 Good evening everyone, we have a very interesting session as a part of a kind of fake news series that has been going on. There have been a few previous talks which I will tell you about. Today's session is about mining fake news using technicals from social media. We have a very interesting panel here and very soon they will be giving their opening remarks. We have Denny George from Tattel which is a collective initiative of sorts to help identify in time fake news. We have Anna Isaac from the News Minute, she's a journalist. We have Sandeep Khurana who is a researcher at the Indian School of Business. We have Pratik Sinha who is with AltNews which is a very well-known fact-checking website. Before I invite my panelists to give their opening remarks, I would like to ask Sinha to give you a brief overview of previous talks given by some of these people here on fake news related topics so that you can catch up later on. Sinha? Thanks Pratik. We put this event together as a continuation of some of the discussions that we've been having about social media networks and mining disinformation disinformation. This conversation started last year when Sandeep Khurana and Pratik Sinha presented their work including the challenges and their own approaches and the solutions that they were developing at the first elephant conference last year. I think Venkatesh is here and we had a workshop sometime in January, February where some of these conversations continued primarily around the questions of virality and therefore will platforms really take an interest in curbing and in monitoring for fake news and misinformation. We've also had a bunch of discussions around technology itself where Sandeep in his talk last week talked about how human intervention is actually more accurate with respect to fact-checking but then you have artificial intelligence which allows you to scale and given that you have all these ones and you have fact-checking you have the platforms. I think so this is an appropriate time to kind of take that discussion forward and bring the people back. Some of the people back in the room and some of the people here. Thank you Anna for joining us today and to continue this conversation. So this is where we are and you will update the link through the videos from the previous talks on the event page and you can also post your comments and questions there. Over to you Pratik. Excellent, excellent. So just for those who are on Zoom whatever you want to request this, unless you're part of the conversation for the time being is mute your mics. I will do the same soon. What I will do now is I will invite each of our panelists to give their take on this topic for roughly about five minutes. Some of them may have slide which they will share on the screen while you do that. Following which I will have about five minutes where I will give my own take on this topic and then we will have about half an hour of Q&A with the panelists. The next half an hour after that 25 minutes to half an hour will be devoted to Q&A from our audience. So whether you're on Zoom, whether you're on Periscope, whether you're on YouTube, your questions will be curated and I will try to have ask those questions to the panelists. So my first speaker for today would be Denny Church of Tatl. Over to you Denny. Hey everyone. I'm Denny. I'm an interaction designer and a software developer. I'm here to represent a project called Tatl. I got into the misinformation space as a concern citizen so I just wanted to provide some back into that. I've been generally interested in using technology for social good and in the past I've worked on civic tech apps in the domain of healthcare, immigrant rights and labor unions. Two years ago I was working on a chat pod that used a combination of natural language processing and human community members to answer questions about workers. So we had a database of frequently asked questions that we could query against in natural language and if a worker asked something that there was a match for they would get an instantaneous reply and if there was no match they would get paid to a human domain expert who could chat with them and answer them. While I was working on this project I was having a conversation with my friend Sarunima about the misinformation we were seeing on WhatsApp that was just being discussed and segregated by our family members. So back then we quite naively thought that a similar kind of community and tech solution should work for the misinformation problem as well. Since then we've been trying to identify the various gaps in the processes, the community and technology that an open source and community driven project could help bridge to address some misinformation on WhatsApp. One of the problems we identified early on was the lack of visibility into some of these platforms especially WhatsApp. So we've been building tools to help by crowdsourcing or automated shipping to collect data from WhatsApp in large quantity and we've also been trying to make it easy to interact with this data by building multimodal search and tag researching so that other researchers and journalists or factors could explore that data. In general over the years vision for us is still the same which is to make it easy to access accurate information for internet and internet users and concretely to that we've been trying to amplify the work of fact checkers, we've been creating open source tools and services for public fact checkers and journalists and we've also been trying to open up our data for researchers incomplete and also been finding safe ways to open up a subset of that data which can be open up to the general public for them. Yeah, this is from my side. I will ask Anna Isaac of the News Minute to make her opening comments now. Right, so I work with the News Minute to put us out based news organization. I've been with them for about four years now. Just to my personal point of view from my personal experience as a journalist, I think the first time I encountered fake news and had to really deal with it was in 2015 during the Chennai plug. I was a reporter with another television organization at that point and the scale and the size of the amount of WhatsApp fake messages at that point was unimaginable. So the things like the one to two zoo being breached and tigers and lions being out on the streets or things like the amount of precipitation that the city would get. So these are the things that we were getting at that point and you know how energy sort of goes into trying to bust that fake news while you're also doing other things on your on the job. Now what I've realized in the last few years is that fake news from a journalistic point of view, it began with you know these big news events. So they were disasters, they were floods, they were for example the hospitalization of the of the chief minister of government. And what I realized from the floods in 2015 and the hospitalization of Jaylanta in 2016 is that the reason that perhaps fake news sort of proliferated was because of a lack of communication from the government side. There was very very minimal information both and communication and transparency from the government, from Tamil Nadu government during the floods as well as during the hospitalization of chief minister which you know allowed fake news to sort of thrive and proliferate. But since then it's become you know it's not no longer your big news events that where fake news sort of thrives on it's become all pervasive and that's the challenge that organizations like us the newsmen face on a daily basis is because we're a very small team, we're a very young team. So your challenge and your dilemma as journalists and as editors is how much time, how much resources, how much energy do you put into actually busting that fake news because you also have other things to do you know as journalists. So that's the dilemma that we're constantly facing but thankfully of course we have organizations like Boom Live and Alt News that also do the job for us. So but yes at the newsmen we do do we do bust fake news so we do fact checks we also put out fake videos in regional languages busting fake news that's something that that we do on our end as well. As far as you know tools and technology go we're very old school in that sense we we do our editorial processes you know basically calling up official sources getting credible information and that's how we do our fact check and bust fake news that's what it means. Thanks a lot Anna. I request Sandeep Khurana of the Indian School of Business is a researcher there to give his opening remarks please. Sandeep over to you. Sandeep it's your audio working. Yeah okay sorry I yeah I was on mute so okay so I'll try and connect you dots and go on some of the lateral issues related to fake news and I see it more as a game between the criminals and the police kind of thing where both sides keep evolving as they deal with each other. So there are two say different examples which I have taken in one example let's say there is a fake news saying you and declared Jan Ganamana as the best national anthem or there is a falsehood stated there is a fact check done that it is not true but instead of acceptance there is a rationalization so what to me it is still the best national anthem don't you think so. Now even if you give a rational reply but that is not the same as you and calling it the best the other side will say no you don't like your national anthem you are anti-national and it is heads I went tails you lose so it is that post proof kind of scenario where the narrative and the the beliefs pre-existing beliefs are what define your version of truth and there is the fact check is actually not even addressing that narrative part. The second example I have is of a dog whistle where somebody says these people spread the virus now they don't say who are these people but they just say this now there are their fans followers to third parties who will say these people who are these they will get elaborated by the third party and fourth party acts on it as a fake news which means a real world damage where somebody goes and does some crime or some some mob action and those kind of things happen there is a real damage which is done by even some fourth party now when it comes to the first party as instigator they are in denial they are saying I never named anyone I'm not responsible for what what's happening and I mean we can clearly think of examples of this kind of narrative over last few months and then there is a victimization also claimed by the by the original dog whistle the person who blew the dog whistle so they'll say that these people always see the devil in us we never said this unnecessarily they are attributing all these charges to us these are the kind of issues that go beyond just fact check and which is where some of the challenges lie as both sides play this game of dealing with each other so this is how the narrative setting goes the dog whistle is blown there is a amplification of that and elaboration of that putting meaning to that