 Hello, good morning, good morning, everyone. I'm going to tell you something about several things, give some background before I tell you about Open Source Program Office. Some background that maybe you know, and maybe some that you don't. So I'll start with telling you something about the scope and purpose of an operational arm of WHO. Then tell you something about what probably everybody here in the audience knows about. This is how the traditional disease surveillance works. So of course, we'll use a COVID example. And we'll identify some challenges and gaps in that. And then I'll tell you more about how we go about addressing these challenges and gaps and what is the strategic plan, a new strategic plan of the WHO to address gaps in the public health and health in general. And then instead of going into something that is closer to my heart because I'm an architect, I won't go into technical side of it. I'll mention a few technical bits and pieces, but I'll stick with Open Source Program Office and how Open Innovations and Open Source as a technology side of it addresses important points that we need to address. So let's start. You maybe know, and maybe you haven't seen this this way, but actually the operational arm of WHO, which is Health Emergency Program, actually has a mandate to strengthen global health security. That's in one sentence. What does that mean? That actually means that we are conducting and performing ongoing global surveillance of public health threats and hazards, including early detection, verification, assessment of risks to provide actionable intelligence for decision-making. So this is the scope. And that's a decision-making throughout emergency life cycle or health emergency life cycle from well, I say prediction in brackets. We are not there yet, but prevention, preparedness, response and recovery. So I guess you're familiar with these graphs and with this kind of simple statistics which are just counting. So we had this all over the news. When we talk about COVID, we usually talk about this information. This is the early days, still March. So the measures that we know now were not in place and we're trying to figure out what's going on. And we had the two states which have interesting, interestingly different information of this kind, South Korea and Italy. It usually started at the same time in both countries. And you can see the difference here in COVID cases counts. So the blue South Korea has significantly less counts after a couple of weeks in an outbreak where it had explosion in Italy. And it was all over the news. The same with deaths, actually. Not too many deaths in South Korea, but then count of deaths in Italy, skyrocketing exponential growth. But this is naive counting. Naive meaning that, I mean, you can guess, there's obviously many biases here. First, can we have any, derive any useful information only from this? Probably not, because actually we don't know. Maybe these differences are because of difference in counting. Maybe it's different because South Korea already had in place measures because after outbreak of MERS, which is not a coronavirus back in 2015, maybe it's a different virus. So this is where we are, couple of months after the outbreak. If we use only this information and you saw only this information and people communicating only about this information all over the place, deaths is even more interesting. Like if you look at the rates, South Korea would be 0.8%, which is around four to eight, it depends on the type of flu, four to eight higher than ordinary flu, where Italy is 6.2, which is 60 times higher. So then if you take this information, extrapolate this to United States. Population, you would have 15 million deaths in the United States if this rate of 6.2 is true. We now know after the fact that luckily it's isn't that true. So what is it? So what do normally public health intelligence analysts or public health detectives like to call them do is they then drill down in more counting. So if you stratify the same information by age, I think to me this is even more interesting now. Now you're seeing South Korea that the most prevalent or the biggest incidents is young people between 20 and 29. Not elderly people. In Italy, it's the opposite. So now you ask, now if you're an analyst, you need to figure out why. And now we know why. Here in this part of the world, at the beginning only people after 50 or people who have some comorbidities or people who have symptoms were tested. The others have not been tested. Where in South Korea they had tests and they had something else as well, which is actually more difficult to figure out and it's definitely not possible to figure out in time. And that is after MERS, their South Korean government basically allowed mobile network operators to share mobility information their users and also credit card operators as well. And even private companies could build APIs or apps for this kind of contract race. So they had information about young people. They send them SMSs suggesting to be tested. They knew that they are gathering, of course, clubs, restaurants, they're having their life. But you don't know that on time. So how can we react? What can we do? Can we do something proactively? No, we need to wait to see what happens and then continue having these graphs. So these are challenges. If you ask why don't we have it do we have some approach? Yes, we do. And we call it public health intelligence. And public health intelligence is really