Artificial Intelligence (AI) and the Challenges for Public and Global Intangible Cultural Heritage Governance

Hanna Schreiber (UNESCO Chair Holder): No one here needs to be convinced about the huge change we are observing and participating in that is related to artificial intelligence. We would like to think about how to bridge culture with artificial intelligence, what it means to have culturally sensitive AI, and what the challenges are.

Bartosz Pieliński (Roundtable moderator, UNESCO Chair Team Member): Our foregoing cooperation focused on implementing AI tools to study policy design and cultural heritage policies. It led us to think about not only how to implement these AI tools to understand how UNESCO works but also about what kind of policy UNESCO should develop regarding AI.

And it’s very important in the context of this panel, to emphasize how culture is important to the development of AI because usually when we talk about AI and the success of AI and what was necessary to produce this success, we emphasize the powers of computers, chips, and the ANSI mathematical models. But we wouldn’t have any kind of powerful generative models without culture. These models are possible because we input culture into them, training them on terabytes of images and texts. This is the third pillar, which is very important for the development of AI and has implications for culture because culture is used to develop those models. And, in contrast, those models influence the culture. This is the perspective framework we would like to discuss here.

Rimvydas Laužikas (Department of Digital Cultures and Communication, Faculty of Communication, Vilnius University), The role of AI as an analytical tool in the field of intangible cultural heritage: the case of gastronomy heritage of Lithuania:

When I think about traditional intelligence, I think more about a tool, but less about intelligence. Yes, it’s a really intelligent tool, and powerful too, and in some situations when I discuss with ChatGPT I feel that ChatGPT is disappointed by me.

Like all artificial intelligence-based tools, we must understand that it’s not some artificial intelligence in general, as it is a set of different tools that gives us different possibilities, and as all tools have their limitations. And we must understand this.

Another thing you need is a real problem. Because when people come to AI without any problem and ask something, and then post the answer of AI on social networks, commenting how weak AI is… We need to approach AI with a deep understanding of our problem, because we must ask very focused questions. I can’t ask questions in general and actually after this I must check the answers. Without a deep understanding of the problem that I am analyzing, I can’t check the answers and it will have some mistakes.

My example of working with AI in cultural research is related to the analysis of historical sources. I will focus on UNESCO’s list of documentary heritage, memory of the world. The current situation of individual intangible heritage usually is a result of long evolution. And understanding previous stages of a given tradition is very important for its preservation. And actually, a documentary heritage is a very important tool in this respect.

The case study I want to discuss is related to the gastronomy heritage in Lithuania. We have lots of documents from the XVI-XIX century that are digitized and OCRed (optical character recognition), meaning that artificial intelligence has a possibility to read these texts. The research problem is that I do not need all parts of this text: I need only very short, very small parts of text related to culinary heritage. Usually in these kinds of documents, we do not have so much information about culinary heritage. There could be 1,000 books without any information about culinary heritage. I have the possibility to read 100,000 pages and find something. But I also have the possibility to put these documents into ChatGPT and find something. The first method will take the time of all my life. In the second – just a few days.

The database I was working on myself was composed of more than 100 printed and OCRed documents. It’s about 85,000 pages in Ukrainian, Polish, French, German, English, and Russian. I used ChatGPT and a tool called AI PDF. It allowed for automatic analysis of my database. It resulted in a list of fragments that were manually checked in second place. I just needed to read these fragments and in a lot of cases, I could find out more information in these pages than in analysis done fully manually. The results of this operation were: one full day of work, around 600 fragments of text found (AI automatic analysis), from which 40% of them were related to Lithuania and 50 fragments were truly useful for the research topic (gastronomy heritage of Lithuania). So, to sum up, I can say that I’m a techno-optimist.

Piotr Górajec (Deputy Director of the Museum of Wilanow), AI in Museum:

The Museum of Wilanow is one of the oldest here in Poland, but also one of the most AI-ready museums here in Poland, with a very robust system of collecting data which could be potentially analyzed by AI.

The first and very important aspect of AI presence in museums is the change in the experience of the exhibition. The ways of immersion into the exhibition will be changing very fast. In Wilanów, we are thinking about such a new model that could be an AI-assisted audio guide who will be able to talk with the visitor and answer questions about the exhibition, artifacts, and context of this story. It’s very easy to do. It’s very simple. You only need good, huge databases and an LLM model (large language model). I hope we will have something like that in Wilanów in the next few months.

AI also gives new opportunities in terms of innovations in inventory management. I think that in the near future inventory tools will be cooperating with AI models. We have now in Wilanow new inventory software and we are talking with the developer about some AI solutions to implement in this space.