using the second party and then because of the velocity and volume on twitter these things start trending and that amplification because it is in trends it goes to media media spillover leads to more social media coverage because again media houses are present on social media and this vicious cycle leads to easy manipulation of public minds using the fake news so this is what I do I use a different tool which is the network analysis the map at aggregate level who are the players what they are talking about and then try to look for patterns and try to look for the specific narrative or specific players who contribute to that and try to make sense of that starting from something very amorphous and defuzzered and difficult to make sense and then try to ascribe meaning and look for the players and people and describe them roles so I covered them in detail in the last week's talk how we go about it and it comes to getting the anatomy of that fake narrative the role players and the reactions and I've taken a live example from that day when I gave the talk how a particular tweet went viral and it was a speech and it was a fake narrative but then it was built into something so this was the details of that at that point it is almost double of this now it's almost 15,000 16,000 retweets and 2000 replies and it is like whatever I described just now so for the past year due to past year of time I'll not go into detail but this is one context which I want to set the second thing is the fact checkers do phenomenal job it is like apart from a lot of say tools intelligence technology there is a lot of courage involved there is a lot of say background work and rigor but more than that I would admire the courage because in last couple of days itself you've seen the amount of attack that has come on them but there is a good value beyond just the first order effect of just debunking one fake news so one fake news is just one part that evidence enables the law enforcement there are patterns which are useful in the sense of building reputation and history for example there was one case where there were 17 fake news belonging to just one Twitter handle and that was debunked by I think Pratik and all news team so those are the kind of things which help everyone it helps research it helps direct action it helps the third order effect of even the public awareness and it goes beyond just to the extent that today there are a large number of fake news say organizations which are thoroughly discredited and the reason for that is the work done by the fake news the fact checkers which are going up to the first second and third level and of course there is the removal itself and the chilling effect or deterrence effect because of which the second and third order effect is also there but having said that the order of problem is also huge and why it is huge we would not all be discussing all this today here if there was active cooperation from all these agencies so there are vested interests so whether it is the social media platform whether it is mass media which there are sections which are compromised or whether it is the regulator government law enforcement agencies they are neglect reluctance or active promotion at times and there are also the lack of scientific temperament in the larger society lack of education those issues and even if we are dealing with fake news we still have to uphold freedom of expression and minimalism and and upholding all human constitutional values so it is under these boundary conditions that we have to due to address the problem if the law enforcement agencies were acting there is enough evidence the fact checkers provide or there is enough evidence the research can lead to or if the social media platform did not have the the kind of commercial incentives they have so those are the kind of issues that make the challenge more compelling and more difficult if these things were aligned correctly in one direction none of this would require us to be discussing this issue here so this is a quick or maybe not so quick say context setting of view of the lateral dimension related to today's discussion because I don't come with a direct fact checker fact checking but I deal with a lot of related issues as part of my research so I just stop here and over back to you Prithviraj. Thanks a lot Saty. I now invite Prithviraj to present his new remarks Prithviraj over to you please. Hi thank you for having this discussion I think this you know while talking about the politics around misinformation and the misinformation itself I think one of the issues that we don't talk about enough is that at the start of the day I think technology has a role to play in as to where we are and how you know Denny spoke about sort of technology for good and largely I have worked as a software engineer for over 15 years and and largely the tech industry has forgotten that aspect where you know profit driven sort of profit driven motives have resulted in in such an ecosystem that most of the software companies especially you know the ones which are directly related to this that is the social media platforms etc they always do just so much that you know that the PR does not go bad they won't do beyond that I mean one of the latest examples is you know the TikTok video where they got without fully accepted to do a video saying that you know forward must go and that is where the message is limited to I know TikTok has joined hands with one specific fact-checking website but again if you look at the sort of problematic content that comes out of TikTok it is not you know it a lot more needs to be done as far as the platform is concerned I'll I'll talk about what my understanding of what technology can do vis-a-vis what I have learned in the past three years so there are three aspects to fact-checking the first aspect is a collection of information that is what is being spoken about in social media who's exchanging what and in case of India it is most of the misinformation is mostly in terms of images and videos especially the viral misinformation if you look back in what you see on WhatsApp it is a two to three lines of text and an image or a video yes there are these long long as WhatsApp forwards but typically those are more about data you know this government has done so much this government has done so much they they're typically not as problematic as these images and videos which are often misrepresented you know all videos are represented as something which is happening in present and very often with the communal narrative so so okay so the first aspect is collection of information what you know who's talking about what that and that is something that can be centralized using using technology I'll just show you in a bit what we have done I'm not sure what are the next questions I don't know whether to show it now or later please let me know but go ahead you could show it right now okay okay I'll I'll show what we have done in a bit so first there's a centralization collection of information and and that can be done using technology so that two things can be achieved number one you figure out statistically what is more viral what is less viral what you know which which story because one of the important sort of indicators as to which story we prioritize is based on what is more viral and over time obviously you know the team that we have at all and I'm sure the boom life we want so we have a pretty good idea as to you know by looking at certain indicators on social media you know how much it's getting shared on Facebook how much it is getting shared on Twitter and both you know all the factors now have individual what type accounts we know what is viral and what is not as viral but but I think there's very little data around the issue of misinformation and centralizing this will help in collection of statistics that is one thing and also help us help it helps factors and prioritizing let me share my screen for just a second okay so so this is what we have developed if you can see my screen so let me go to manage requests so what we have is that we have an app on on the Android Play Store we don't have one for we don't have one for iOS yet we focused on Android because that is what most Indians have so these are the requests that are coming in we have I don't know each 33,000 pending requests the design issue that we you know we realize later is that we made the app into a sort of a call center kind of interface that you send a request and we'll respond and that obviously cannot work with us you know with a 10 member team so the app is going through a redesign we are thinking about more things but so I'll show you an example so these are the requests that are that we are getting in today and we also get it on WhatsApp but earlier we you know we we tested WhatsApp a lot to give us the API which they refuse to and but now we think that that may not be the right approach because again WhatsApp is also like you know there's unless you have a robust API on WhatsApp you know automation won't be that easy but here what we have done is for example you see this one video coming again and again brutal assault on Rindavan press brutal assault on Rindavan press so when you see this thing this thing coming again and again that is multiple users and we have anonymized data except for the MAC address of the user we don't collect anything so it is not that people are worried as to which side of the political spectrum they are from we have no demographic data about the people who are putting in who are sending the data so so the people who are sending in the data they don't have to worry about their identity so but here if you see you know you you're getting this again and again which means that this you know in a short span of time the people who are on the app have sent this four times and they are from different MAC addresses if you look on the right you see that they are from independent MAC addresses you know some of them are from the same MAC address but very often you see the examples that they are from independent MAC addresses which means that it's vital on social media because it is you know so we look at the difference you know how often we are getting a request and how you know how how many are getting certainly the velocity of the request and so that is what I mean by centralization right now the centralization is not there because it's only the app feeding in the request we have an API where we're going to try it on social media platforms so that we send all the requests in one place that is Twitter Facebook all of the requests essentially come in one place now the second part is fact checking fact checking is mostly you know driven by humans you need a human brain to figure out what is true what is false but there are exceptions to that now the exceptions being there are a lot of things that these social media platforms can do to make things easier for example one of the most common forms misinformation is old videos and just used in the present context that is something that the likes of Facebook Google