 We are back. Our next keynote speaker is my friend José Luis Flores. José Luis, how are you? How are you doing? We can hear you. How are you doing? Everything fine? Everything fine. Very excited to be here and it's a pleasure. Even more after Pedro Domingo is starting this world of machine learning and so on. It's great to be here. Well, you are a statue. You are a statue. Your presentation is probably one of the most expected because you are talking about epidemiology, about tracing contacts, about I think probably the staff that is more interesting nowadays for someone that is dealing with artificial intelligence. So I don't want to take your time. So it's your turn now. You're 30 minutes to talk about this, but something important for our audience. You can make questions. We are insisting all the time on this. So make questions in Spanish, in English. Spanish is fine. You can ask questions in Spanish or sometimes in English. We will make our best to translate them. So English is fine. Spanish is fine too. So we are waiting for your questions. José Luis, thank you. Your time. Thank you. Well, thank you for this glad invitation. As I said, I'm really very excited to be here with you in this event to talk about something that I think is interesting because of this difficult situation we are facing. Well, as you see, I've chosen to start my presentation with this image. It can be a cave painting 13,000 years ago. I think it's a good way to start in the sense that it's quite extraordinary and also disturbing. Now it represents probably the ancestral dream of trying to survive. It seems even that they are even reaching us from the past with these hands moving. And I think it's interesting to stay while thinking in the vulnerability because of the nature of the disease of these people in the past. They only could explain this because of kind of living punishment. And we now, seeing in this ancient of this ancestors now with the mindset of only a superior piece in the sense that we know much more about the world, what is the nature, the disease and so on. But things like the one that is right happening now makes us think that we are not the owners of the nature and we are only a part of it. We are facing a very big challenge that is quite clear. Probably all the generations have to face their own challenges. But in this case, I think we are facing two actually. One is the tremendous change that we are living because of the technology and all the aspects of our life, our society and our work and so. I know this new thing, the thing of the epidemic that has had a tremendous impact for the life of everyone. In December 2019, we start having some quite diffused information coming from China, speaking about respiratory disease. And probably it was not so alarming because some was something that has happened in the in the past with the MERS, with the SARS-4. And normally this outbreaks were quite limited weeks, months, but quite local in certain areas of the world in certain regions. And well, at some point we started receiving information about not only about the people who were ill, but also about deaths. The number of deaths was increasing following an exponential pattern that's quite normal in these kind of situations. And what made the process even more terrible for us in our Occidental mindset was that we start to have these problems in our own home. In Spain, in Italy, in different countries of Europe, in Latin America, the US. And I think it was the first time in our recent history when we realized that something, a threat, could harm ourselves. Because in the past, when we talk about terrorist attacks, normally these attacks were in other countries or were quite limited in time. Or when we thought, when we think in the tsunami was in very far regions. This is the very first time we feel that we are vulnerable. And this is something that makes us so similar to these ancient people in the past. Well, my intention during this presentation is speaking a little bit about data science, about epidemiology. And why not, about my own experience as a data scientist, in terms of how we manage this crisis. I had the opportunity to lead the team, to work with people, freedom, people at means rate, in collaboration with the Spanish government to put some technological solutions on the table. I think it's been a very interesting journey, very challenging one. What we are living is a big drama and I want to share with you my thoughts. And I want to start speaking about this epidemiological quantitative analysis. In the last century, in the 20s of the last century, was the first time we started to apply mathematics to understand the behavior of such a complicated model as the evolution and the spread of a disease. In 1920, Lofka and Volterra started with a model that was able to measure and to understand the equilibrium and the balance between praise and predictors in an ecosystem. And based on these results, we started in the early 20s with the SIR model. The SIR model is a model about the susceptible guys, guys who can be infected, infected ones, and they recover it. And it's based on a set of differential equations. Differential equations are useful because they are less to analyze how an event or behavior, something that has happened, is varying on the time. And this is what we try to do with the epidemiological models. We want to understand when something that is linear, at the beginning normally it doesn't seem to be dangerous, when it starts to have higher volatility and this volatility starts to have a chaotic behavior. This is what we try to do, chaotic behavior and then an exponential work. Well, the good thing about the differential equations and the SIR model to explain or to understand the quantitative aspects of a disease, of a pandemic in this case, is they are really very flexible and very adaptable. Because the solution of this system of equations is a set of