I work with the Victorian Indigenous Research Centre. My question is around, Indigenous collections. There's seems to be a big rush on the pace that we cannot match. A lot of the metadata isn't recorded for Indigenous collections and material in our institutions. We're still learning ourselves as people and human beings and cultural practice and the ethics around material.We can't match a machine, and we can't train a machine with things that I or we don't have personally. The Indigenous people of Norway or Sami people, or Native Americans, have they had experiences with their Indigenous collections being used ethically with AI?0:52:49 - 0:53:52
Ines Vodopivec
Yeah. In Norway, actually, they do the model for the Sami community. So they train on the materials also for their use. So that's one of the language models that they do. But they have – I don't know if you heard about it– but the, the government is supporting the development. So the government is supporting also the modern authors.So it's kind of a common loop together. So they provide the access to the material, but they also reimburse. So the authorial rights are also paid. I mean, the government takes care of that, that everything works in a, in a common system. For other purposes, it's like, okay, I can speak as a Slovenian because we're small language model and something similar.So we are working on the governmental level as well. So it's always about the cooperation between the policymakers, the institutions who have the material and the authors, publishers and the ones that can, I mean, that would need to be heard and also, addressed. I don't know if I answered the question.0:53:52 - 0:54:52
Jeremy, Victorian Indigenous Research Centre
Yeah. I think we have, some policy changes first before we interact with this learning machine. There's a few steps before we’re there I think. There's a lot of larger languages that we deal with. So I'm not sure if you're aware, there’s 250 Indigenous languages and 800 different dialogs across this continent.Yeah, and then about 256 years of of colonisation. All of our collections come from a colonial lens as well, too. So that affects how a machine would pick this up. So, we’re, as colleagues, in the work of doing, unlearning things and addressing that and the cataloging and the metadata aspect. But I think the human comes first, of course, to interpret that before the machine.And I think that's the my question is just a time frame in what order we do these things. But I think you answered it with policy first and then people power. Yeah. Thank you.0:54:52 - 0:56:50
Ines Vodopivec
Okay. In some cases, not long ago, I think October or November (2025). I'm not sure anymore. There was a presentation of student project, research project on Maori New Zealand community, a language based AI model. Also because as I was I already saying before, if you want the language to be or languages, as you say, 250 integrated into the modern tools and services, it needs to be trained on, and somehow. So at one point it will need to be integrated, otherwise it will be left behind.And that's what we don't want. Another possibility is and I showcase here. But maybe it's interesting is that, for example, Stanford developed AI because it said 250 languages. Stanford has that in their collection, and they have a huge map collection, geographical collection. And you know that in map collections you have rivers and mountains in all of different directionSo we cannot do the OCR or HTR because it's going in all directions. You don't have the lines to, to, to work on it. Okay. So what they did, they developed an AI tool that detects, each letter as a symbol.l. So we don't need to have a language because the text symbols is a pictorial analysis.So when I was there, we used some old Slovenian words that don’t exist anymore. And actually it was wasn't before the territory was different than it is now. So it's actually a part of Austria at the moment. But it's in of course, it's in their geographical historical collection. And the tool found all those Carinthian words everywhere in the collection, even though it's not.I mean, it's not Slovenian language, it’s not German language, it’s just a historical term. So it was found because it was searched by the symbols. So each letter, one symbol. That could be maybe a case to think about.0:56:50 - 0:57:06
Jeremy, Victorian Indigenous Research Centre
Thank you. Yeah. And I'm familiar with the Māori one too, because they had already previously recorded all the material through a former radio station to where they had recorded the dialog, and they had it [unintelligible]. And I think we're still yet to do that. Something like that. Thank you very much.0:57:06 - 0:57:08
Ines Vodopivec
Thank you.0:57:08 - 0:57:16
Ana Tiquia
Just to wrap up, I was wanting to ask you Ines if you could just say a couple of words and what you see is a good future for libraries and AI.0:57:16 - 0:57:58
Ines Vodopivec
Yeah. As I said already, I think it's very important that we, educate ourselves, but also our users, that we think about out-of-the-box so that we think about where we are in the global system. I mean, global system laws are not of the world, but, in the network system of information. So what is our role, what we have to focus on, how we provide the users the data, and that we go out of that librarian frame that we typically have in our heads.So, that is framed with the databases and catalogs and out of that, to provide the information in a way that new generations want and need.0:57:58 - 0:58:00
Ana Tiquia
Brilliant. Thank you so much, Ines.0:58:00 - 0:58:01
Ines Vodopivec
Thank you. Okay.0:58:01 - 0:58:03
Unknown
[applause]