and Twitter who have really deep pockets they can start indexing all the videos and images and whenever an old video is viral they can mark it on the platform saying that look this is an old video they have not done that yet but that those are the things tools that can be developed in the process of fact checking itself to make the job of fact because easy and to eliminate certain kind of fact checking if if a Facebook video is posted and it comes with this thing that this is a two-year-old video like many journalism platforms are doing let's say even if you go to Bhajian if it's an old article it'll say this article five years old or this article seven years old because you know many people tend to share articles thinking that this is something that has happened it doesn't while it must have happened three four years back so those are the kind of solutions that Facebook you know that comes like Facebook and Twitter can look at in terms of automating fact checks the third part of fact checking is disbursement of fact checks so that is once something is fact checked how do you automate the delivery of fact checks so this you know this request log that you're seeing what you're seeing is that the system that we have created that is automatically responding to fact checks which have all you know queries have already been fact checked you saw some things in managed requests which are coming in and which we have not which we have not fact checked yet but what you're seeing here in is our request that are automatically getting responded to without any human intervention so we are taking the human intervention out so I will stop sharing yeah so this is you know this is something that we have implemented and you know there's a there's too much talk about AI and AI ML around this issue we use I'm a big believer of using very very simple technologies which address 95% of the cost and reduce the infrastructure costs AI ML will always mean a lot of competing costs which means your infrastructure costs load up what we do is very simple comparisons you know for example if a video is viral on WhatsApp there's a cryptographic hash called MD5SUM you just compare MD5SUM and you know that these two videos are the same so I'm a big believer in you know very very simple technology which can address 95% of the issue I'm not gonna I'm not bothered about the rest five percent which maybe AI ML can address I'm also I'm not completely convinced that it can so yeah so these are the three bus fact checking collect centralization of information collection the fact check itself where I said there can be some amount of automation and disbursement of fact check so that is how I think technologists should look at fact checking of how to automate these separate parts and work on solutions and as I said you know fact checkers work on a very very tiny budget so if you're going to give us solutions which are going to need you know lots of AWS and lots of Google compute and all of that it's not going to work out you need simple solutions which work for a majority of the population which is not which does not cost thousands and thousands of dollars advice infrastructure system then yeah so those are that is from my side thanks a lot Pratik and now for Venkatesh of Boob to present his opening remarks Venkatesh over to you thanks so much can you all hear me yeah good so I'm going to share my screen in a second following Pratik is always a call order but he always makes you think and I love that and Pratik your framework of collection and then the actual process of fact checking and distribution is a very very nice framework I'll try to fit my framework onto this I'm presenting my let me see if I'm presenting right so can you see my screen can Pratik please just a thumbs up yeah okay so I feel this is a lot of contains a little bit of what Pratik has been saying and also a lot of my own thoughts and I feel I think I will take us through what I believe are the broad trends I think at the very beginning there are two broad trends I see in the technology community a lot of the resources are being used to understand misinformation still I mean it's probably the fourth year since you know and we're still trying to understand the phenomenon and a lot of research is still trying to figure this out I mean using data collecting data yes but also you know models and predictions to see what the patterns are on various platforms and I understand why that's the case because you know big news or whatever it is is you want to call it misinformation disinformation information disorder is like a shape-shifting monster and you know the moment you feel your you've addressed one issue another issue crops up so this is a one broad trend the second trend I see is there are different types of solutions and sometimes the solutions are okay I have this very cool approach whether it's a you know AI approach or whether it's like a some other tech-enabled approach let me see if I can fit this approach and let me see if I can find a problem sometimes there's a mismatch and this is where what Pratik said tends to echo with me because he says we need simple solutions 80 percent of misinformation out there you know it's image and video based at boom we carried out a piece of research on all the coronavirus related fact checks we've done it was about 280 or 290 must be more than 300 by now and we realize the trends are 30 roughly video based 30 percent I mean the numbers are a little different but these are the broad trends 30 percent are 34 percent are video something like 29 percent are audio sorry images under something like 29 percent is x only and audio based misinformation accounts for some 2 percent so we don't need like huge solutions we need like small solutions incremental kind of solutions and I've seen also when you look at technology from my perspective as a researcher and someone who's looked at this field for a few years I see some poor tech specific trends there's big tech they get a lot of pressure on them and they fund research they fund efforts to deal with misinformation they also fund fact-checking organizations the international fact-checking network of which both boom and alternatives are a part of you know these tech giants route their funding through IFCN but also through other means that is that is that is happening but you know you have other approaches as well you have an abundance of technology startups that are trying a multiplicity of approaches for example there is something called credor CRE DDDR that says it's rotten tomatoes for news so basically it it brings in people who are going to rate the quality of news you have an organization called RepuStar that is bringing in activists to help deal with misinformation so you have a variety of approaches and universities are also trying different approaches they have their labs they're doing their research they're holding convenings and then you have individuals and interested groups I think even in among the panelists today we can fit ourselves in some of these categories Denny for example is an interested individual person and then you know he's also a technologist and so on there's nothing from our perspective there's nothing wrong in this and we need the more approaches we have the better it is ultimately not every approach is going to work out and we've seen that as well these are the examples I mentioned rotten tomatoes for news for news activists what we feel is needed is in India specifically we need a scalable solution and the scalable solution has to take into these realities the fact that the same piece of fake news comes up again and again there are multiple languages in India and you know one comes up in Marathi the same thing a month later is going to come up in you know Odia or maybe at the same time you have multiple platforms you have WhatsApp you have Facebook you have so many platforms we have share chat you have tiktok you have hello and then you have uneven literacy so you have as you go across the country you have people who need debunking in in in in various formats audio is a is a is a big way to debunk but I don't know if we have explored that sufficiently and you also have to take into account uneven internet access so this is the reality we're dealing with what's needed to start with at least from a boom or you know other factors perspective we have huge organizations around the world like Reuters and Associated Press a lot of them are using these little programs bits of computer code that essentially write their stories and a lot of fake news debunking involves repetitive tasks and if there is a way we can automate this process it freezes up so essentially the process of fact checking itself is automated or we could have like bots on various social platforms that you know each time somebody tweets something or you know a known offender tweets something you have a bot that says you know this is something that my colleague Karen Rebello was pointing out she says there are some organizations that do that and also you have platforms that are cracking down and we did fake news we need more of this so this is a broad kind of landscape landscape from our perspective we need a solution that is scalable but when we talk about scale you know if you look at the silicon valley model it's the it's a bit like saying okay here is an organization that's throw lots of money at them let's scale up really fast we've seen that that kind of thing doesn't tend to work so I mean one thought experiment just for this session I thought was we can either all try to scale up in multiple ways it's individual organizations or we can think of ourselves as thoughts in this gigantic organism that is essentially dealing with fake news so essentially everybody in this chat and all the people are watching this chat we all have a role to play and if we learn to collaborate I feel both non-tech people and tech people and tech people you know amongst each other and everybody together that's the only way forward that's basically it thanks Vakatesh thanks guys it saw some very interesting perspectives I thought I'll share a few of my own as well here I have I will be sharing a lot of what I plan to talk about has been covered so I'll go very quickly wherever it's my screen visible yes yes great so there have been lots of talks before on fake news so I'll skip there are many kinds of fake news there could be boxes there could be conspiracy theories malicious rumors outright fabrication or versus wrong reporting etc what I found interesting was this article by the BBC with very nice illustrations about seven kinds of people that said fake news and I keep in mind that why I'm doing this is because the end goal is to talk about maybe AI or ML approaches to detecting fake news so it might be useful to know who are the kinds of people they have 4chan and even small-time jokers who want to see the world re-construed there's cameras you can make money politicians you know they're very much part of this ecosystem I don't want to name names conspiracy theorists did China do biological