How Multilingual AI Bridges the Gap to Inclusive Access

as a bridge to democratic access. Switzerland is very pleased to contribute to this global conversation at a pivotal time, a pivotal moment for responsible AI. Our message, which was supposed to be delivered by our president, is very clear. AI can only serve the public good if it serves all languages and all cultures. Today, linguistic exclusion remains one of the most persistent barriers to digital participation, ensuring multilingual access is therefore not only a technical challenge, it’s a democratic imperative. This discussion forms part of the international arc that began with the Paris 2025 public interest AI process, continues here at India AI Summit 2026, and will advance further when Switzerland will happily host the Geneva AI Summit.

The Geneva AI Summit in 2027. Our shared objective is continuity, cooperation and genuinely global approach to AI governance. Switzerland is proud that this session brings together partners who embody open and collaborative innovation. India’s Barshini Initiative, current AI that emerged from the French AI Summit and then many partners from the broader network of academic and policy institutions of ICAIN, the International Computation and AI Network. Such partners as ELIS, NTU Singapore and of course the Swiss partners ETH and EPFL. ICAIN really reflects Switzerland’s commitment to equitable access to compute data and multilingual models. A notable example is Apertus, which maybe many of you have heard of. It was developed by ETH Zurich and EPFL, fully open and transparent multilingual model designed to support public interest applications across diverse linguistic communities.

As we prepare for Geneva 2027, Switzerland views multilingual AI as a foundation for inclusive digital public services and for strengthening participation across societies. Allow me to briefly, just very briefly outline today’s agenda. We will begin with the announcement of the launch of the three new joint calls under the lead of the Indo -Swiss Joint Research Programme, JRP, which is making a further strengthening of our bilateral ties in science, innovation and research between Switzerland and India. This will be followed by a panel discussion. We have distinguished international guests and I’m very happy to announce that this will be moderated by my colleague Nina Frey, the Executive Director of ICAIN. Thank you so much for attending. I will hand over the floor to the next speaker, Professor Thorsten Svede, President of the Swiss National Science Foundation.

Very warm welcome. Thank you.

Your Excellencies, ladies and gentlemen, namaste. It’s my great pleasure to be here today with you. It’s a moment to highlight a particularly exciting moment in the Indo -Swiss research collaboration. As many of you know, Switzerland and India have long -standing trusted partnership in research built on reciprocity, on joint excellence, and on shared priorities. Today, this collaboration is stronger than ever, and I’m delighted to announce three new calls for joint research projects, as well as the launch of our new Indo -Swiss research framework program, between the Swiss National Science Foundation and the Swiss National Science Foundation. and our Indian partner organizations. This is a really remarkable convergence that underscores both the depths and the breadth of our bilateral engagements.

The three calls for joint research programs span a very diverse range of disciplines and are designed to foster cutting -edge, high -impact research. The first two calls that we launched earlier this year are in the geosciences and in the social sciences. Together with the Indian Ministry of Earth Sciences, we are inviting proposals on natural hazards in mountain regions, a field of great relevance for both our countries as we are each facing very unique geological challenges. In parallel, our call with the Indian Council of Social Science Research opens the door for joint projects on pressing social and societal questions, again strengthening our collaboration in a domain where cross -cultural perspectives are significantly enriching the research outcomes.

And two weeks ago, the Swiss National Science Foundation, together with the Indian Department of Biotechnology and the Indian Council of Medical Research, launched a third call focused on One Health, a topic of real global urgency. This One Health call is particularly important for us. It reflects many months of preparations and close coordinations with our Indian partners and embodies a holistic approach needed to understand interconnected health of humans, animals, and the environment. The challenges we face in this area know no borders, and international collaboration is indispensable. We therefore anticipate a very high uptake and interest and participation of researchers in both our communities. Taken together, these three simultaneous calls represent an exceptional moment in IndusVis research cooperation.

