 Welcome to this new episode of the robustly beneficial podcast so today we're going to discuss an important topic on social media which is misinformation in general and in particular in the case for vaccination there's regular debates on the internet between pro and anti-vaccination and there's a recent paper in nature that analyzes this debate and in particular how different communities interact on this debate depending on whether they are pro anti or undecided and this is a very like I think it's a very important topic especially with the COVID situation where like it's been argued by many biologists epidemiologists medical doctors that the main way maybe the only way out of this pandemic so that we can really have a normal life again is if there's a vaccine or a solid vaccine a reliable vaccine that's developed but also it needs to be implemented it needs to be accepted by the general public because we want to reach herd immunity probably at least 60% if not a lot more of the population should be vaccinated and this means that if there are like 30 or 40% of people who are skeptical and don't want to vaccinate themselves then we may have an ongoing endemic as it's called meaning that the coronavirus may become a seasonal or may come again and again every year in which case it would lead up to potentially hundreds of thousands if not millions of deaths per year. Yeah so as they say providing good information concerning vaccines it's a it's a question of a it's a life of death situation and and this is what this this paper is quite striking by the the numbers it tells us concerning a vaccine misinformation that are extremely high and quite surprising so what they did in the study consisted in analyzing Facebook pages so how many members of the each page on Facebook has and how do they link towards one another and for this they classified the pages into three categories the anti-vaxxers the provaxxers so pages for provax vaccination are pages like the Gates Foundation, the WHO page and also the majority of the pages were in the category undecided where and decided where groups that have talked about vaccines but without taking any clear side on on their stance towards vaccines so for this decide for example parents association for for schools or similar things and the first thing we see that could be the reason why it's quite surprising to to to see these numbers of the of the scale of misinformation is that these pages are separated in the in two very distinct clusters there is the cluster of provaxination content the cluster of anti-vaccination content and if you are in the in one one of these two two clusters you most likely nearly never see content from the other cluster so this is the concept known as a filter bubbles pushed by by recommender systems and so finally to say the the numbers out of all these pages they they found that the the pages concerning anti-vaccination that they had contained us about five five million members and the pages for pro-vaccination contained about seven million million members so but what I found surprising is that it was approximately the same amount on a on both sides so nearly 50-50 speed even though the the scientific consensus on vaccine is there there is no 50-50 speed on that it's a vast vast majority of scientists except one or two exceptions that that consider vaccine to be extremely useful yeah this case are pretty huge like it's a study of that involves like clusters that add up to hundreds of millions of of individuals and what they did in the study is they particularly studied the interactions between different clusters so cluster is like a facebook page and what they call an interaction or link is when a given facebook page shares a link towards another facebook page and this allows to see how the different clusters interact with one another and sort of the question we may ask is well what do most undecided pages link to and how this like how does this cluster that can make can imagine you have a graph now like some some clusters we link to some clusters that link to other clusters and you have this graph that is connected in a certain way and one thing they did is to try to to visualize this graph by a very simple model where essentially each cluster is like naturally repaired to one another except for links that pull them together and like you can imagine that this dynamic the simulated dynamic creates like large clusters of clusters and what they observed is that there were essentially like three equal big clusters of clusters created this way two of them are somewhat with the pro vaccination like more dominated by pro vaccination than anti-vaccination cluster facebook pages but they are also smaller clusters of clusters and there's one bigger cluster of clusters where you have most of the facebook pages and this one is actually dominated it's like well infiltrated that's sort of like the visual impression you can get by looking at the image by the anti-vaccination facebook pages so it's like the anti-vaccination pages even though there are fewer of them and even though each of them is also smaller in size like they tend to be smaller whereas like the pro vaccination pages tend to be very large well these anti-vaccination pages are somewhat better connected to undecided nodes and this may explain why currently these anti-vaccination groups are growing faster than the pro vaccination groups yeah just to correct on what you just said actually the anti-vaccination pages contain less members so slightly less members but they are more pages so they are more numerous pages talking about anti-vaccination than pro vaccination and this was quite a big difference like there is between two and three times more anti-vaccination pages than pro vaccination yeah yes so it's quite interesting like in a sense like there's a non-going war or battle maybe war is too strong of a war but there's an ongoing battle on the internet for on this debate and you can really almost see these as territories that the different sides are trying to conquer and what's sort of nice with this paper it's that it kind of gives us a map of the ongoing battle and it allows you better understand not necessarily the intentional strategies but at least the effective strategies like indeed what's going on in this war and sort of the the insight that this gives us is that pro vaccination are probably doing a very good job at convincing the people that they reach out to but they don't reach out to most people they like they