warfare and come up with the coronavirus insider sometimes may be well-meaning you might have a doctor who thinks that some information is worth sharing that information may not be correct in India I think we have a term called the whatsapp uncle a relative and of course you see in celebrities right so this is what they say are the seven kinds of people I think Sandeep has presented this in his last talk but for those of you who are not there this is a very famous study published in the journal science last year two years ago what it says is that fake news is more novel than real news and as a result it propagates faster deeper and broader on social media so the depth of the cascade people who you don't know who are very very far away from you socially might also be subject to your fake news in an information cascade and one thing that they do is both sentiment analysis and topic modeling topic modeling suggests that political themes fake news with political themes tend to be propagate faster broader etc than the terrorism natural disaster financial information I think they do is look at replies to fake news now all of this is on a massive twitter data set by the way so the tweets and all can be mapped and they show that fake news illustrates fear, disgust and supply elicits fear, disgust and supply in the replies but real news is more of anticipation joy, sadness and trust anyone here who's familiar with topic modeling or sentiment analysis would be familiar with how this may have been done another thing that I found and this is actually a very sobering thought this is an experimental study again in a very famous journal of experimental psychology general they say that we kind of know this but they did it very rigorously is to show that repeated statements in fake especially fake news the more you repeat them and the more convincingly you repeat them tend to be perceived as true and interestingly enough if you do a fact check and tell people that you're believing something which is false they may not believe you or they may not update their beliefs now there are two parts to it one is fluency so if somebody very confidently fluently and repetitively repeats fake news and we know there are legend on it who does this quite often people have a heuristic a lot of people have a heuristic that there is a higher fluency rather than knowledge so in simple terms a clip liar we easily believe this is a little sobering because you know at the end of the day we are trying to correct people's misconceptions etc psychologically it may not be working so this is one perspective that an academic research so now fake news there are two aspects to it that machine learning researchers are interested in one is characterization and characterization is based on psychological foundations social foundations in traditional media so now you know that there are some media houses between mistakes there are others and increasingly and disturbingly so we are seeing there are certain media houses that seem to be doing this on purpose I think two days ago there was a story about a so-called tiranga virus in tomatoes that was again Maharashtra and according to S.C. minister Prithviraj Javan who has just put it on Twitter he claims that it caused a crash in tomato prices a tiranga virus may sound very stupid to you and me but it had real consequences of course social media you know that there are these malicious people these echo chambers etc more interesting is detection and this is where so there are parts to what Prithviraj said and one of the major ways of detecting fake news today stands based which means you try to see if this media house has stands left right etc and without taking names maybe a certain stance is more prone to fake news you can make your own conclusions what that stance may be but other is a very interesting method called propagation based this is very similar to what Google does with the credibility of web pages using the page rank technology so if credible sources are linking to you then your credibility increases this is one way to do it but there are other ways which are knowledge based which is actually manually fact-checking or style based now there are certain stylistic elements that you can see in certain fake news for example you might see UNESCO coming up a lot in tech news etc and these are some style based detection method so for any of the listeners here who are interested in learning about fake news detection algorithms these are some of the considerations in place very quickly now most of you spoke about different initiatives of social media companies very quickly I'll just share with you what I found different companies are doing Facebook seems to have the most comprehensive on its own Facebook blog this a lot of fake news is financially motivated based on clicks etc okay you disrupt that by prevent if you know it's fake don't propagate it so reduce the cut of their accession supply building new products to curb the spread of fake news I'm guessing this means something to do with their algorithms again and effective pages how do you help people who are exposed to fake news google google and youtube on the other hand they have they are investing at least in India to train journalists and fact-checkers that's what at least public reports say globally also they have developed tools especially for google news and of course similar to facebook changing their algorithms propagation algorithms especially on youtube whatsapp has a forwarding limit which they recently did a search message feature which is if something is very very vital not all messages something very very vital you can search it on their user interface and see and look at its velocity and they have a big FAQ I did not find much from twitter except that they plan to label coronavirus related suspicious messages as suspicious and of course prefix book about the much forward campaign and an in-app reporting feature that ticked off head so I'll keep this slide here I'll keep this slide here because this is what the platforms have done from this point onward what I would like to do is start asking a few questions to a panelist here typically what I do is I'll probably nominate one panelist a question but please feel free to join in okay so so I'll actually stop the share here because I have my questions on my thing so the first question let's let's start with the use of computers and AI in tagging fake news now I thought two perspectives one is Pratik who said that he's not that big a fan of AI he would rather preface implement it right so which is something like a database match I mean I'd like to ask you one thing so see you know that AI has a lot of bias in it right we have seen AI biases in sentencing in airport risks etc with racial bias in this particular case would there be a possibility that there's a stance by it so you may have a lot of right-wing propaganda which makes right-wing it which makes right-wing even legit people more prone to be mislabeled as fake news is this kind of bias possible yes so surely like the AI or machine learning or or any even statistical assumption in all this is that the trends of the past hold on to the future so if I do machine learning on a training data set and my primary assumption there is that whatever is being learned from by the machine that is what I will be applying it to to then get the the best results now that assumption is violated like in in case of fact check it is it cannot be assumed that the every time the the story coming will be a repeat in fact not just the story but the whole narrative the terms used the jargon the images everything is very dynamic and it changes so it is a big limitation of AI based models so even if you take vocabulary so the terms used for certain communities for terms used as part of say even misogyny those terms more and then those terms may be used in languages in which we may not have AI models or the training data it could be English it could be Hindi or it could be the font also is different or even the in English font it is the it is the say a English word combination of Hindi English so there is that serious concern which which extends not just to the the biases that you said if the if the training data set is being say designed by a body which composes only of one race that is the kind of example which is there in research and same way it could be here the training data set researchers own values also get reflected in the choice of the labels that we have checks and balances by way of interator agreement multiple people doing it and those things but I would say it remains a concern so one is the opacity of algorithms which the platforms don't reveal we don't even know on what basis they are shadowbanning a certain tweet we don't know on what basis the views are increased or decreased it could be commercial motive it could be a real scientific stuff it could be there is an algorithmic bias in the background and opacity is one part because of which we can't even get to know or comment and the bias is the another part which sometimes through secondary research analysis we can even point out like there are those research papers you kind of refer would anyone else like to pitch in here I won't come in here and I want to ask the question you know why are we even talking about AI and what is AI going to solve because there's every conference tech conference I go they talk about AI from the point of view of misinformation at least from the Indian point of view number one what are the four kinds of misinformation audio what is audio in where somebody said that look Lakshmi Mitchell said this and we find out that Lakshmi Mitchell has not said this can you detect that using AI unless you have everybody's audio samples and even if you do that you cannot detect that kind of misinformation using AI okay number two you have videos and again I'm talking purely from the Indian perspective I'm not talking about the bright parts of the world you have videos which are you know preceded by two lines of text now what is AI AI is essentially you have a database and you predict that a new piece of information based on your existing database what does it say anything about the new piece of information broadly now this new let's say a new video comes with with two lines of text number one NLP you know with two lines of text with just two lines of text your natural language processing you're not able to going to be able to detect bias or at least in an effective manner there will be too many false positives as far as your NLP methodology is concerned because you don't have a large large enough sort of block of text as far as videos is concerned now the only way to figure out where the video came from is by looking at an existing database of videos that is what we are trying to create in a very very small manner as I said Facebook Google and Twitter of the world have enough money in the pockets to create a gigantic database of videos and images which one can query which factors can query to figure out where a video came from number number