functions, is a set of family functions called exponential curves based on several parameters. The parameters can be different. In some cases, is the number of contacts, is the number of infected individuals, the window of infection, the duration of immunity. So you can establish different conditions and basically the idea is once you have a contact, you measure the probability of this contact to generate a new infected, a new positive. And this has to do with the information you have about how this infection happens. But at the end, the solution is always a set of exponential functions with different parameters as the parameters I mentioned previously. The advantage of this is very versatile, very adaptable, and that's also the problem, sells for the trap. I have seen during the last months the emergency of many, many, many different approaches based on differential equations created by research centers, by engineers, by data scientists. I even received some letters, some emails of people saying, I have a model that's very well, what is happening right now. In all the cases, what normally happens is people consider that the certification of the goodness of the model is the fact that the model adjusts very well and matches very well the historical data. But of course, data are historical. It's something that has happened. And this set of equations is very, very flexible and very adaptable. And it's a consequence. It is something that you can get. You can define the right parameters to tell the story. So that's really very, very interesting. But what is the different thing is if you're able to make predictions with this model, and this is something that is not possible in general. Sometimes you make a prediction and it's okay, but if you are right once, it's quite difficult to write twice. And the reason is the reality is much more complicated than the model we have created. A model is a simplification of the reality. We need to simplify the reality, because if not, we cannot compute and we cannot make calculations. As a consequence of that, we try to go to a formulation that is much simpler than the observation that the reality. We also have a lack of data that allows us to have all the information we require. So that's the point. I think now, as human beings, we are obsessed with two different things. One is telling stories, and the other one is accounting. Accounting how many people, how many cows, how many infected guys. These are the two obsessions of the people. And this is also the obsession of the data scientists. The good data scientist uses data to calculate, to count. And with these numbers, with this information, it's able to tell a story. But sometimes happens just the opposite. We have a story to tell, and we don't talk to the data until the data is saved. It's a story we want to share. And this is the danger with this kind of model. It's something that happens with some of the data scientists working in this space right now, because you have the sensation of, well, I have the power to control what's happening. It's something that's happening probably. So more roughly with our politicians. Well, the top-down approach has these problems. It's difficult to make predictions in the short term. It's impossible to make predictions in the medium term and in the long term. It's a chaotic phenomenon. It's the same problem that we face when we try to make a forecast of the weather for the next month. It's impossible. You can make a good prediction probably for the next two days, three days, but probably no more than that. Here happens something very similar. So this is one of the limitations we are facing. And we'll be talking about limitations in the first part of the presentation, and then we'll try to find some solutions probably not to solve this problem of chaos in the phenomenon we want to believe, because it's something unavoidable, but in terms of how we can use data to take the decisions. There are other types of approaches when trying to understand that we can tell you in this kind of models. This is a reference. This is a good example, a graphical example of this. It's quite good to understand how this differential equation models we're using work. And when you see this, you see, well, that's fantastic, because you have the exponential roll versus linear, then this phase of exponential roll. Then you have a plateau and then all it goes down. Yeah, but the thing that what you have here is a Bronian movement, something that is quite random. It doesn't represent the reality of a social behavior. So we have other methods to try to solve this. And in the last years, and it has been great to have before Pedro Domingo, because he had a great job in this area, we also use social network analysis to try to understand this phenomenon of influence. How when you have a group of people infected, how this infection is going to be spread on a population according with the social relationship. It is very important. We have now talked about particles with a Bronian movement and then there are interactions following some low, but there is a social relationship among the people and we want to, let's say, try to understand and define what kind of relationship are they. Well, in the network model, basically we had a node. Its node is a person, it's an individual. You have some of them who are infected, others are not infected, they will win. Then you have a probability to infect. And here, basically you have two different approaches. One is the classical epidemiological approach. In the classical approach, I have a probability to infect you. The other guy has a probability to infect you. And it's like if we were using a random event to say, I can infect you or not. And then you apply the same event. You're launching a coin or something like that. If one thing is infected, if the other is not infected. So this is the classical approach. But the classical approach