They showcase our commitment enabling ambitious science from fundamental research questions in the natural and the life sciences to complex issues shaped by society, geography or technology. And with each call, we reaffirm our shared belief that long -term co -created research is the key to addressing the major challenges of our times. So building on these strong foundations now is the right moment to announce a new strategic long -term collaboration, the Indo -Swiss Research Framework Program between the SNSF and our Indian partner organizations. We aim to create a program in which all researchers wishing to contribute to the Indo -Swiss cooperation can find appropriate support. Thematic calls on strategic areas will be launched together with our Indian partner partners and remain at the core of this program.

And to this audience, it might not come as a real surprise. that one of the high -priority topics we are currently considering is artificial intelligence. In addition to these bilateral and multilateral calls, I’m also pleased to announce that we are launching several new measures and funding schemes to support collaborative research. With our brand -new Explore, Experiment, and Expand grants, we want to give consortia the opportunities to explore new collaborations, new networks, new partnerships. We want to allow them to experiment with blue -sky thinking topics and methodologies that haven’t been tried before, but we also want to allow them to expand on already established functional collaborations and build them in an innovative way into the future.

We’re also increasing mobility funding for existing consortia to make sure that every project we fund by our program can lead to a durable collaboration, impactful events that connect with the wider world, and the wider society. and early career researchers can truly benefit from the mobility and the capacity building. We plan to hold frequent flagship events, both in Switzerland and in India, to keep connecting our various partners of this program from funding actors, beneficiaries of the calls, policy makers and prospective applicants and early career researchers. So make sure you follow our website and social media and there’s more updates coming soon. I want to extend my sincere thanks to all our partner organizations here in India for their continued trust and collaboration and the two research communities in both our countries that show a lot of enthusiasm and engagement in these programs.

So I encourage all interested researchers here in the room and out there to take advantage of these new opportunities and continue building the bridges that make our partnership so successful. Thank you. Thank you very much for your attention.

Thank you so much. Thank you so much also from my side. My name is Nina Frey or Katharina Frey as my colleague or former colleague, Markus Reuvi, has introduced me. I am the executive director of ICANN, which is this network linking already academic partners from Europe, Africa, and Singapore. And I’m very glad that I have many representatives from the network that will be on the panel and actually also one of the board members sitting in the second row from the Finnish Supercomputing Center. Thank you, Damian, for coming. So we have such a big panel representing ICANN that there’s not even a space for me, so I will be standing here. And I would like to invite my panelists to take seats on the different names.

I will introduce you and hand over the mic to you. In a minute. Please have a seat. turned out there was a seat for me yeah I know we do a group photo at 1225 ok oh now so you have to bear with us this afternoon ok you can have a fast smile I have to stand oh we have to stand how can I get in ok ok so we have to stand ok thank you now we can all take a picture do I have a mask to join me here no ok we have we have a mask to join me here ok ok ok ok ok ok ok ok ok ok ok ok ok ok ok ok ok ok ok ok ok ok ok Thank you.

Thank you. Thank you. Wonderful. Thank you so much. Thank you so much for bearing with us, for taking pictures. We actually talk about language, but let me think about an analogy to pictures. We’ll dive right into the importance of, I would say, as to the language question, obviously also the cultural and the contextual embedment of different AI in the different settings. So again, allow me to extend my thanks to all my distinguished panelists for coming, for also allowing us to show how this ICANN collaboration works from very different angles. The idea of this next 40 minutes is really to try to give a red line, I think you say, between the different summits. Actually, it started obviously in Bletchley, and I hope we can then showcase how this topic of language and cultural diversity was somehow present in all the different summits and unites us all.

Since we’re here in your host country, allow me to hand over the mic to you to talk about the ICANN collaboration. And also to share with us, like, why did Bajini been funded? you had presented your work this morning to me and Alex. It was very impressive how it translated immediately live from Hindi to German to English. But please share with us maybe the next five minutes what your work is, what has it been, and where you’re going. Thank

you. Yeah, thank you very much, Nina, and thanks for inviting me here. Bhashini stands for Bhasha Interface for India, so it is basically looking at 22 languages which are enshrined in our Eighth Schedule of Constitution, which basically says that we will conduct or we will have these languages as the languages to start off with for our work in the regions. We started off as a program for transcending the language barrier using artificial intelligence. In these 22 languages, we have been able to do a lot of work and we’ve been able to do a lot of work So it’s been a very the first place. We had our own challenges but the methodology which we followed was to collaborate with 70 research institutes across the country and the problem statement was actually divided in between all the 70 research institutes.