three images same as videos you know there's no difference image for video is essentially a combination of multiple images and number four is text so text you know a large block of text there maybe you can run your NLP algorithms to figure out what is the bias etc but is it going to be effective I mean if you go out and tell somebody you know you also spoke about how to convince people even with a legit 100 person fact check very often you're not able to convince people are you going to be able to convince people that there is a 80% chance of this being misinformation and 20% chance otherwise you know so I you know this based on my current understanding of technology I think at least from the Indian Indian perspective EIML is of no use if I may respond to yes please two three points one is of course AI is not deterministic it is not going to give a clear yes no answer and there is definitely not just the the probability or deterministic part but also the accountability part because of which the human element has to be there and the nature of the game itself is complex enough where someone has to take a manual call so there is no doubt that there has to be in every single fact check there has to be a human element that is one part the second thing is that I'm not if we are not able to answer using AI and ML an individual say a story or an individual fake fake news but suppose there is an aggregate level let's say a particular handle is islamophobic or misogynistic or a hate speech now that there is an aggregate pattern which gives a high probability or low probability which then gives a clue that okay these are the things to be looked into just as what Pratik says that okay this is coming in more often this story is coming in more often and this is what people are referring to us frequently so there is that element here also that at aggregate level the accuracy is better but at individual story we cannot do that the third part is when we use these kind of say tools it is classifying the the volume of work which could be let's say in millions of tweets into red orange green now in red we are very clear that these are hate speech in green we are very clear these are not hate speech there is a chunk of orange in between now in the absence of that what do you do you can't go through all million so at least it prioritizes what we do yes it is not good for fact checkers because they deal with only one story at a time but if I look at the aggregate level in using these tools even for fact checkers where I would say so some of the area where I'm working and I got into this hate speech as part of the research there are screenshots posted now just a moment something I would like to hear distinguish between hate speech and yeah now there is a lot of intersection between the two but they're not necessarily the same yeah so let's let's stick to the fake news aspect of it irrespective of the motivation and I agree that in hate speech it will be useful because hate speech you can recognize certain characters as in hate speech but misinformation does not necessarily have those characteristics in fact you know in your presentation you use the word dog whistling in case of misinformation there's so much dog whistle you know that he is doing you know and those kind of phrases in different languages where you don't specifically target anybody you know there's a lot of dog whistling and how do you sort of capture that using yeah I'll move to my next question though I would like to get Denny's views because you know yeah very quickly I think I just want to chime in that I totally see how you know like like looking at the dashboard that Pratik saw where they were like 30,000 pending you know videos right let's say at best AI might just flag that some of these are very obviously fake right but like I don't know I'm not sure if that information is that in that in itself useful to him you know because as like the amateur Pratik at my own home when my mom shows me something whether this is fake or true like knowing that just by you know being just by having seen enough you know fake news before like I kind of knew whether something's true or false right away I do see the application of AI on our side because like before we if we collect mass data and in order to provide some kind of overview or aggregation or insight on that data maybe you know AI on our end can be useful but like maybe when you when you create a technology that's directly interfacing with fact checkers like I don't think you know that okay okay thanks I'll move on to the next question to Anna this is regarding mainstream media now there are two parts right one is mainstream media has taken has historically had fact checkers and we do see some sort of scale back the budget in the scale back a be the other thing I would like to address is let's forget about the motivated mainstream media who are the propagandist now that's a separate story but we do often see genuine well-meaning media often fall prey to fake news do you think at some point it is possible to have an ecosystem where you don't like we have auditors for our financial reports is it possible that a mainstream media lets the time of India or the news minute or whoever ties up with a boom or an alt news in place that we are verified by them is some sort of business model like that person well I'm sure in terms of collaboration yes we have think you know picked up all news stories in the past we picked up a boom like stories in the past so in terms of collaboration definitely in terms of you know fact checkers as a as a you know specific details specific role in the organization in the organizations that I have worked at you know over the years we I've never encountered anybody who's a separate fact checker the newsman also we don't have somebody who only fact checks everybody is expected from the reporter up to you know the editor point of view who then you know okay is the story and then put it out it goes to obviously various levels of fact checking before we publish a story depending of course on the nature of the story it has there will be you know multiple levels of fact checks from the reporters point of view you know each editor may have their expertise so it'll go to the editor of that expertise before it actually gets here and published so I mean at least that's how the newslet operates where we have multiple levels of check especially for very sensitive you know news that we're putting out yeah so I hope I want how do you check here is a question though if if I give you if I call you up and give you a tip how do you verify if I am motivated or I'm reliable or whatever so I mean it depends on what the what it is right we do get multiple tips on social media with a digital platform after all and we obviously try and get other people to sort of either share have the similar kind of experience depending on whatever it is and you know therefore and thereby cooperate it and then you know depending on what that that tipoff is then obviously you get the official version to see what the official version is so that's how we sort of you know try and get multiple voices to speak on the same issue if they have the same experience then of course we will go ahead and again of course and get a my next question will be to Denny this is about community and crowdsourcing I also see a few questions on the chat and the Q&A I will come to that you know may I chime in on the whole media aspect please please please okay so it's like the problem with the media is you know it's like the russian doll phenomenon there are so many aspects to why the media calls for videos why the media does not verify enough and the newsman it is an exception actually most news media in India the problem is this is a problem that affects everyone is a business model problem your incentive is to create as much content as possible but as little time and effort as possible in order to generate the maximum revenue so you know unless the organizations media organizations start pivoting away from this business model and many organizations have started and unless the demand from readers and viewers comes with greater frequency that you produce quality news and the willingness to pay for news then until that happens I don't see you know any easy solution you have media organizations that themselves call for and they publish it and they don't do it because they intend to they just do it and so they don't have the structures in place to add verification and when you talk about verification you know when I was in TV news verification meant you independently get cooperation from two different sources in today's day and age that may not be enough you know two is not enough you might need more independent properties you might need to get away from the breaking news mindset then will that work and that's a huge problem you can go on and on talking about the kind of issues the media and the pressures under which they kind of operate but this is at the very heart of it thanks yeah thanks I want to address come to one more aspect of this and I'd like to ask Denny this question this is about community involvement now one very intuitive thing is can you crowd source right now people like boom fact news etc the crowd source at least the reports so I can report something that I find suspicious etc the next level that probably could be possible and maybe you're looking at also is crowd sourcing of the verification itself now let me just give you one anecdote that I came across a long time ago it was a very unfortunate Boston marathon bombing that about some years back that we all know about and I wasn't ready at that time and there was a very well-meaning reddit community that was trying to crowd source who the suspect would be and one thing that they with lot of triangulation lot of evidence etc people producing is that some Indian origin student was missing maybe a one day before the mommy happened etc this guy must be and they went on and on and on about it to the extent that reddit had to officially apologize when it came out that other people were involved and so we can see that maybe sometimes you know missing aeroplanes maybe crowd sourcing can work in scanning images of the ocean etc but in many cases crowd sourcing can have its limits what are your thoughts on this I think I definitely want to distinguish between what's like really relevant to the Indian context right now and then talk about you know crowd sourcing fact checking as like like as like a hypothetical solution right I think right now you know I've seen this thing talked about a lot in the fact checking circles is that the reach of already fact checked information is very limited right so it's not like misinformation is not being factored it's actually being factored very thoroughly using you know I have seen guidelines or you know those those structures but it's not reaching out to enough people so that's one thing right so so maybe like the in the current context that's probably a bigger problem but yes I think well you could argue that there's some