doesn't take into account the characteristic of its node, how this person is. What's the kind of relationship among the people? We are very close friends. We are acquaintances. We are relatives. We know very well because we have been in the same place a couple of times. Or it's the first time I've seen this guy. It's very difficult. Sorry, very different in one case and the other. And you don't have information about the context of the interaction. We are together in a party, in the charts, in the school, the office, in the vass. So this is very valuable information to being able to model what is happening behind. So basically what we're trying to know is once I have a set of infected people, these three guys, you say K guys, P1, V2, VK, in the paragraph, what's the number of people who is going to become infected after a time T? Sometimes your objective is to maximize this number because you want to sell something and you're using a viral marketing strategy. Or in this case, you try to say what happened if the majority of the infection is happening at home or is happening in the shop or is happening in the school. So it gives you not only the capability to make a prediction in terms of numbers, but also to make a prediction in terms of where the content is happening. This is something qualitative that is tremendously relevant to make predictions. The problem of this, the main problem of this is data. You require a huge amount of data. You have to create a network of sensors to gather all this information. But we have a problem with this approach of social networks and here we have a problem again. I repeat the story of the data scientists putting the numbers before the story, putting the story before the numbers. That's very important because what happens is this. In order to understand and to measure the future impact of a set of people who has the disease in the moment t equals zero. What is one to have in the moment t? I require to know a function of influence, an influence function. An influence function that means that when I have a group of people with the disease infected, what is the number of people that I will infect in the future in this time t? It's a response function. And as mathematicians, we ask for two properties to this function. The first property is that it has to be a monotone increasing function. That means that the number of people infected of a group A is bigger than the ones infected by a group B. Where a group B is part of the group A. It seems quite natural. You say, well, I have more people infected. The group A is bigger than the group B. The group B is included in the A. The effect in terms of new infections is going to be smaller. It seems natural. The second one is called submodularity. It's a property that is very important in mathematics and especially is very important when we talk about optimization, maximizing or the opposite. There was a problem. And here when we talk about submodularity, we said this thing. We said that the marginal impact in terms of contagious, of an individual X, is decreasing according to the size of the group where this individual is, is increasing. It's called a law of diminishing returns. And here we have a problem. And the problem is when you are talking about, for instance, a physical interaction. An example of physical interaction is you have a grid with ferromagnetic particles with different spins. And according with the magnetic field, they can modify when one particle can modify the spin of the neighbors. And then it's something that happens yes or yes. It happens the same way. It's deterministic. And this way it's true that this submodularity is happening. But when you are talking about something that is social, submodularity has been shown. And we have some research on this area talking about submodularity in the spread of behaviors in a telco operator, for instance. And submodularity doesn't, doesn't happen. It doesn't happen at a local, at a local level or it doesn't happen always. We use submodularity from a mathematical perspective because there is a very interesting theorem that says that if submodularity happens in the influence function, then we can use a very simple optimization algorithm in order to get a very good result, a very great result in terms of the number of future contiguous that this group of A people is going to, is going to make in the time t. But with submodularity doesn't happen. We don't know anything. Everything can happen. Everything is chaotic. And this is what happens in a social behavior. What's our surprise? Our surprise is when a disease is a pandemic disease, the behavior of the disease is not an epidemiological behavior, it's a social behavior. Because what is more important to determine the impact of the disease in terms of new contiguous, in terms of close context, in terms of what is the context of this new context is your social behavior. So what that means? That means that we cannot do anything because we lack of instruments. Now, it means two things. First, predictions are really very hard. It doesn't matter the kind of approach. They're really very hard. I would say that predictions in the middle of the day are not possible. But if we can gather data about the type of relationship, type of context of the contact, we can give you information, very rich information about what's the probable impact of this behavior in the future of the disease, in the future of the pandemic. But again, I repeat, this requires a lot of data. And requires a network of sensors. That's one of the main aspects or concepts I want to share. A network of sensors. A network of sensors about the behavior of the citizens of the individuals. This is not enough to have isolated cases. There is a case and there is a paper that says that there were a wedding, there was a wedding with 555 attendees, 200 infections, seven deaths in level two and level three relationship. It's great to