Another area in which using AI will be bringing some benefits is the area of digitalization and virtualization of collections, as well as an area of conservation, monitoring, security, and monitoring visitor satisfaction.

And the fact I think it’s most interesting of this, it will be changing the conservation, monitoring, security, and visitor satisfaction analysis because we have a lot of data. In last September we talked about some solutions in this area but it was very difficult to do it very fast and cheap. Half a year later it is possible. I hope at the end we will develop a tool that could be a personal assistant with AI for the conservators.

The museums have to change with AI, like the whole world. We have a problem, of course, with the staff, with the money, with the ideas too. But in my opinion, it will be easier and easier technology and from the next months or two or three years, we will be at a different point. I agree with the CEO of the NVIDIA company who said that in the future we don’t need a programmer because we will all be programmers, using our natural language, with the help of AI.

Przemysław Biecek (University of Warsaw / Warsaw University of Technology):

I’m a programmer and computer scientist and I’m here because actually I believe that to build good AI models, it’s not enough to have good computer scientists. It’s not enough to have a good computer, chips, and data. We also need to have good research on culture and heritage. I’m here because I believe in doing my work as computer scientists working on AI models, I need input from you – culture experts. I need your experience. And I’m here to learn from you.

Let me share some experience from a project developed by my students. They were working on large language models and other models. They were checking how models recognize images. And they found out that models recognizing different types of food are doing pretty bad on dumplings. They were only able to recognize Chinese dumplings, and no other types of dumplings. Why? Because the training data used was from the United States, so the area of the world where dumplings are most often associated with Chinese dumplings. Does it matter for big tech companies? Do they care about dumplings?

The same problem happens also for other features like gender or race. Research that was conducted a few years ago found that models for face recognition are working much better for white people and males. They’re doing a pretty bad job with recognizing females, especially in the darker skin tones.

These biases against women and against race are actually pretty big problems. Not only because of the media but also because of the large markets, like in Africa and Asia. It’s not just a funny story about dumplings, but it’s also something deeper. So you might now ask: what can we do about this? It looks like the data is biased. What can we do? One idea was just to make the training data more diverse: let’s make some representatives from different skin colors or genders. It sounds like a good idea. And actually recently we have a model that was trained like that. The models were asked to generate some illustrations of soldiers from Nazi Germany, images of a Pope or Founding Father of America. And the result was, for example: a black woman as a Pope. So the conclusion is that making these models’ data more diverse is not a solution if we don’t understand our heritage. If we don’t understand our culture, we will end up with such stupid mistakes, creating realities that do not exist. What could we do about it? I think that the answer is: in AI research we need input from a good researcher from a heritage area.

I asked Google and ChatGPT who actually cracked Enigma and I was surprised to learn that the answers from these two tools are different. From Google, the answer is Marian Rejewski. While from the ChatGPT answer was Alan Turing. Of course, the true history is more complicated, but it’s kind of surprising that tools would affect the reality that we perceive.

We need input from heritage and culture researchers because there are a lot of questions with no single answer or no single global answers. Questions that can be answered differently in different regions. I think that we have really good research on how heritage and culture is encoded in models that will be used massively.

Discussion: The question of younger generations and transmission of heritage

Monika Stobiecka: I’m working on a habilitation book on digital heritage. What I see is very problematic with digital heritage is often the fact that the collaboration on digital heritage is not conducted in synergy with programmers. And the last group that is usually omitted in designing this digital heritage solution options are younger generations, and here I mean that the solutions are usually designed by people coming from the millennial generation that have a completely different approach to technologies than for instance Gen. Z. Putting a bunch of tablets in a museum or Goggles that were already something new in 2011, but they really failed as an innovation is nothing interesting for Gen. Z. So I think we need to be extremely aware and maybe a bit techno pessimist in designing digital futures.

Piotr Górajec: OK, I agree with you. Technology can be a barrier between a visitor and heritage, but I see that the solution could be natural language interfaces. And I’m pretty sure that we need young employees in museums because we are old.

Przemysław Biecek: I would like to actually hire young people and people with diverse experience, but this is super, super hard in Poland. I spent six months hiring a postdoc. She’s female after psychology and it was super hard to hire her as a computer scientist because for people in our discipline, it was really hard to believe that after psychology you can do research in AI. But we succeeded after six months. So we need more diverse places like that and diverse groups like that. So I really believe in this category of diverse experience.