trying to, the main objective is to detect occurrences that might lead to a disease or any other health threat as early as possible. And this is not possible, unfortunately, from the formal sources like through public health institutes or even from hospitals. So what we do is actually we use publicly available sources, informal sources like from the web, web media, other type of media, news, blogs, social networks. And the idea is really to try to detect early, to verify, to be able to provide actual intelligence faster and save lives, ultimately, and economies. And we also know that the world is changing and the massive urbanization, people are connected even virtually and changing each other's behavior. So really, the pace of change is really high. And for example, take massive urbanization and we're reaching the nature where we haven't been before and we're getting in touch with animals, with plants, with different pathogens and microorganisms that we haven't been in touch before. So we need to take into account something that we call one health, which is connection and interaction between animal, human and environment health. And this is also not enough. Then we need to take care of any hazards. If there is a mass gathering, if there is a conflict like a war, you don't have a healthcare. So that can lead to health threats. If you have disasters, man-made disasters or natural disasters, the same. So it's a quite comprehensive scope, but we have no other way than to try to tackle this. So if this seems like, this is a new thing and it seems like something that should use, that is innovative and that would use, for example, technology, information, communication, technology, in all the ways it's true, but it's not a new discipline. I won't go into details for the sake of time, but even before this 1881, before second part of 19th century, there was a British epidemiologist, John Snow, who actually is very famous for mapping cholera cases in Soho District of London. And he managed to find a correlation between water pumps and the cholera cases. So this is what we're trying to do. Really, context is important. And if it's needs now, you would think of it immediately. You all wear masks and maybe you don't have COVID, but the context is what makes us understand things in one way or the other. Again, League of Nations had approached and at the end, World Health Organization, which is founded three years after you and replaced League of Nations. So people are going about addressing the problem of, and the complexity of understanding the mechanism of disease spread and trying to mitigate it, at least if not prevent. So our ambition is big. We, the only way to go about this in efficient ways to actually create, because we don't have it, a global network of data information, insights and knowledge and linked data, insights and knowledge from diverse information systems and data sets that we all know are built using different technologies. We have myriad of systems there and developing. It will always remain the same different technologies and people are using different technologies. I mean, if you look at one molecule name, you can have 300 names run the same molecule and things of that. So what is the solution? Can we harmonize the technologies and agree to use one technology? I think we don't need to discuss that. It's impossible. People are still trying to harmonize, which is interesting. Talking about harmonized data models, standard data models and things of that, but we know it doesn't work. So our approach is to let people use their technologies, actually apply, if you want, domain agnostic standard web technologies and apply worldwide web principle. Open world assumption. Anyone can say triple A principle. Anyone can say anything about any topic. So please use your systems, build your systems but if you can please develop this virtualization link data layer, put your eyes to your data types and let's make web of data finally as we have started planning 20 years ago, but we never did it. I mean, Googles and other social networks are having great success with that, but this is not open. It's closed. We decided to go that way knowing that this is problem mission impossible. It's really difficult, but we'll try to use the power of WHO and the power of WHO lies in that it's international organization which belongs to every single one of 195 member states which is practically the whole world. So we're not there yet. And this is one important part of what we need to do, but what we're doing, we're trying to do something. So yes, we do have some systems and some communities of practice that are trying to detect, to find a needle in the haystack and to detect an occurrence from a messy information, unreliable information, facts, opinions, disinformation, misinformation, rumors, conspiracy theories, all over, for example, web media, including social networks. I would say all over media in general. So we're using open source intelligence and people who are from the intelligence sector. We borrowed the terminology from the intelligence domain and from security domain. So open source intelligence is really trying to understand what is going on and create some actionable information out of public information from publicly available sources. The EUS is, and you can later check it if you have interest, there's a web link there with more details. The EUS is Epidemic Intelligence for Open Sources which is primarily a community of communities of practice. It's