have this. Because it gives you a clue about what is happening or the kind of variables you require to gather. But at the end, what you need is a systematic, a systematic approach to gather all this information. Well, based on this, we started an initiative in March this year. And we work in an application for our citizens called Calitmonitor. Calitmonitor was what is called a geotracking solution. Basically, what this application does is this. First, you could download the application. It was anonymous. You don't have to enter any kind of personal data about who you are. Then you add information about your symptoms, about what's your state at the moment when you introduce the app and you can update this information in time. And what's quite interesting is with this application, we can follow its citizen using a geotracking, using GPS location. We do the same with all the people who have the application. And this allows us to create heat maps of people in different places. And according with the symptomatology that we have from the people, or even if we have positive cases, confirmed positive cases, we could create a metric. And I think it's a very interesting metric for the citizens. It's a metric of the exposure to the virus. Your exposure is low. Your exposure is high. It depends on what you're doing. If you're staying at home and you're staying at home since two or three, four weeks, your exposure is going to be low. If you are walking your pet, your exposure is going to be a little bit higher. You're going into a place where there are a lot of people and some of these people have a high risk and your exposure is going to increase. So it gave a way to measure the degree of exposure of this person. And having to account that we knew what was the exposure, not only of a device or person, but also the exposure of a place. We were able to say, well, if you have to go for shopping in the next hour, two hours, this is the best time to go. This is the best time to walk your dog, and so on and so on. So it was a way to give something, something valuable, some information to the end user. Well, the thing is, it happens the same in the running apps, you know? You have information about your accumulated exposure. You have information about your instant exposure. You have information about the number of people that has been close to you in the last hours. The number of people has been close to you with a high probability. We defer the sending of this information to avoid any risk to know information while third practice. But this was the idea. And the idea was trying to understand where the close contours happened and where the positive cases were, was the correlation between behavior and activity and the positive cases. This was the main idea of this solution. We have, it would be more than 200,000 points of interest, schools, transportation networks, charge offices, supermarkets and so on. The first thing, probably some of you, and I think it's also a consequence of the noise in our society related to this kind of noise in the last months. Probably some of you say, well, I don't want an application being able to track and to know where I am. I don't trust in this kind of application, even if the government is behind of that, or sometimes because of that, the people don't trust so much. Well, the thing is, and I want to start with the ethical dilemma here. We face an important dilemma putting as confrontational two different concepts. One was privacy, the other was freedom. And basically this is what happened. People said, no, all the tracking apps are forbidden. In all Europe, in all the world, the geolocation apps were forbidden. I think the debate was not very correct. I think to some extent, I'd say the debate was very polarized in two different groups, the black and the white group, without any kind of gray. The thing is, first, you don't need personal information. Second, we don't need information about where is your exact location at each time. Third, the computation can happen in your own device. So the only information we require is profiling or the type of activity behind of you. And we can give all the transparency to these kind of apps. Why we think this kind of app shouldn't have been discarded and where, not only in Spain but in all the international community. Because we have mathematical models and technical solutions to solve all the privacy statements and all the privacy doubts that we can face. One is using a projection of the dimensional GPS information in a multi-dimensional space and higher dimensional space, keeping the property of closeness, of being far or being away in the new projection. And second, because we can control what is happening in your device and what is happening in a central server. And with these two elements, we are able to control the privacy of the entire process. What is the advantage of this? The advantage is mainly that we can understand what is happening, what is the impact of certain type of closing counter. We can answer the question of what we have to do with the restaurants, with the schools, with the hospitals. What are the best practices? And so this is something that we don't have. Now we are blind. And we are blind because in this very polarized debate what happened was something like this. This is Peter Field. This is the CEO of Palantir. He is a very well-known investor in Silicon Valley. And he said something that was extremely provocative, but probably true. At least in the mindsets of somebody. He said, quick, just libertarian and AI is communist. So what he says is the kind of polarization we are living also with technology. There was not the dilemma based on science, based on technology. It was based on feelings, basically on opinions, but without any kind of solid foundation. With our carrying mastering of the technology, we can do almost everything. We can keep the information as private as the society