Rimvydas Laužikas: When we think about the relationship between technology and generations, I think that it is very important to look much deeper than only technology. Because when we think about how technology works for society and for different people, they can find the different informational behaviors related to technology. But actually, if I think about, for example, the target audience of young people, from 10 to 15 years old, I must look at the technological solutions that they use. And at the deeper technology-related information behavior which is typical for this target audience. What does it mean? It means that if I understand technological behavior, informational behavior, I can use all possible technologies. Recognizing the patterns is crucial. And, for example, one of them is gamification. Computer games are very, very popular among audiences of the younger generation. But if I want to understand what kind of gamification could be used for my educational work in a museum, I need to analyze what kind of computer games are most popular in this audience. And probably to understand what gamification solutions are engaging.

Actually, there is another big problem when we analyze the younger generations, especially when we think about differences between my generation and younger generations. I like the metaphor of Marc Prensky about digital natives and digital immigrants. What is a very big problem? It is that while analyzing informational behavior of another generation, I will be using the concepts related to my generation. For example, if I want to describe creativity among younger generations, but I will operationalize the creativity concept by way of my generation, it will lead to misunderstanding. My research will show how many people in younger generations fit into the my generation concept of creativity.

Michał Malinowski (Director of the Storytelling Museum): I think the topic of artificial intelligence is so major that we need to make one or two one-week
sessions because issues are so important and artificial intelligence is challenging our lives so much. When you talk about the experience in the museum, it will
be completely personalized due to artificial intelligence, offering a unique encounter distinct from any previous experiences we might have had. However, we
need to think about how to flourish, how to use these tools to really empower youth and the young generation in an experience of intangible heritage. Because
this is the most important question: an issue of how to transmit to the younger generation intangible heritage and how to use artificial intelligence to do that. How
to enable the younger generation to get interested in traditional culture because there is a gap between these traditional tellers, transmitters of the tradition of
culture that are not completely familiar with technology or AI, and the younger generation, which is completely growing up as not these immigrants but natives.
How can we educate children to get interested in intangible heritage, which is everything?

Piotr Górajec: We are in a very, very interesting situation because we don’t know where we go, but we go very fast.

Rimvydas Laužikas: I think that ambition to transmit all intangible heritages is a big ambition, and this never happened in history because in all periods of history, in all evolution of traditions, some of them got lost, some of them changed, some of them were adopted to new realities. Our problem probably is that our new realities are changing very fast.

We need to think about how to preserve some parts of our heritage, but in the new realities. There is a very good example of Ukrainian „wyszywanka”. They understand that to preserve the tradition of dressing in full national costume is impossible nowadays, because people live in other conditions and they can dress the national costume only occasionally in some situations, but not in everyday life. But Ukrainians make a very good solution to preserve it, choosing one element of national costume and importing it to everyday life of contemporary people living in urban areas. It could be helpful to understand what ideas from our experience are most fruitful and most engageable for digital natives.

Przemysław Biecek: Let me share a bit of a different perspective. I have two kids and they are very interested in heritage but not necessarily our heritage. They consume content through YouTube, Tiktok, and other media. They are very, very interested in many topics which are exposed, recommended, presented to them, so I still think that in our discussions, it’s not only about how to use existing tools, but actually how to shape these tools to tell stories that should be stories about our heritage.

Agnieszka Pawłowska-Mainville: One of the key things that I’m involved in is actually language transmission and language revitalization, especially of indigenous languages such as Dakota or Anishinaabemowin. So my question is, while I see a lot of young people using AI to document languages, I’m wondering if you think of the chance of digitizing and personalizing experiences or doing work in Intangible Heritage actually diminishes linguistic ability and valuation of learning a language because everything is now personalized to your own language. Therefore, there is no need to learn another language. Where do you think that goes?

Rimvydas Laužikas: When we think about languages at different levels on how language works, one is the transmission of information. But actually on this level, we have more or less good tools and I can read the Arabic text and understand the Lithuanian. Another level is some deeper semantic levels of language. We have a language-related worldview. And I can understand some languages technically for transcription, not translation, but I can’t, absolutely can’t understand the whole view of these people who spoke in this language. If you go deeper the role of artificial intelligence also grows. Because we need very powerful tools for translating or transmitting the meanings which are much more detailed.

Michał Malinowski: It goes much deeper… every language is a way of thinking. It is a separate system of thought and if we have, for example, African music and African rhythm, it is adapted to the language and the difference in rhythm from one tribe to another constitutes a difference in language. I can’t understand the difference in the rhythm of the two ethnic groups. The music sounds very similar to me, but African ears can perceive very big differences. And I think that’s the point, intangible heritage highlights the diversity of languages, something we must strive to preserve.

Bartosz Pieliński: I’ve recently read a few papers about the Rammstein song “We live in America” that was translated from one language to another using AI-driven models. We can understand the translations, but, as it was said, the semantics level is all English dependent, so we have meanings produced by this particular culture, not others.