an initiative, global initiative where WTO is just coordinating the global community of practice. And of course they collaborate, One Health, all hazards, One Health approach. And the technology side of it is this system that is getting more and more complex. It's basically a web-based cloud application that in general consists of NLP Engine and the user portal. User portal is on the right-hand side. It's a news desk. Basically you can filter news according to your criteria and categories and labels that these news are labeled for. And you can filter them. You can do text search as well for more details. And NLP Engine is the one making that possible. So there we have the usual suspects, language detection rules based on classification, name and recognition and some non-usual suspects. For example, new stress worthiness. I mean, if you look at, I'll try to use this one if it works, yeah. If you look at this here, this peak is actually night between 30th and 31st December, 2019. So the same Engine, no change. We have, by this NLP Engine, we have 10-fold increase of news articles and other text items that are potentially relevant. Actually, it's mostly from China, Wuhan province, about five cases on unusual unexpected hospitalization cases that are hospitalized with unusual respiratory symptoms tested negatively for SARS virus. The rest you know, I don't need to tell you. So this platform actually is tapping into more than 13,000 web sources and 30,000 news feeds, 80 different languages covering. We have now more than 100 million text items. This is just Twitter accounts. So predefined Twitter accounts. We don't have social media processing yet. So this is what we do. This is what we can do with more or less success. But then, you know, the problem that we faced with unexpected problem was this. So now the whole system is much less useful. If useful at all. We have so many different news articles that you need to go through to try to find a needle in haystack. So is this a noise? And we knew that we have a problem with trustworthiness, with misinformation, disinformation and whatnot. So I'm giving you this an example of the need for open innovations and open source right now. So this is how we went about this problem. So we decided early on, no fact checkers. If there's somebody knowing more about fact checkers and you will know why. They're actually not fact checkers. They're biased as well. And it's known for decades. But that was not the main reason. Also the other reason was everybody's doing fact checking so we might well try something different. So we're thinking some of us who have some life experience reading news and we can't check facts when you read news. You don't have time. You know, I know that I'm better and better reading between lines and I'm much better than I was 25. So the hypothesis is that maybe we can have an algorithm that can learn and can predict trustworthiness. So binary classify trustworthiness or not from the writing style and tone. So the hypothesis is that in any given language, so in English, people are using the same writing style and tone when they want to, I mean, when they want to write something that's trustworthy or when they want to deceive you. And then if they want to deceive you, we are also concerned about the intent. And then we have a mess of classes like political and junk science and misinformation which is non-deliberate misleading and this information and whatnot. And they're overlapping if you think about it. So basically, I'll come back to the challenges in a second. So basically the experiment was simply to reuse Google Bird transformer based model. So basically selected five layers. It's five layer of feed forward neural network for those who know something about it. And of course, we applied transfer learning. We had some pre-labeled news open there in some GitHub repositories and then we also tried with crowdsourcing and failed because actually we were not able to pre-label big enough for Bird, for this transformer, for our use case, our pre-labeled training dataset was not good enough. And we don't know even if it was good enough. You know, taxonomy of this terminology doesn't exist. If you check Webster or Oxford dictionary, it's really difficult to, I mean, these terms are not there. I mean, we first called it credibility because my English is not first language. Then we called it reliability. I think trustworthiness is the right term. But you know, so even terminology, so without terminology, without semantics, we can't go far. We can experiment and they're all experiments, but it's shooting the dark. So what do we do in WHO? We are not IT company. We don't have million of people. And we also checked this idea and we discussed this. We have our collaborators and colleagues and friends from different places. So what we decided very early on, okay, we won't go and try to improve the algorithm. This will be open source because we engage with AWS or Amazon Machine Learning Lab from Oxford. We needed to agree with them on the type of license and then the DevOps has to be thought through properly because it has to be open source and community building we already did. It's a really multidisciplinary life social computer science. It's academia, McGill, Cambridge, Imperial College, London, University, Texas, Austin. I maybe forgot a few. Private sector, I mentioned the AWS or Amazon. And supranational organizations. I've tried to be exact here. EU is not international, supranational. So