could require, but we don't have to renounce to have good information about the disease. My own question is what would happen if the disease would be worse? And instead of having two or three percent of that, we had, I don't know, 30%, 40%. That's just my first question. So the solution, and I repeat again, not only in Spain, but in all the countries, was a contact tracing solution, contact tracing apps. And it's great if we have the critical mass that we require to have these close contacts happen, between people who are infected and people who can be infected, because it has to stay a long time, have to stay a long time with someone infected. It's a contact tracing. I think anyone knows how this works. We are using here in Spain the DP3T protocol, created by a group of university research centers, the University of Lausanne, many others. It's great because it preserves the privacy. All happens in the mobile telephone. All happens with absolute privacy. It's absolutely great. The effort we have done to create this app for the Spanish government has been great. We've been in a rush all the time, because it's really very high demanding. I think it's also a milestone for the Spanish government to have launched an app that can release open source. I think the openness is one of the most important things in this kind of technologies. And the technology is great. When you deploy a positive, you are sending the information about your keys to all the people who were quite close to you during the last days. If you stay more than 15 minutes, less than two meters from you, then you're going to receive an alert. The problem is, you have to download the app, you have to have the app open, working. And of course, you have to add your positive key, your code. And that makes it not be very intense in terms of use. In the past, three or four months ago, we have said that you require 60% of the population using the app. Now, we think that probably with 20% is enough, but that is really very, very challenging. But from an epidemiology point of view, the problem of the contact tracing app is not the technology itself. It's the fact that different states, Germany, France, the Kingdom, and some others in Europe and the States have their own ideas about how to deploy this application. And there were also two models, the centralized model and the decentralized model. In the centralized model, the information about the close encounters was in the central server and the authorities were able to see what was happening. Now, we have longer chains, so probably something bad is going to happen very quickly because we have a long chance of transmission, for instance. But the thing is, there is the other perspective that is the decentralized one. In this decentralized one, all this information is stored in your own device. All this local network is stored in your own device. We don't have a central repository with all the network understanding all the behavior, what is happening and so on. And this is an important limitation from an epidemiology point of view. It was something that was this way because of the strong position of Google and Apple. At the end of the day, Google and Apple, they were the owners of the operating system and the majority of the mobile telephones. And as a consequence, they said, well, this is the way this is going to work. For us, the privacy of our plans is a top priority. And this is what we are going to do. There were some response, Germany, France, and Kingdom, but all of them, they had to say, well, yeah, we don't have any other option, but doing what would you say? Because again, the power is the power of the corporations. And here, also from an ethical point of view, it's something that we should think about because there is a very important lack of confidence in the states. When the states is asking for you to download an app, people say, well, what's the kind of information my government wants from me? I don't trust in my government. You don't trust in your government that you're trusting in big corporations in a foreign country, and you don't ask questions to this, to this, to these corporations. And this is what happened. All the people said, oh, we want the code. The code has to be open. We have to audit the code. It's great. I firmly believe in openness, but the thing is, why not we ask him for the same transparency with the big firms, with the big players, with the Googles, with the apples of the world? That's an important question. As you see, contact tracing is happening almost in all the countries in the world. We don't have now an information but if it's being useful or not to fight against the COVIDs, the only thing I'd say is, I think there were much better solutions from an epidemiology point of view. And even if you have to work in a limited population, even if you have to give some kind of reward, I don't know. But having this information is so extremely important that we should do something to have this information about how the spread, how the chance of transmission, how is the context of the closing context when transmission and contagious happen. This is very important. But I don't want to be narrative. I think we have evolved a lot during the last months, talking even about the governments and the relationship of the governments with the civil society. And of course, speaking also about the civil society itself, there is a movement called the volunteering technical unit that is happening in different countries around the world. This is a movement that started in New York at the beginning of the pandemic. And it's very interesting because it's also happening in Spain. And here, engineers, mathematics, physicists, and so we are collaborating to create solutions for the problem. Solutions like the CAD Air Cap Wader, this is a very nice initiative that allows us to parametrize and set up the airport fires in order to reduce the probability