We were solving five problems to work on. First was automatic speech recognition that means the digital systems should be able to understand what we are speaking in all 22 languages. Then we are looking at the second piece of it which is text to text translation and again bidirectionally in all 22 languages. Third was text to speech which was basically again that the digital system should be able to speak to you that is again in 22 languages. And then we are looking at optical character recognition in 22 languages and also our digital, our dictionary. Which is basically the vocabulary in all 22 languages are not digital. So there was an attempt to digitize all the vocabulary which is around.

That includes names of places, people, companies, etc., etc. We have till now achieved 22 languages in all the modalities. We also have increased the number of languages. Incidentally, in India, there are 100 languages which are spoken or written by at least 100 ,000 plus people. So our journey is not complete when we do 22 languages. We are moving ahead with more languages. So we now have 36 languages on text and we are going to add more languages to move forward. We also have languages which don’t have script and those are basically in the tribal area. So we are attempting to digitize that also and that is being digitized. One of them has been digitized and will be launched in next few days.

We. We also have. So in all of this, we had one basic challenge, which was non -availability of digital data. So the non -availability of digital data, which is oil to the AI models, was basically done for the first time in the world as a brute force data, digital data collection. So what we had done was that we had about 200 -odd people who would go down on the field and, you know, speak to the people on a certain subject. Pick up a picture or any other things so that it becomes the topic of discussion. We will create the monolingual corpus by requesting them to write the same thing or bilingual corpus if they are, you know, having two languages.

And that is how we build the bare minimal digital data. Obviously, when we have done these things, the model is like a child. It only read 100 books, so it will be as intelligent as those 100 books. So we realized that… over a period of time we need to collect more data that means give the child thousand books so is more intelligent and more and that journey continues so we have taken AI as a journey but we haven’t waited for some things to become perfect so that we are in a position to launch them as a product we launched them and built a narrow use cases a narrow use cases in the sense that okay let’s build something for the farmers I will try to give two examples for want of time is one is that we have built up an interface for the farmers where farmers in their own language can ask a question about agricultural advisory and he is he or she is answered in that particular language so it’s a voice first and voice journey so that means I will be talking in voice and you know the answers will be coming on voice so that’s a voice journey sequence the other thing which we had actually experimented on our working is this is a deployed system so it is actually a very large system we are now we are working which is one of the things which have been displayed here is a project called Gyan Bharatam where you know the manuscripts have been made interactive.

Plus we have multiple other use cases perhaps I will come to them during the discussion but means we have about 20 odd of them displayed in

Thank you Amitabh. Thank you so much and I somehow assumed everyone knows it but obviously I should introduce you as well so apologies for that. Mr. Amitabh Nag, he is the CEO of Bajini, the national language initiative and we will be collaborating Alex will be mentioning more later on on that. But before that, I would turn a year back to Paris, where obviously Current AI was started and came out of the Public Interest AI Working Group, if I think. So, Mrs. Aya Bedir, you say? Bedir, sorry for that. She’s the CEO, quite recent CEO of Current AI, a very, very important initiative that amongst others also wants to thank for the topic that we’re talking about.

But please, Aya, I know you come from a wrong background also in hardware. You are launching, I think, this afternoon something very impressive. That also helps… the importance of language diversity. Could you share with us some of your key focus interests and also why you so focus on hardware? Thank you.

Thank you so much for having me. So, my name is Aya Bedir. I, yes, did join recently, about a month and a half ago. Exactly. So really feeling the very warm welcome in India. Current AI was an initiative that came out of the French AI Summit. The founder, Martin Disney, was the special envoy of President Macron at the summit, and the initiative essentially has a vision for AI that is global, that is collaborative, and that is collective. And so the idea is that if we acknowledge that some of the biggest tech companies that are really governing our lives and really governing AI and the way we consume it in day -to -day, they are a handful of these companies, they are big, they have scale, they have a lot of financial resources, and they are very ambitious.