merit to opening the process of fact checking itself you know to make it so that people can actually I mean as a way to build trust with your community right so if let's say the people can see how our use is going about fact checking something or if you know you can attach let's say first hand information that you have maybe that will you know build more trust with the community but I think it's hypothetical you know yeah it's worth trying I guess anyone else has thoughts on this I can come in here so please so for example I'll talk about the story that we did yesterday where we fact check a claim by opindia where they claimed that that boy a Hindu boy was sacrificed at a musketeer in Gopalganj Bihar and it was a complete completely fabricated claim so that is one kind of story and this even though I have never been trained as a journalist for say I've only been trained as an engineer but if you look at that story that only has a journalistic aspect in the sense that there's where there's no boogal rivers in my searching none of that right so we contacted a journalist on the ground that person went and spoke to multiple people and all of that we got the postmodern report we reached out to a forensic expert there's the same guy who had done forensic work earlier from the tarika links and all of that so really high-profile guy who works in Sydney right now and so it was a regular journalistic piece so in here when you you can't really talk so firstly but what you can do especially for small organizations like us what we have done is we have been able to get in touch with people everywhere and this is not the first time we did this we get in touch with people on the ground who are also equally interested that this particular misinformation be debunked and they often help us so that is one kind the other kind the other kind where the fact checking starts with the help of a digital tool that is you use something like an in vit or a Google reverse image search to first get to a point where you figure out where this you know what this incident pertains to sometimes it is also just clean observation you look at a license plate or you look at a billboard to figure out you know to find the location of that incident and after that you actually have to get down to regular old-fashioned journalism trying to get to people sometimes it stops right there that is it is an old image you have a news report and you publish it so these are the things where crowdsourcing and technology can help that is you you know somebody even if we don't have the technology you know even if you have parts of technology that is okay this is viral somebody does a reverse image search till the time it is automated say okay this is the original source of the thing you know a crawler goes and finds out you know what is the core of the text in that article which you know it says that this incident pertains to this and then creates an initial draft of a fact check which which an editor can quickly which a copy editor can quickly edit and push it out so these are the kind of things that can be done to to simplify and to fasten the process of fact check which is because eventually we will have tools where fact checks will reach more people I mean what what we are doing we are essentially creating an API that is that we are going to try to you know telegram all I mean Twitter and then we have other ideas as I the original design of the app itself was not the right and we are trying to redesign it but but eventually we will have tools where fact checks have a much larger audience but then we will have the process of fact checking which itself takes too long so we have to figure out ways to sort of start doing that and there are solutions and I know we will eventually get to it then you want to do something here you know I was just thinking the way Pratik was describing this whole thing and I kind of almost thought of you know conspiracy theories then like the kind of pizza gates up that happened in the US like I can totally see how some guys just get extra motivated to you know find loopholes and you know spread half-truths and you know conspiracy theories that don't need to exist yeah so there's a there's a other angle okay I'll take some questions from the social media field uh this is a question from Torsha Sarkar who says social media platforms have been at the forefront of the conversation regarding the regulation of fake news and are mostly known for being slow responders what role in the opinion of the speakers should these platforms play in facilitating and helping the fact checking community and making the process seamless I'll just add one point here I think a lot of these platforms have their own uh the uh financial incentives here you need daily active users uh you need activity going on plus you need sponsored content a lot of fake news you can see on twitter at least I see a lot of times the really malicious stuff is being uh propagated for a fee so on one hand they actually have uh an incentive to make more money increase the valuation on the other of course they need their credibility plus they need to pay plus they need to be seen as caring uh about fake news etc right uh what do you think can be made to make these people more responsive to the fake news idea to uh to curb fake news anyone oh Venkatesh do you want to comment Venkatesh can you rephrase that question yeah so basically what Torsha asks I just added a few points is that uh some of these platforms have been slow responders do not give API access properly etc right so what she asks is what role should these platforms play in facilitating and helping the factory community should I can say uh I'll add another question how can we incentivize them to do this how can we incentivize technology I mean how can we incentivize Facebook or WhatsApp to share their API or whatever let me wear you know slightly opposing ads okay and thanks Prateek for crossing my name with that thing to answer this question one is there is this um uh this analogy about cars I mean we've had technology since forever right each time there's a technological breakthrough there are a host of problems that come along with it so when cars are invented it took how many years 30 to 40 years before seat belts were put on cars uh then you have the you know the breakthrough in genetics right which Saman Subramanian talks about in his in his book uh so now the breakthrough in genetics along with that came with this movement uh the eugenics movement that was cooperated by the Nazis you know this obsession with creating a pure form of a human race that kind of so each time you have technological breakthroughs the pro technology argument is it takes some time for uh the world of technology and everyone else to demand accountability and we are currently in that phase uh so that is a pro technology argument right and I there is some limit to that uh I mean merit to that argument I believe uh you know you have the Facebooks and the Google's who set out to dominate the world but they also have good intentions and then they realize they're blind to their uh to their own creations this is the analogy of the Frankenstein monster also and then you know they kind of lose control of what's going on and then they try to fix the problem uh so that is a one uh that's one way of looking at it the other way of looking at it is uh you know they have to be more accountable they have to do much much more because we already have a society that is completely ripped apart uh you know how do you deal with it unless you have the most resources for the most you know the craziest problem at this point in time and here one good way or one uh dangerous potentially dangerous solution is having governments uh you know being demanding more accountability from social media platforms uh you know what the European Union is doing uh vis-a-vis privacy for example uh you know we have we can have uh you know there are people who are people who are calling for a coalition of governments around the world to demand greater accountability accountability from tech platforms uh the danger of course is you know this can be hijacked by special interests and nations that want to uh benefit from the state status quo uh so this is that's the problem you know the big problem for me is every time we start talking about fake news it just very quickly becomes uh you know talking about society talking about all the problems we have in the world everything and then it can get overwhelming really really fast uh um but so i i am increasingly of the mind that we need more accountability from technology box thank you Pratik yeah so you would buy something yeah yeah so uh number one if you look at the pattern you know Myanmar happened and that is when facebook realized that facebook zero is a problem and uh they uh i think it was not called facebook zero it was called free basic free basic there it was called free basic there uh so Myanmar happened and then that is when they realized you know free basics is a problem the child kidnapping rumors leading to the mob lynchings happened and that is when what's happened you know desperately try to do something so again and again around the world we have seen that it is only when you know that and there have been many indications of misinformation causing this kind of a catastrophe but only when it reaches to the point that people start dying to do these platforms suddenly realize that okay we have a problem so that is problem number one but in in case of India there's a bigger problem the problem is that um that the corp you know we keep up talking of ease of doing business and as Venkatesh pointed out we have we end up talking about everything over the society and that is a fact that everything in the society is you know misinformation is a societal issue it is not just a platform issue it is not just uh cultural issue it is not just a issue with platform you know uh for example it is it is related to the governance themselves i mean uh you know one time twitter suspended a few handles and uh the parliament you know called twitter asked twitter to come and represent themselves that is the kind of uh that is the kind of influence that that government in India has over these corporate vikis uh so for them to get anything done so my basic point is that you know try try bringing down some of the big pages which which which put pro pro you know pro-government propaganda and they put out misinformation and i know if you try bringing down some of them there'll be all hell will break loose so the issue is multi-fold and yes platforms can do much more but in the present political uh environment in India they also have the limitations uh and and this is i'm not speaking in favor of them i'm just saying that everything is true also privacy issues right at some point these guys are going to say give us details of who has done what twitter just to add to dosha's question one more response yeah so uh most of the uh current responses by the platform as even prathik highlighted with tiktok doing that uh campaign