are kind of isolated whereas the anti-vaccination pages seem to be what better at infiltrating different other web pages so another thing that they did in the paper is that well they try to model the dynamics of the of this battle in a sense and so they fit some some models so the model is quite sophisticated it's quite hard to explain or just to understand because again we understand it at all but it's essentially a model that tries to fit with the current evolution of the dynamics of the battle and then they did projections for the future based on the model so it's sort of like saying like we have this well the the current situation seems to be quite explainable by or by a certain dynamic and then they projected this dynamic for the future and what they observed what the model predicted is very concerning since it raises the possibility that within a decade anti-vaccination views may become predominant which would be very very bad if this happens arguing brings us to maybe one experience we have I don't know for later so whenever I presented our book projects or when I give a talk about why we should care about robust recommender systems and why we should care about like robust machine learning for commander systems and I motivate that by vaccine like up to up to last year people tell me like but this is not a big threat like like this is this is not a big deal but arguably it's it's growing like its potential of growth is scary and we are seeing it now with with with covid because if it keeps growing like now now there are reports that the anti-vaccine movement is there is an anti-vaccine movement against a covid vaccine this is growing faster than researchers are are getting closer to having a covid vaccine so by the time we would have a covid vaccine you should expect that there would be already like communities anti-vaccine communities would be ready they're like would already have their talking points and their influence on social media would have been established yeah yeah so this is like the stakes are high I think and especially the covid situation really shows this but more generally well health public health is especially vaccines have saved millions probably hundreds of millions of lives in the last decade or something this so yeah we're talking about big big numbers and one of the key feature of vaccination is herd immunity so herd immunity is the the fact that if sufficiently sufficiently large proportion of the population is vaccinated the intact tree protects all of the the population and the reason for this has to do with the spread of of of of a disease and the famous you've probably heard of the famous famous reproduction number by now this RRT and and this R number if it's smaller than one then the the pandemic just disappears by itself like it's exponentially decaying and so you save everybody well not everybody because there's few people die because of this but you like 99.99% of the population is saved but if this R number gets larger than one and if and this happens if there are not sufficiently many people vaccinated then you have an exponential growth of the pandemic and it can reach like 10 percent of the population and if the virus is is quite deadly this can represent effectively well millions if not hundreds of millions of of deaths so I think that the stakes are high and it's not only about public health as we see in the covid situation like just trying to fight a pandemic puts a lot of stress on different other parts of the society like for instance economics so like unemployment has skyrocketed in the US and there's a lot of concerns about people not being able to afford a living in many countries in the world especially developing countries but also it creates stress on and political tensions and political decisions and and there can be a lot of good coming out of this maybe this will question some some bad ways of doing things but it can also create bad situation like essentially a bad economy can cause something very bad in terms of politics and we have a historical examples of this so arguably the stakes here are huge like I think we're really talking about the future of mankind here in many regards so one thing that we're interested in doing is somehow knowing how to fight this this this sort of misinformation and spread the correct information so in the paper they they mentioned that doing the mapping like they did and they they also mentioned a few metrics like measuring the centrality of of a cluster so of a facebook page somehow it says how much that facebook page can influence other page or on itself so correctly mapping the battlefield kind of helped you change the way you communicate so tell you which page to target to better influence the this this battle that's happening over misinformation and yeah and also like yeah so so there are different interesting insights like another interesting insight from the paper which is more qualitative but I think it makes sense is the idea that the anti-vaccination movement relies on many different narratives and suffices that one of these narratives even though the narratives are mutually incompatible or something like this it suffices that one of the narrative is convincing to to to one user or one cluster for this cluster to to be convinced and to adopt it so this is another way that this anti-vaccination movement can grow is by exploiting very different narratives and it's very hard to fight against all of these narratives at once because there are multiple and you have to spend energy on each different narrative and so this is an interesting insight to show that it is very very hard to to combat these misinformation yeah in our discussion we we agree that features like how pedagogical is some content and how engaging it is not only how scientifically accurate the content is the matters for for spreading this kind of information so if we if you want to correct inform people about vaccines not only be scientifically accurate but also try to be engaging and pedagogical so that the so that the message is heard like you you want to reach as many people as possible and for this scientific accuracy is somehow boring and and hard to understand and and so it won't spread and it the message is usually not as much yeah and this may be a quite a big limitation in today's strategy of of pro vaccination let's say or or like i think more and more companies so for instance if you search for anti-vaccination on facebook one at least when i did like i don't