And so the initiative acknowledges that to be able to stand a chance to be an alternative, and to be able to do something about it, and to be able to do something about it, and to be able to do something about it, and to be able to do something about it, and to be able to do something about it, and to be able to do something about it, and a counterpart to these large companies, we must fight scale with scale. And so obviously there is lots of interesting work happening in public interest AI around the world, but oftentimes the work is distributed, the work is decentralized, and sometimes it’s duplicative, and it’s not always additive.

And so as a result, current AI has this vision that we need to sort of bring together and bring more collaboration into the space, but also raise the level of ambition and of financial scale that is taken on. So current AI is a public -private partnership between philanthropy, between the private sector and government. It has initial commitments of about $400 million, but the ambition is to get to $2 .5 billion and hopefully more. The initial… commitments are from the French government. There are also partners, multiple other governments, including the Indian government, the Kenyan government, Moroccan, and many others, as well as from MacArthur Foundation, Ford Foundation, McGovern, and a few others, and the private sector, so Google DeepMind, Salesforce, and others.

So it really is a public -private partnership with the intention of kind of bringing everybody around the table that has sort of the same commitment to public interest AI, to AI that works for individuals and for the public good, and one of the main vehicles of doing that is really investing in open source. Language has been a priority for current AI ever since its inception. The initiative was called Multilingual Diversity, which I know is something everybody here is committed to, and we’ve been hearing a lot about over the past few days. I joined about a month ago, and I’m myself very passionate about the topic, and I sort of expanded the topic to be about culture, diversity, and culture preservation.

So it’s really not just about language. It’s also about acknowledging that culture exists in many facets. Language is one of them, but there are also behaviors, there are norms, there are also artifacts, physical and digital artifacts, and there are many things that are digitized and non -digitized. And so we now talk about culture preservation as one of our big priorities, and it’s something that we’ll be doing a lot of work in. As part of the culture preservation work, also when I came in, there had already been conversations between Current AI and Bashni about doing a collaboration together for the summit. And to be honest, I fell in love with the work that Amitabh and his team were doing and the care that they were taking with their diaries.

And I was like, oh, my gosh, this is so cool. and really the fact that they were going to sort of the source and getting a lot of this knowledge, not just data, this knowledge about the language from individuals and from communities themselves, no matter how small they were. And so we ended up collaborating on a device that will launch later today at 3 .30 in Room 10. I hope you all can attend. I’m not going to say much about it because there’s a drumroll situation that will happen. So you guys can come see. You all can come see for yourselves. But the intention of the device is to really get as far, as close as possible to the individuals and the communities themselves.

There is one concern I have that could be kind of a negative repercussion, I think, of having so much attention on multilingual diversity in a society. And I think that’s a really important thing. And I think that’s a really important thing. And I think that’s a really important thing. And I think that’s a really important thing. And I think that’s a really important thing. And I think that’s a really important thing. And I think that’s a really important thing. that a lot of the big companies and big players have to do all the work. And so, you know, it’s interesting and positive that, you know, the big tech companies are saying we’re going to make commitments to more multilingual diversity and more languages.

That’s good. But oftentimes when they are kind of in the leadership taking these positions, there’s a brute force kind of methodology that they deploy because of the scale at which they operate in. And so oftentimes it’s about scraping data. Oftentimes it’s about taking data without licensing it. It’s about treating individuals and communities as data, whereas they are people and they are not data. And so that’s sort of my concern in this area, and I believe that we have to get as close as possible to the communities themselves and invite them and support them in doing that kind of work themselves. So it’s really about them preserving their own. Their own cultures and languages. and not about us doing it for them in this sort of like somewhat condescending way.

I’ll also say one last thing, which is I myself grew up in Beirut in Lebanon, a very tiny country, but that everybody has heard of sometimes for good and not good reasons. But, you know, Arab language is also very concerned about AI and representation in AI, and we have thousands of different cultures and dialects within Arab culture, and we also have varying degrees of resource availability across Arab countries. Some countries are very resourced financially from a government perspective. Others have very scarce access to resources. So I’m also very concerned about thinking about AI that is more resilient, that operates from scarcity. operates from frugality and operates from a limited amount of resources and looking at that as a positive as opposed to a negative.