of don't forward or that kind of thing those are either very passive or soft or just advisory suggestive they don't really raise these stakes for the uh for the people spreading fake news so the actions should get deterrent or punitive or maybe at least even if you block account today the cost of opening a new account is also pretty easy but then that is one part where the actions have to get we get more uh deterrent rather than just soft suggestive actually second thing is recently facebook announced uh third party governance board it is an autonomous 20 member body where people can complain against facebook decisions so let's say uh some facebook removed a post and that was wrongly removed uh that scope of that kind of third party governance body is uh is limited but otherwise i get confidence from it being 20 member people of eminence those are 20 people who are eminent global say journalists or politicians or some kind of researchers or or professionals and with that large size of body and with that eminence as long as their scope is also extended so for example if uh they design they they approve a different process which is autonomous facebook or those are the kind of things which will separate their incentives from the the part of fact checking otherwise today as everyone on the panel also is suggesting in some way or the other that because of those incentives they are not aligned fully to fact checking yeah i'm going to ask this question uh yeah so to all panelists probably but uh i often see people using the using hashtags while publishing fake news isn't that actually helping propagate fake news and we see a lot of this propaganda tv channel is really useful actually i think i did mention that uh i often see people using the hashtag that is uh publishing the fake news in their counter post doesn't that boost to fake news okay okay good i don't think either or he's a boom doctor you just say something uh what the question you said you asked was uh yeah we see the same hashtags proliferating fake news i'm just uh uh pasting something here for everyone's benefit which is uh paper by data and society which is run which i think partially funded by microsoft research and uh run by dener boy saying that it's like giving oxygen uh don't give oxygen to fake news so don't uh you know uh when you you know when you're on twitter when you're trying to debunk something don't uh rt it rt that tweet take a screenshot of that tweet instead so people don't find the original tweet that kind of thing i'm just thinking that in here maybe you might uh maybe it could be shared uh then uh whoever is at the back and maybe it could be shared with the youtube participants as well but also the question itself i mean for example i know boom uses boom factor hashtag we use all things factor hashtag so i or covid 19 or corona virus those are the only hashtags we use so i i mean we don't really use hashtags what vegetation is different thing that is don't quote tweet and say that you know this is misinformation and uh i know we often do that uh uh but yeah that is a different issue whether we are amplifying the misinformation further but that has also led to many people deleting uh you know when because when you quote tweet there are a lot of people who go and tell them that look this is misinformation it becomes a credibility issue if you keep calling them out like that it becomes a credibility issue and they tend to delete it so uh if i don't know if many of you notice but uh madhu kishpa is not posting anything fake these days so well she says uh uh shocking if through not anymore i mean uh it's been such a long time since we fact-checked her so it all depends on the debunk as well you know versus Pratik you know rt you know quote tweet something then uh the quote tweet will probably get more uh amplification than the original but you know i'm talking about it from the perspective of all of us you know when we are on uh twitter or uh facebook it's a good idea not to amplify this you know i agree i agree and there definitely there's something definitely that that needs to be thought about whether in fact these days uh when we take especially when articles we only take screenshots and say that this is false so we we also started doing that because we don't want people to go and read that article okay i have one more question this is for ana uh i don't know who has asked it uh the chat window says this from the enab only she hasn't mentioned who it is from what are the challenges with respect to costs and resources that the news minute has to face in collating filtering and dispersing fake news uh i will bundle that with another question which has come in the two and a window this is from nadika while all this fact checking is great how far does it travel on the ground uh related to that i will bunch one more since uh i see some similar things uh which mosaic from youtube has asked uh which is what are the reasons that fact information doesn't reach people as much as the fake news itself so in terms of costs and resources obviously um i mean our fact checking which actually happens in house so you know that's the cost involved and obviously in terms of resources because we don't only i mean of course we fact check every news article that goes out but in terms of fake news first day uh and dealing with debunking fake news uh you know in terms of resources obviously that's something that we we don't do on a daily basis i would say but in terms of collating and filtering what we choose based on our news values so example if a politician or political leader prominent political leader is obviously busting i mean is obviously you know giving out fake news that's a priority for us to to debunk it of course we also go in terms of virality uh that's also you know in terms of virality we would choose to debunk it so we don't obviously do it on a daily basis so it really does depend on the news point of the news value that we see in uh in uh choosing to debunk that fake news um you know i mean the biggest challenge obviously is uh reach of debunking fake news because we are using you know uh social media to debunk that fake news so the question is how much of our debunking is actually going out and reaching that audience so i think that uh you know it's coming to nadika's question um again we're an english news medium we do put out some debunking fake news in the regional uh in regional languages but again that is very very minimal so i think that the problem here is that how do you reach your man on the street and you reach your man on the street who may not necessarily get that fake news on whatsapp or you know on twitter uh how do we reach him and i think that is the biggest issue that you know we need to be sort of addressing at this point is how to reach the man on the street and that's simply through either your regional media your traditional regional media um or uh and basic basic understanding like human awareness needs to increase uh in like uh you know in basically trying to identify what is fake and what is real so samjit maybe you may want to come in here the next part of the question uh there are psychological mechanisms right why the fake news travels fast but the fact it does not okay i saw again borrowing from the paper by wasavi which prithviraj had also mentioned as part of his his slides so the the fake news does travel faster deeper and broader which means that it goes viral it goes and the reason for that in that paper when they got into the why part of it is that because of the content of the of the fake news it is meant to generate certain emotions emotions of maybe outrage or anger or maybe it is a confirmatory bias for our own existing beliefs and the the biggest part which they found that there is a novelty factor if something is new and to be new the kind of content we mostly find is that something which is alarming oh this kind of thing would happen it is like very novel it is alarming and because of that i tend to forward it to two others so that raises the amplification and virality and uh and that is where the statistical validation because of the 126000 new stories being studied by them to full level of propagation how they are going and then doing the robustness checks so i think that that is where the the research kind of suggests that these emotions and the novelty that that lets the fake news go faster and wider and of course there are other behavioral aspects of confirmation bias or hindsight bias or those which are also at play people may not consciously design based on that but based on experience they will know what kind of content goes viral and then because of the likes economy they all want to generate more likes they want more followers and because of that they will kind of fall into that trap of sticking to that kind of narrative and sticking to that kind of generation of content and of course the costs are low if you get caught also the most that happens is you withdraw your post mostly the law enforcement agencies won't act unless it is a rare case so those are the kind of things which are at play the next question so i'll bunch up two questions by mosaic and critica and i'll start with denny and then maybe take opinions from both the fact-checking guys the critique and Venkatesh what is the role that boss open source type communities can play in curbing fake news uh so mosaic on youtube and critica i don't know which platform i have asked very similar questions here i i think uh you know i am in some ways much closer to the audience on this video stream than the fact-checkers and i can like you know speak to some empathy to you know because i was i also had those questions that how can be as let's say technologies help in this space and what we have realized is that uh a lot of the technology and you know uh uh ai or whatever these you know tools that you know are open source and they are pretty easy to get started with and like start developing stuff with but i think what's uh what's lacking is some sort of a standard data set uh about the unique misinformation data that's you know uh circulating in india right and so i think that uh more than open source software there's also some need for open data sets which uh because a lot of this data is collected by fact-checkers or researchers across you know the world working with working in focus of on india but it's kind of like exist in silo and i kind of know why they don't open up open it up for general public because it might you know lead to further application of that fake news but they must like you know if the people on this panel were part of some entity that could you know make these high quality data sets available for you know uh amateur uh you know uh software developers or researchers to like you know so that we to fasten up their tech development and not worry about data collection which probably has already been you know addressed at this point by then and just like to take views from both prathit and uh this do you see either odd news or zoom uh turning into uh mozilla or a gnu kind