 Hi, good afternoon. My name is Erin Gallagher and I'm a multimedia artist, a translator, and a writer with Revolution News. I've been blogging about the Ayotzinapa protests last September 26th. 43 normalista students were forcibly disappeared in Iguale Guerrero. And in the course of that reporting, I kept running into this weird software. So I started reporting on that as well. I had hoped to present here today with Alberto Escorcia from Lo Que Sigue in Mexico City. He couldn't make the trip, but he has helped me tremendously with researching his topic. And he is really the expert with bots in Mexico. Alberto is using a variety of open source tools to detect bots in Mexican Twitter. I have a basic idea of how his software works. And I'm going to show you several graphics and videos that Alberto has made demonstrating the bots that they're finding in Mexican Twitter. And really, I've come here today with more questions than I have answers. There's also just a quick warning. There's some bloody pictures in my slides. So if you need a trigger warning, this is it. A 2014 survey by the competitive intelligence unit in Mexico found that 96% of Mexican citizens between the ages of 20 and 30 are using social media. The mainstream media in Mexico is quite bad. So they use social media to spread alternative news. They also use social media to organize both digital protests as well as street marches. And they're using Twitter in some rather innovative ways, creating networks of people that ultimately help benefit society. Bots have been active in Mexico since 2012. The current president, Enrique Peña Nieto, used bots to spread political propaganda during his 2012 presidential campaign. They've been known as Peña Bots ever since. Initially, they were used only for spreading political propaganda. In October of 2012, they started attacking hashtags. Since last November 2014, the bot attacks have escalated dramatically. And as of December 5th, Alberto has counted 75,000 automated bot accounts in Mexican Twitter. And he estimates that since last December, there are many more. This first video is a recent hashtag attack. A Twitter user recorded their screen while it was going on. And it doesn't look like much is really going on here. But these bots are censoring a trend denouncing nepotism. The niece of Enrique Peña Nieto, Sobrina means niece, and EPN is the president of Mexico. His niece recently got a job at Pemex, Mexican state-owned oil and gas company. So the bots are doing two things here. They're filling the trend with spam, which is a sensor in itself. It drowns out the real conversation with spammy repeater tweets. They're also sort of gaming the Twitter algorithm by altering the novelty index. So they're regulating the speed of the hashtag, which we believe triggers some sort of spam mechanism causing the hashtag to be kicked out of the top 10 trending topics. So the hashtag gets bounced out of the trending bar. So I've observed these bots doing a variety of activities, and I break them down into five categories. Hashtag spamming. I've also seen it referred to as hashtag poisoning, creating fake trends, spreading smear campaigns, sending death threats, and spreading political propaganda. The first time I ran into a live bot attack was last December 1st, during the 1DMX protests in Mexico City. This was a protest for Yotinapa. There were large street marches in 60 Mexican cities, as well as international solidarity actions. Also, protesters in Oaxaca shot down the airport. You can see they also redecorated a little bit. So the march in Mexico City was massive, and it was entirely peaceful all day long. In the evening, towards the end of the march, a hashtag rompe el miedo was activated. Rompe el miedo means break the fear. And rompe el miedo is a network of media workers, so journalists, photographers, activists, and human rights defenders, who use this hashtag to document and aggregate human rights abuses during protests. When the hashtag first started trending, this map was one of the first things that was tweeted. It says, la zona roja repleta de granaderos. The red zone is full of grenadiers. Evita la zona. Avoid the zone. So this map was a warning to tell protesters to avoid that red zone and safely exit the protest area. Bot centered the hashtag maybe 20 to 30 minutes later, and the trend was full of spam. For myself, I was reporting remotely in the safety of my own home. I was in contact privately with an activist who was on the ground with human rights defenders, but for the people on the ground, they were not receiving these safety notifications. Arbitrary arrests began shortly after. Protesters were cuddled. These men in white are human