So that’s something that current AI will be prioritizing in a big way and we’ll hope to do more of. So hope to see you all at 3 .30 and hope

Thank you so much, Aya. Let me hand over because you mentioned obviously the many announcements that were made as well from private companies to start collecting data. I think it’s fantastic to see that governments can do that as well and that you also invest in this PPP so far. And allow me to hand over to my colleague sitting to my left because I think you can also showcase how also public institutions like universities can also train a model multilingual from scratch. Scratch, not stretch. It was probably a stretch sometimes. Let me introduce you to Dr. Alex Ilich. He founded and is the executive director of the AI Center, a co -founder also of ICANN. And please, could you share your experiences with Apertus, which is this multilingual model, and maybe also mention something on Swiss AI and how the Indian languages we can maybe then present next year in Apertus.

Alex, please.

but basically we were able to build this model and one of the key bottlenecks that we also identified is it’s not just the infrastructure where currently a lot of money is going in but also the talent. Outside of big tech, you have maybe 100 people on the planet who have the experience and capabilities to build such foundation models and that’s not enough. And I think that’s something where academia can change it and I think that’s why it’s important we not just need supercomputer and data centers for the companies, we need it to empower academia. This is very, very critical that we also push this very, very strongly. We named the model Apertus, Latin for open, because we want it to be a foundation where everyone can take it and build on top of it.

So it’s not something that we force up on someone but it’s something that can be a thriving community where each university, each project, each country gets a step further. And I think we will hear later a little bit from the perspective of the Apertus Foundation and also from Singapore, from India. We already heard… There are not many countries that recognize how important that is as a public infrastructure that you really take it serious to develop your own benchmarks and your own data sources as well. Because today, still, if you read LinkedIn, whatever, the majority is driven by benchmarks that the big companies are publishing. And surprise, in every benchmark they publish, they are, of course, the best because they pick whatever metric is usable.

And I think this metric should be driven by what do we want it to be in the cultures and the regions to empower this. And so we have 1 ,000 languages included because we trained it with data on the Internet. As you know, the Internet is not the most diverse data source there is. 60 % of the data in our training set is English. 40 % is non -English. And so what we’re thinking about now strategically is how can we… Increase… the number of languages that are close to the performance we see in English, step by step for the next hundred languages and so on. And this is, I think, like important because many companies that are going in that area and say, oh, we sponsor a data collection effort, they just do it on best effort.

Like you, let’s do something and you don’t know does it actually move the needle. So the next step for us is that, you know, with all the experiences, you know, in Boschini and other parts, I think we can find out now very strategically how much does it cost us to raise the bar significantly, not just make a check mark out of that. So that will be also the hope for connecting forward through the mission of ICAIN and also Geneva next year that we can present also, you know, how far of a progress could we make, like where do we stand today that is really usable and economically usable and to elevate this. I think that’s super critical on that side.

And, yeah, we’re also very happy to be here. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. of sea lion that’s already built on Apertus, and we want to extend the collaborations now globally. For the researchers, we also have a very strong international program where we share basically our compute infrastructure. So that’s also very unique, and we would like to see also other countries to do that because we know that for where we stand with AI today, we’re maybe at 5 % or 10 % of the potential to train the next models that are, you know, including more data, becoming more aware of the physical world and so on.

We need more compute. We need to team up, and I think that’s also a question of how can we collaborate more and share more. And with ICANN, I think, like, in the beginning, we had this, like, the bottlenecks, you know, our compute, that’s why we have strong compute representation. It’s data and benchmarks, and it’s talent. And I think on these three capabilities, we need to jointly increase, and whoever doesn’t have it should be able to easily get the data and the benchmarks. And I think that’s where we’re headed from. sides to do it themselves basically. Thank you.