of COVID to stay in the air. You know the problem of the air results and these kinds of things. You have apps for small commerce trying to help them to survive because it's very difficult for them to survive with a single website. They are proposing improvements in the COVID, to contemplating apps we are using in the different countries in order to improve the results and the performance and to measure in a better way this blue to low energy app monitor that allows us to measure what is happening behind. They are fighting very hard for open information and I think probably it's one of the best things that have happened during this pandemic. People trying to put their talent at the serve of the society trying to find new solutions and new ways to do things. And finally, what are the lessons for the future? I think the first lesson is important is openness, transparency and flexibility. When we as a society work together when the government is able to trust in the experts in people working in research centers in companies and so I put in their talent and their best intention to give solutions I think it's great. But this is not something that can happen in a very chaotic way. You need a government for that and I think a government in terms of the method to be able to create with interdisciplinary groups. This interdisciplinary collaboration requires a space and a way to be done. Third, it's important to avoid false dilemmas. Privacy against freedom against security. It's not a true dilemma. It's being a dilemma that has been fooled by politicians by citizens with not very technical background and as as experts in data as data scientists and so I think we have to help people understand that technology allows us to make many, many different things and we can improve a lot in the kind of information and the kind of answers we can give with the good data. That's the fourth point. If the analogy requires data I read one thing during the pandemic and probably was the worst thing I read was an epidemiologist saying that they don't require data. Even he spoke about the telepidemiology saying that I think this is not true. I'm on the opposite position. We require data and the data we have are not very good. Sometimes even counting the number of that is not easy. The criterion used is different in one region and the other even with sometimes it's because of political interest in some other case because the criteria depends but the quality of the data even very, very basic data has been really very, very poor and it's something that we have to improve a lot. Because if we compare the last slide we compare what happened here 100 years ago in the left you see the Spanish fuel in the 1918 and in the right you see the Fema Palace in 2020. Yes, we've improved a lot if you compare with our ancestors in the caves we probably have learned a lot but we could have made this much better and this is my my last thought. As a data scientist I think there are long, long space to improve and to make things better. So that's all from my side. Thank you and take care out there. Thank you. Thank you, Jose Luis. Let me make a couple of questions. The first one you talk in your presentation about different reactions in different countries. I'd like to know your opinion about local reaction here in Spain and how the Spanish government has used these kind of tools including apps by way to react to this crisis. In your opinion is our app good enough? It could be better. For the next time, there probably will be a next time we should change in the years we have made of data. We are asking our app rather COVID is the same kind of app you can find in France and Germany is the same basically because you have the same protocol so basically you are talking about the same app it's like a clone from this point of view. I want to say that in the development of this app using the DP3T protocol we found that parts of this protocol were not very mature and we have played a very active role to create new code, new libraries and so to make this more robust. It's something that here in Spain we have had a good contribution but it is the same in other countries. The problem is not the Spanish deployment of the app I think the problem is the app itself the big question is if content tracing is going to be useful but the same problem is in Spain or it's in other countries we don't have less usage of the app than if you compare with other countries like Germany. Germany is a very good reference. It's something that we have to check probably in the next month there are going to be more downloads of the app, there are going to more positive cases Madrid and Catalonia have introduced their healthy systems in the app quite recently so let's see, I have my reservation on that and that's the reason why, following with my answer I propose a more aggressive approach. The last one because you were talking about this dilemma, this false dilemma that it's trying to debate to choose between freedom or privacy or between health and transparency. Do you think that this debate has changed with this pandemic after all this, after Covid probably the public debate will be different, that our opinion our feelings about the use of our data will change after all this Unfortunately I don't think so I think there is a a politic underlying behavior in the sense that it's quite if you belong to this party if your ideas belong to this party then you have to be okay with this or if not it's the opposite so when you have this kind of black or white approach it's very difficult to solve something because when a problem is complicated the solution normal is in the middle it's a combination of grades and I think the population in Spain is quite polarized and it's still polarized The answers are as usual the answers are more complex than this white or black debate so thank you for being with us and sharing your thoughts Thank you Bye bye Thank you