know what you would get from your facebook not even anti-vaccination just search for vaccine on amazon i like one like if you search like vaccine on amazon books you get anti-vaccine books yeah so maybe they've changed this when we had the book online and so in november 2019 i did yeah and i remember vaccines on amazon books give you anti-vaccine books no no anti or pro or just vaccines yeah yeah but what i meant to say is that i think that there's a lot of work going on from facebook from youtube from google and so on to try to remove these contents so when i searched the other day anti-vaccination on facebook i was only recommended contents from world health organization john hopkins university that's it that's you searching for anti-vaccine so yeah yeah yeah of course like it's only my data but i'm guessing that they've done quite some efforts from uh to try to to to promote more uh why i mentioned amazon because i could this did this search from an anonymous browsers without without being connected yeah so it was not influenced by my profile yeah yeah yeah so i think that there needs to be a lot of work like the the these recommendation platforms then they they're not neutral like it's not clear what it means for them to be neutral um like on every topic should you show 50 50 of every side of the question like if someone is searching one plus one equals two like do or you're going to show like half of the contents explaining why one plus one is not equal to two probably not uh and so uh and we've been neutral like uh that would we call it neutral to show 50 50 of the videos saying that one plus one is not close to two uh yeah probably not uh and uh yeah i think one important thing to i think to reflect on is that this concept of neutrality is a lot more complicated at this than the the simple idea that you should do 50 50 on both sides on every question uh and uh and i think you should actually try to to to be more robustly beneficial given the stakes especially for for this kind of of topic where it's a matter of life and death and at large scale uh and so you might say okay so i'm going to try to to to shift the battlefield like to to to to to to favor uh like one scientific consensus uh i don't think it would be that controversial uh for many topics at least uh but uh in any case like this is something that many uh big social platforms are trying to move towards uh at least from what they say also from some results of of course i've made but the concern i still have is that uh the way they're doing this right now is by promoting uh systematically contents from the big uh John Hopkins or from University or from WHO and the contents that's uh spread by these uh very trusted or trustworthy uh institutions is not the one that may be the most engaging or the most compelling uh and instead of this maybe if you get some influencers to to convey the same message it would be a lot more impactful of course it's also harder to identify this kind of uh of content uh from influencers because like it's easier to for for YouTube for instance to say well we just recommend systematically what WHO is saying but it's arguably also less effective that's uh what I meant to say so it's a very complicated uh topic overall uh just the question of what should you recommend on these uh on these platforms and I think there needs to be a lot more research in this direction yeah and uh adding to this concerning what we discussed last week about uh somehow the backfire effect the fact that if you want to show content saying uh not X to someone that believes in X it does not necessarily make them believe in not X right away and in some cases it can make them believe in what they believe in even stronger by opposition to to to what they are shown so uh as as they say so showing contents from uh from the big groups are doing scientific research on vaccines it's not totally obvious that it won't have this uh this effect that people would further distance to to to what they read uh because it's opposite to to what they currently think yeah yeah and overall it's like intersects with a lot of different fields but uh understanding the understanding the psychology of users and how they react to this or that kind of content even though it's coming from the WHO from the World Health Organization uh yeah I think it's a very open question uh and we need better understanding of how people react to science communication in general we also discussed about uh the possibility of banning such uh such facebook page or such groups and that it might it sounds desirable but it might not uh not be fully effective like youtube has been banning some some some creators especially far right content creators um and uh also like recently you had this story of uh of a twitter uh removing a tweet from uh Donald Trump and then putting it back as I think it's very complicated to and there's a huge backfire risk and in particular in the case of the far right uh video content creators uh a lot of them have uh moved to another platform called Beatshoot uh and uh and this well if you just get one in one of if when you ban people people still produce their content and they bring their followers with them to another platform then arguably it's a huge cost in in the battle I'd say because like you you you are using a lot of the influence you could have had on these people and um in particular the algorithmic recommendations or that could be made more beneficial on youtube wouldn't have no effect on on Beatshoot uh for instance so that's clearly this risk and people may even be more angry more uh I think a lot of the issue with conspiracy theories and particular anti-vaccination is that they're often related to being uh against the system in general or anti-authorities uh and so you can imagine that removing them uh from this platform just justifies their narratives and will increase their influence uh on other platforms it's at least something to consider when you uh I think when you considering the possibility of uh of banning people from platforms next week we will discuss um computable philosophy which is a proposal a course proposal by Li and me in which we tackle some aspects in computing, computability, computer science, the historical aspects of computing itself and the the the interlinks between computing and judgment and stuff like that and why and to actually just like why why computer scientists should care more about philosophy he wants to give us another name okay see you next time