Thank you so much Alex and also for emphasizing the talent need and I think if I can just add that you mentioned the talent or the capabilities in knowing how to train a model something that and again I’m looking at the supercomputing representatives I mean it’s also a talent in knowing how to build up such an HPC so I think that’s something we could also add to the table but allow me to come back to the models themselves and the really very concrete application. Now I’m turning first left to the north to the Finns. Petri you’re here as obviously as a founding member of ICANN but those representing ELIS Network but also the team Finland if I can say that but you were also member of the age lab of the UN Secretary General so where you one of the recommendation was also exactly this that we collaborate and could you mention maybe more from the perspective of the Nordics, you had already your own language models, but maybe also you can share some thoughts on why you recommended that to the world, if I can

Yes, thank you. So happy to be here. So indeed, I mean, as you all know, Nordic languages are not the biggest major languages in the world. So obviously we take the kind of the preservation of our languages and cultures very seriously. Talking about the H -Lab of the UN, there was just upstairs a kind of a handover to the new International Independent Scientific Panel on AI. So maybe one thing I learned in this UN advisory body was that, I didn’t know this, but like one thing I learned was that access to language and culture is a human right, one of the human rights that all the countries in the world have. And I think that’s something that we have to accept it.

So to me this was a surprise and pleasant surprise because as also like language is already important because we operate with language. But like what Aja was saying, like even more important is the culture behind the language. We have different value frameworks and norms in different countries. So if there’s one size fits all English version AI that we all start to use, what is the value framework behind that? So that’s kind of I think this is a very critical issue. Another thing I learned in UN was that I mean like and that’s why is that there are several global initiatives towards like making this more accessible to all countries. Seven of the UN member states, 193 countries are included in all these initiatives.

119 countries are included in none. So initiatives like ICANN or current AI. So all these. this summit are very important to make this more inclusive. So I now shamelessly steal a quote from Joshua Benjio, who was just upstairs, saying, like, we need to make sure that all the countries in the world are invited to the dinner table, but not part of the menu, but they are dining guests. So I think this was hilarious.

Thank you so much. Thank you so much for sharing. I didn’t hear, but I think it’s a good thing that we can take up. Because also, obviously, food is very cultural diverse. Thank you. So let me turn from the north first more to the south. And to Singapore, you were also quite recently, I think, at the NTU Singapore, which is also the newest member of ICANN. You had already developed, and I think you will share something on the sea lion model, which is obviously, for the ASEAN region, the famous language model. but you also had already collaboration with Apertus. If time allows, because I would also allow, I need to speak, but if time allows, you could also mention something on the importance of sovereignty and language.

It’s wonderful. At NTU Singapore, we’re the newest members of ICAIN, but it’s fantastic that the… And I’ve only been at NTU Singapore for six months, but the conversation that we’re having here is the same conversation that we’re having, about the importance of multilingual diversity, the importance of getting close to the ground, the importance of culture as well as tech. And I’m the dean of a college of humanities, arts and social sciences. I’m a historian. And it’s where my college is in the lead, collaborating with computer science at NTU and engaging with CLI and thinking about AI in the context we’re talking about. So… So, you know, I just point to very, very important… that it’s about culture and thinking about cultural diversity and how AI models, et cetera, reflect culture, how we engage with culture and history, et cetera, as well as simply technology.

And I think that’s something that’s very evident in this conversation. So it’s great to be part of this club. So C -Line is a language model that reflects 13 languages across Southeast Asia. In fact, it includes Tamil, because Tamil is a Southeast Asian language because it’s an important national language within Singapore, with aspirations actually to kind of expand, potentially connect beyond to other parts of Asia. It’s a nationally funded initiative, but it connects from Singapore. So it’s part of Singapore’s public infrastructure, but then it connects regionally and is used by, has good connections with private sector providers across Asia. Southeast Asia platform. in Indonesia on various things, etc. So it’s, and you know, as we were hearing a moment ago, I mean there’s a number of different versions of it, one of them is built on Apertus, so there’s a real synergy here.