of project in the world recognition you know see this is a good question to ask but not just of boom or odd news or any other fact-checking service you know there's someone mentioned open source right uh i mean there is also open source not just of technology but also open source of uh you know open sourcing resources or crowd sourcing uh in other words um you ask the question is there is there an opportunity for us to turn it to a mozilla kind of thing i would say there is a potential opportunity though i don't know how far fetched it is for us to turn it to a wikipedia not a mozilla because you know uh wikipedia is probably not what last time i checked three or four years ago it had 20 employees and they're pretty robust fact-checked there right yeah i had to deface a wikipedia page yeah but look at this uh if they have i don't know how many employees they have right now but let's assume the number is below 100 and they are sustained by something like 170,000 or 200,000 volunteers around the world but out of that 200,000 volunteers about 80 to 70 to 80 percent of fact checks are not fact checks wikipedia articles are edited by just 1 percent of that 200,000 yeah right so what then would happen you know a boom would need if we had you know 20 employees we would need a cat i mean let's just count both odd news and boom and you know other fact checkers there are about 100 people in india doing this we need a 200,000 people of which only 10,000 people ultimately will actually be an active source so uh that that is my answer to your question which is we need to get everybody involved unless we get everybody involved there is no easy way out unless you know it's like you know what are the worlds you know that edgy wells uh novel and that movie that you suddenly have a threat and the entire planet is being uh terrorized and the the aliens just die or start dying on their own because they cannot take limit unless you have some kind of uh you know abrupt end to this crisis we need to get everybody involved nothing you want to climb in here yeah so uh first of all you did that read with me about wikipedia defacement yeah go and check the alt news history the wikipedia history or my father also has a wikipedia page and i keep you know it gets defaced somebody anyway uh but uh yeah wikipedia i mean one very recently this what's his athish to see if he wrote a uh piece before 2019 elections against mr modi and immediately his wikipedia page was defaced to claim that he works for congress that scheme was circulated everywhere uh and so no wikipedia get you know the thing is that the defacement does not have to last for too long all you have to do is deface it for and it allows wikipedia allows anonymous edit so all you have to do is deface it for those two seconds where you can take a screenshot i mean you don't even have to do that right you can you can just edit the html in you know in place but not many people know how to do that uh but you know what they do is they go and deface it for two seconds take a screenshot and then secrete it so anyway but uh coming to the core question number one uh i agree with two so there are two things number one yes uh alt news will be open source thing all the tools that it is making uh you know including this we will be releasing it i'm not trying to what license whether it will be lgpl or mit we have not uh decided on on what license we are going to use uh but we will be open sourcing it now alt news uh articles uh are already under creative commons so we already have a permissive license anybody can use it to the point of uh dataset uh all the factors use the claim so in case of alt news it is a json uh a json block in the code which can be scraped uh so you already have a limited database there but it does not you know you won't have what image or what video it is there because because the schema.org uh json structure does not have a field to point to a image of video which is related to the fact that one schema.org has that we will update so that automatically becomes an open source database especially if the factors are under a creative commons license that is number one and secondly uh i'll share my screen real quick uh so you know we have something called seed data here and so these are the articles that go in you know these are the factors that we uh ingest and eventually we are looking at automating the ingestion process as well but this will also ingest uh eventually ingest uh the json data and uh we are looking at uh sort of uh uh uh api that is protected with some kind of authentication it is not going to be open to all it is going to be per request space it's not going to happen right now i think it's it will probably take another uh at least another year for us to make sure that this is accessed to a safety but uh per request basis we are looking at giving access to academics and you know and researchers uh to this data okay uh crispy i think we should wrap up in three minutes so it might be dinner time for everybody but we've taken a lot of the questions and put them on the comment section of the event page so the researchers can respond later one more thing and the wikipedia model that Venkatesh was talking about this especially uh will work for translations where we can you know once a factor is done in one language the wikipedia model can can be used to do tactics in multiple languages and that is the problem that we are facing somebody has a question about reaching the ground the problem is all of us are right in English or in Hindi but uh but you know it is so many other languages that that it needs to be translated to and that is where especially where the wikipedia model will be extremely useful absolutely thanks nothing okay uh so since we're out of time i'm going to ask one last question which i will also use as you're concluding in the map uh from each of the panelists it's going to be the same question which actually somebody has asked to one person but i thought it a valid question to everybody uh where does the human element come in um if you're a fact checker i'm sure you're being exposed to a lot of toxic stuff if you're developing tools i'm sure anybody in this ecosystem is uh is uh i think uh subject to a lot of abuse right verbal abuse maybe and then there must be a human element to that so i'll go one by one and maybe use this as a closing question so uh let's back with Denny uh sorry could you repeat that question yeah the human element of uh fake news in fact check you know there could be negative parts of it right like threats or being exposed to toxic stuff or maybe positive parts of it also like people look up to you and thank you etc in your opinion what is the human element of this entire thing oh i mean uh i guess just being able to inform somebody offer you know a new thing uh like when we talk of our service we we don't say that okay this will tell you whether something's true or false we kind of say that this is going to give you more information to add nuances to the issue you know so i think uh just building that level of health is skepticism and curiosity in people like this is you know uh the positive human aspect of you know doing okay ana would you like to comment um yeah i think in terms of i mean in terms of journalists the human element is i mean at least from traditional media it's always there right from the point of fact checking to obviously to putting that news out and also facing the tirade that comes our way uh that's i think part of our job irrespective of whether it's uh fake news or not um i think obviously in terms of the the challenge like i mentioned earlier was the challenge of reaching that man on the street um that is something that we will always struggle with in this you know in in this sort of landscape so i think that is the major human element that we need to sort of conquer okay thank you uh by the way this question was asked by by somebody called Pratik so i will ask Pratik to answer this last uh Sandeep so uh i would look at the human dimension in the fertile ground for fake news the fertile ground for fake news to spread is an eco chamber where there is social homophily a lot of people with like-minded biases they get into an eco chamber and then that is a very fertile ground for fake news to go which finds acceptance much easily because there is a confirmation bias of what beliefs they already have and there is something known as information avoidance where fresh information which challenges their existing beliefs they tend to underplay or ignore it even though it is a fact so the human element involved is in that uh being in that silo and being in that uh eco chamber where you just want to get your own voice reflected and not open to new views great thanks uh thanks Sandeep i'll go over to Venkatesh Venkatesh your comments on this as as a concluding comment i mean for me the human element uh doesn't bring the mind fact checking so much as what i'm increasingly preoccupied with which is media literacy or educating everybody you know so just to use a if you permit me a corona virus analogy uh it's not uh it's not a antibiotic that i mean we need an antibiotic or an antiviral but we actually also need a vaccine so i believe that the human element for me is uh getting into why people share misinformation and educating them on why they should not share share misinformation or believe this which is a huge amount of us and this question has been asked by a namesake yeah so uh completely agree with Venkatesh that you know there was an earlier question how can fact checks teach everybody it is never going to do and the the expectation that uh that fact checking is going to solve the issue of misinformation uh uh is not the right one what we need is an educational approach to this and that is that is what what is going to help as far as uh uh you know trawling et cetera it's concerned that is now anybody who's on social media and anybody especially fact checking which is uh has such a political aspect in most of the misinformation is has a political nature there is uh you know there's going to be a lot of hate a lot of trawling and it is something that we have to we have to train ourselves to deal with i don't think it's the ideal scenario but that is how it is going to be thanks thanks Pratik uh thanks everybody this for i for for sorry for lack of time i think i will have to conclude it here thanks to all panelists here uh thanks to uh denny thanks to thanks to thank you thanks to Pratik thanks to Venkatesh for uh very enlightening news here i myself had a lot of fun and i learned a lot about perfecting the ecosystem today uh thanks to Zainab for hosting this and uh thanks everybody at hasi thanks a lot Zainab uh do you have any last words to say nothing we will uh all questions unanswered we put them on the comment section of the event page so we request for the speakers to respond accordingly and we can continue the conversation there on that note i think everybody can go get dinner and good night and goodbye have a good Monday evening thanks everyone see you bye bye