we have partnership. We joined research center. It's more than collaboration I would say. And I will just skip through. I already mentioned we have a number of initiatives and we established a term collaborative intelligence and created a new WTOB hub for pandemic epidemic intelligence based in Berlin who is serious about the objectives that I told you. And one piece of the puzzle is open source program office. Before I just go into quick details and I'm now over time. So I hope I can spend a few minutes. Our reality check when we start engaging in these projects that I mentioned is that we have a problem, we have a challenge with multi-display collaboration. I mean, it's romantic to say different disciplines sit together and then work together, but the reality is different. You know, when you put engineers like me and doctors like my colleagues who are epidemiologists together, you don't know whose ego is bigger. So there is a challenge there. And then private public sector, we are having collaboration. That's why open source program office, among other things. It was inspired by our attempts to work with colleagues who have expertise. But then, you know, we had a challenge to find a sweet spot between global goods. So that you cannot own, but you need to be profitable. Of course, if you're private sector, otherwise it doesn't make any sense. So how do we go about this is a challenge. Competence, learning new skills requires substantial investment of time. That's why I mentioned that project. You can imagine, I mean, it requires real machine learning. You can just use scikit or some library and then button click and, you know, off you go. You need really to think about the model and try to experiment and do real, real modeling there. So, you know, it's not easy to find these people. Steep learning curve, of course, when you engage people. And if you go into knowledge modeling and semantic web, it's even steeper now. So there's a deficit of people there and it's also an epidemiological level. I mean, these concepts are completely new for epidemiologists. I mean, they used to use R and this exploratory statistics or maybe some inferential statistics, some linear regression type. And that's all, they stop there. So now introducing these concepts to them, including using, you know, network analysis and graph theories in the analytic workflows is really difficult to discuss with them and to propose. So there is a problem. How do we address this problem? One piece of the puzzle, engage everyone, citizens, communities, anyone who is interested. This is global good, this is health, it concerns all of us. And in, we did the homework before we got the green light to establish this organization and the homework result was this. The open source program office should have been benevolent leader, it should provide advisory role to any project who is actually leading and they know what they wanna do. So if they need legal advice, of course, licenses come to mind, procurement, technical capacity building, community building, they should provide advice, maybe write some policies and guidelines and simply be that competent center that can help individual projects to accelerate or make a right decision. And I'll skip this one, but you know, there is a capability maturity model there and we had internal embracement of inner source, which is open source within boundaries of WHO, so sharing between WHO teams, that was not possible before. And this was embraced by chief technology office, but community building capacity management maturity that is required for real open source, they said, no, no guys, we don't know how to work with people outside. Nobody worked with outside in WHO before. And yes, that was a problem. We needed to go rogue because we didn't have a platform to share our code, that algorithm for example, we had SVN. So how can we do it? Github came proposing to work with us and offers us a pro bono help when COVID started. We embraced this. Github is now, and this is Github World Health Organization organization. You can check it there, you can see a mess right now. You can see 70 or I think public repositories. I think none has any license information there. It should happen soon because we're establishing open source program office, but yeah, this is our platform. And I'll end with this, in the spirit of open source. So it's an open question. It's really multi-intelligence is, and public health intelligence therefore, or health intelligence, if you go down to individuals, diagnostics, IE diagnostics, it requires multi-disciplinary knowledge. So anybody's invited. And we are really serious about creating this collaborative environment. We have organization already. It's politically supported. It's in Berlin. All this is currently pretty strong from these political points of view and pretty much at the beginning from the technical point of view. So open invitation, if you want to collaborate with us, I mentioned profile is a listed few here. But also if you want to join with us in this paradigm shift and creating semantic web. So if you want to put link data API on top of your software that you build to enable to expose or put data on the web, give us a shout because you might have some contextual information that might lead us to early detection or early understanding of some disease pattern. I'll end with this and here's some details if you want to contact me and I can put you in touch with others. And thank you very much for your patience.