And I think, you know, I just want to flag the connection between Singapore and Switzerland, they’re both multilingual, multicultural, you know, kind of relatively small societies, so there’s a very obvious kind of, there’s a very obvious collaboration there. And I think another, echoing again, something that I was saying is, when we’re thinking about AI and we’re thinking about the relationship between culture and language, we’re also interested in frugality, we’re also interested in using resources effectively, and in thinking about how we can you know, draw on sort of deep truths about language and culture without vast amounts of data, you know, kind of with relatively small amounts of data. I mean, one of the, you know, we have languages like Laotian, Khmer, etc.

within C -Line, and so, you know, colleagues are really thinking hard about how you leverage relatively small amounts of language to then produce an effective model in the discussions. Just a couple of additional points, and I’m looking at the clock. Sovereignty is the big word within the AI summit. I’m a historian actually in some ways of sovereignty at the moment. Sovereignty means power. It’s a power that we want for ourselves, for our communities, for our nations and states. But in a sense it’s also about individuals as well, and there’s a kind of complicated relationship between those two things. And so I just wanted to reflect on the importance of sovereignty in that we’re talking about the sovereignty of societies that are neither the US or China in this discussion.

The two big superpowers maybe. And this discussion is about how we can think about a world that is multipolar and is multicolored. and reflects the fact that sovereignty actually is dispersed in the world in which we live in, and that’s very important. And that’s Indian principles of non -alignment that go back to the 1940s and so on. So I don’t know if I’m allowed to use that phrase in today’s India. But anyway, it’s a similar set of principles that we’re talking about. So the dispersal of sovereignty that we’re talking about here and power is important, but it’s also as part of that, I think, to reflect the limits of the nation state, I suppose, and the limits of national approaches to language.

In that, we all live in environments in which people speak complicated… They’re multilingual societies in a minute -by -minute way. People code -switch. They’re speaking Hindi in one minute. They’re speaking English the next, Swiss, German, etc. Similarly, in Singapore, people speak Mandarin, Chinese, then they speak a Chinese dialect, and then they’ll speak English. And so the… Sovereignty is crucial, but we need to… If we’re interested in the… sovereignty of the individual and the power of individuals, then we need to have a more nuanced account of language that allows for things like code switching, and dialects, etc. And that’s something that we’re very much interested in in NTU.

Thank you so much, and thank you to all the speakers. Also allow Annie to speak, because I think you’re obviously from South Africa, but now living in Switzerland and the US, and you lead these linkages between medicine and the AI, and I always think you explain very concretely in your work what if you just take an English -speaking language and train it in a tiny set of local data lakes, how you experience that in the medical field in reality. So if you could share something on that, and obviously also your role in ICANN. Thank you, Annie. Professor Annie, she’s at EPFL at the moment in the interview. She’s in Yale. Thank you.

Yeah, so thank you very much, and I think I’ll take it down to the ground then about the consequences about what happened. happens like really when you are at a patient’s bedside and you ask questions that are high stakes. And something that I do to just test these models in different places because we are rolling out these tools in different hospitals around the world, I ask the same question which is a very high stakes question of how to treat diabetic ketoacidosis which is a diabetic crisis in a child. And I did this recently in Ethiopia in a language that’s not very well known, Afanaromo, and it responded to me, thou shalt not eat insulin on a Tuesday.

And I did share this advice with the, because I thought it was actually very good advice, you should not eat insulin any day actually. But I did share this advice. But it comes to something that’s really, really important. I’m stating the obvious. But it means that if you do not, because it’s obviously only trained on the Bible, right? That’s something that’s very available. That’s the one. book that is available in every single language in the world and so you have these biblical kind of terms but the Bible isn’t like very necessarily very accurate in medicine or other things but but depending on where you’re coming from but the thing is that you can’t rely on these models to make these decisions because they are inequitably inaccurate in the places that need it most so we know that they’ll be inaccurate but the point is that we actually have to if we collecting this kind of information we have to make an effort to collect it in the highest stakes environments and in those contexts so if you have use cases for collecting language it’s interesting to collect it in maybe like historical texts or to represent culture of course but I think something that has a much bigger urgency are the urgent questions these are high -stakes decisions that we are making and people will believe that the model performs well if it only speaks the language but they might get sense of security if we don’t really train it to be accurate in the questions that people are relying on these tools for the most.

And so this is why we actually have to, when we collect languages and when we are trying to test these tools in reality, we have to make sure that we represent those kinds of contexts. And that’s what we are doing. So I lead a lab called LIGHTS. It’s the Laboratory for Intelligent Global Health and Humanitarian Response Technology. So obviously I’m interested in these high -stakes environments and these cultures that are so underrepresented that they will never be represented with any kind of large commercial enterprise, right? No commercial entity has ever said there’s a great place to make money and it’s that war zone. Okay, unfortunately they have, I suppose. But the point is that people don’t want to represent that kind of place because it’s not in their interest.