rights defenders. They formed a wall to protect some protesters from the police. This woman was not involved in the protests. She was a 43-year-old mother who was beaten by the police. Her 19-year-old son was with her, and he jumped in to protect his mom, and he got arrested. Activists were beaten. Protesters were beaten. And the end of the 1DMX part, protests ended in brutal police repression. It's possible that these folks still would have been beaten regardless if they were getting the notifications or not, but clearly this was putting them in real danger not being able to access this hashtag. So the next hashtag I'm going to talk about, I'm going to backtrack a little bit from Rompel Miedo. Maybe you've heard about this. It means I'm tired. Jamé Cancé is considered the most powerful hashtag in Mexico. It trended for over 27 days straight at a constant speed, and amassed over 4 million tweets before bots took it down. After Jamé Cancé was overthrown by bots, tweeters started using Jamé Cancé 1, Jamé Cancé 2. They kept adding a number 1 after Jamé Cancé in order to keep the hashtag trending while avoiding the bots. And Jamé Cancé had in total 34 iterations, 23 of them made it into the trends bar, so 23 Jamé Cancé's were top 10 trending topics. Locasigue, Alberto from Locasigue, recorded this short video, and he wanted me to put a disclaimer at the beginning of this. This is a software called Flocker. Oops, let me pause. It's a very short video. This software is not a precise tool. Alberto describes his process to me. He uses Flocker initially in order to detect if bots are present. Some of these accounts are not all bots, and I believe what he had told me was that Flocker only shows bots that are interacting with other accounts. So if the bots are not interacting with anyone, they're not going to appear in Flocker. But this here was a 22-second video, and you'll see, I'll play it twice if it goes too fast, the bots are on the outside. So this sort of demonstrates the effect of what happens when they're spamming a hashtag. This is a graphic Alberto created using a software called Gelfi. This is a visual representation of people tweeting into Jamé Cancé. The pink and purple are real people tweeting, and the weird formations in the corner, the greenish blue and beige, those are bots. Alberto calls those tumors. He has a very long video. It's not very long. It's about 10 minutes. This is the interface of Gelfi. So this is what Jamé Cancé looks like in Gelfi, and he's zooming in here on one of the tumors, and you can see he's pulling apart the bots. This software, as far as I know, is not very user-friendly. I don't know how to use it. But this is one of the main tools that he uses to detect bots, and it's very precise. And I just wanted to show you the interface to see what it looks like. So the second activity that I've seen bots do is creating fake trends. They're doing this on a daily basis now in Mexico. Alberto says they're averaging two to three fake trends per day. And the fake trends, basically they make these absurd fake trends that overpower real trends. I covered a protest in February. This happened in Acapulco, Mexico. There was a set de guistas from the Mexican teachers union, said they had a protest in Acapulco on February 25th. The police beat the crap out of everyone. The injuries were brutal, and the pictures went viral. Later on in the evening, a Mexican journalist, Pati Peñalosa, denounced that she thought bots were censoring Acapulco. There were two other trends in the top two slots that night. Soya Montede and Donde Firmo Para. Alberto ran a comparison in Topsy, and you can see that despite the Acapulco has over 300,000 tweets, Soya Montede and Donde Firmo Para are in the number one and two slots. Soya Montede means I am a lover of, and then you sort of fill in the blank with what you love, and Donde Firmo Para means where do I sign for this, or where do I sign for fill in the blank? These two fake trends pushed Acapulco down to the 10th position and eventually bounced it out of the trending bar. This is another visualization using Flocker. So this is what the real humans tweeting into the Acapulco hashtag looked like. And this is the Soya Montede trend. So these are bots. And the bots always look like this. Humans make a massive network. They connect. The bots never make a network. The bots that created Soya Montede were not tweeting spammy repeater tweets. I'm sorry if you can't quite read that well, it's kind of small, but so they tweeted sort of the same words only in different arrangements. So there was a repetitive feeling to it, but they weren't just spammy repeater tweets like the regular hashtag spamming. This