And this means that it is so important for academia to play a role. We don’t just play a role because we’ve got expertise. We don’t just play a role because we have expertise. We play a role because we do something that commercial… entities cannot provide, it is, we are neutral and we create a neutral space for this kind of collection of data to represent the needs of people and also to make sure that we can test it in reality, right? This is why we can do open science, it’s because we don’t have like any money in the game to lose, right? And the most important thing to do is actually to see when we do represent these languages, not just to represent them and be happy about it, which is the first step, but to go the extra mile to actually test whether the languages are being represented as you expect them to be.

So some of like, like some of my patients for example, it might speak their language but does it speak their language in the way that they expect and do they follow the advice or don’t they? And this is a really important thing to test in these high -stakes environments. My patients will come to me in South Africa. In South Africa we speak 11 official languages, and in Kaga, a way of explaining certain things, it’s very different, and it gets translated into English in a strange way sometimes. And so one of my patients came to tell me, you know, I’ve got elephants running in my head, right? I know exactly how to respond because that’s my culture.

I’m South African. But what would an AI respond, right? And I have a pregnancy in my knee, right? I’m pregnant in my knee. That’s what the patient came to tell me. And actually it doesn’t come from a mistranslation. It comes from the way that people understand how their bodies work. And this is very, very cultural. What is the next most likely word after pregnancy in my knee, right? So it’s really important that we understand how it works when it’s in our body. And we understand how it works when it’s in our body. And we understand how it works when it’s in our body. and making sure that we get feedback from reality, this is what we’re trying to do.

So we have, starting with ICAIN, a flagship project that we made. It’s called MOVE. It stands for Massive Open Online Validation and Evaluation. And it’s about getting these real -world signals from real people in high -stakes decision -making processes, from our doctors, from the people on the ground in different countries around the world, and to get that information from how they are using any tool, because we are neutral. If any tool comes out, any new model, we can test it. And then we get how it works, and then when it breaks, we don’t just say, oh, this model is bad in the setting and this model is good. We really try to get that information and put it back into the model to continuously improve it.

And so learning from reality, learning from the real workflows of how people use models. And I think that’s important, to represent reality. And not just the language. but the reality that the language functions in. So last thing I’d like to say about this is this does cost a little bit more money and it’s not the traditional kind of way of working in science, and people don’t appreciate that implementation science is science. And it’s such a fantastic opportunity where we can actually do impact, like actually measure it, the impact of the models that we are making, we can measure it and feed that back into our models and really create impact driven models. And to run these trials, it’s ambitious, but we do need to start asking like different kinds of funding and being more ambitious, and I think academia does need to be more ambitious because we are representing something that’s actually very important these days, which is, and very rare, which is this neutrality.

When OpenAI updates it from 4 .5 to 5 or more, it’s a very, very good thing. And it’s a very, very good thing. And it’s a very, very good thing. And it’s a very, very good thing. And it’s a very, very good thing. And it’s a very, very good thing. And it’s a very, very good thing. And it’s a very, very good thing. And it’s a very, very good thing. And it’s a very, very good thing. And it’s a very, very good thing. And it’s a very, very good thing. And it’s a very, very good thing. And it’s a very, very good thing. And it’s a very, very good thing. point one, did they ask your commission?

No, right? Did they ask the doctors who had validated those models for their context? No. We need control. We need to know how these tools work in reality and we need to be able to control the tools and so sovereignty for me is control of tools and control of the environment and to understand how these models work in reality so that we

Thank you. Thank you so much. Thank you, honestly. Thank you, everyone. Thank you, everyone, for keeping the time and for making sure that we are actually creating the menu and controlling the menu to also steal Professor Benjo’s words and for contributing here and I think we will be more than happy to update you hopefully next year on our more important work. Thank you, everyone, for joining Collaborative Verge. Thank you, everyone, for coming. Thank you for coming, for staying with us, and on the speakers. Thank you. Thank you. Thank you.