activity is not as obvious as hashtag spamming. If you don't speak Spanish, you might not realize that they're repeating certain words in this, but you can see here, we're here at Valerie, or we'll look at Lucia. She says, La verdad, Soya Montede, dormir juntos me gusta mucho. The one below her, Soya Montede, dormir a su lado me gusta mucho. Soya Montede, ir de viaje me gusta mucho. So they're using the same words and just flipping around the arrangements. Bots are also spreading smear campaigns. The campaigns are usually against students and teachers, also against journalists, certain media outlets. The smears can bleed into the mainstream media narratives. Alberto estimates at least one smear campaign per week is spread by bots. And some of the most notable campaigns were run against John Ackerman, Lydia Cacho, San Juan Martínez, and Rosana Regillo. And I'm going to talk in a moment about what happened to Rosana. The setegistas, again, had a protest. This protest was in Chilpan-Singo. And this picture was tweeted. And I don't know who Daniel Gomez is. I'm picking on him at random because he was one of hundreds of accounts that tweeted this picture. The tweet says, teachers from the seteg throw rocks at riot police in Chilpan-Singo. This protester in the photo is what they refer to in Mexico as encapuchado. So encapuchado is someone who wears a hood. These are people who mask up to go to protests. The encapuchados are frequent targets of smear campaigns. They're usually seen as the troublemakers at protests, also anarchistas occasionally. So I don't know which tweet was first sending this picture out, but a deluge of tweets followed. And you can see this is not organic activity. This is a fraction of the tweets that were sent out using this picture. It was almost kind of comical. So these were smearing the setegistas saying that they were attacking the riot police with rocks and also spamming the Chilpan-Singo hashtag. Another smear campaign was run in February. At the end of February, there was a protest in Mexico City. Protesters painted some graffiti on the Angel de la Independencia. This is the main monument in the middle of Mexico City. And Mexican artists tweeted a translation for us, eternal hatred for the exploiters. Bots began to tweet about the graffiti on the monument. They used a similar technique to the fake bot trend in Soymante de, using the same words over and over again in different arrangements. Although they didn't tweet a hashtag in their tweets this time. This trend was also directed at encapuchados. They were protesters who were wearing masks that painted the graffiti. Excuse me. And these tweets have a very sort of indignant tone to them. I translated a few of them. Bad people at the Angel. They are encapuchados. It will end badly. Bad people at the Angel. They are encapuchados. They gain nothing by painting it. Careful at the Angel. They are encapuchados. It's all painted. So you can tell the tone of these tweets was rather condescending. The encapuchados graffiti made the news the next day. Two people were arrested. Bots are also spreading death threat campaigns. In February 2015, Mexican academic and blogger Rosana Regillo started receiving death threats on Twitter. The threats continued for over two months and spilled out of sight of Twitter into other social media platforms and personal emails. She received tweets, everything from memes to pictures of burned bodies, indicating what was going to happen to her for supporting the Ayotzinapa protests. And initially, Rosana thought there were only five accounts attacking her. After analysis was run, it was discovered there were actually massive botnets supporting the attacks against her. This is another graphic from Gelfi. And Rosana, you can see, is the green target in the middle. This shows the last 200 interactions of attacks directed at Rosana. So the number of bots that participated in this attack is unknown, but it was a massive attack. Finally, spreading political propaganda is probably the most obvious activity we would expect bots to do. They've been spreading political propaganda in Mexico since 2012. In May, digital news outlets, seen embargo, got a copy of an agreement with a company called Agaves Digital. The agreement was for Agaves to procure an army of Twitter trolls for the Green Party in Mexico to support the June 7th elections. Instructions were included on how to position an organic hashtag as a trending topic. So, Agaves, their website describes them as an online media agency specializing in performance strategies through search engine marketing, social media, and web analytics consulting. Since they have eight years' experience in the internet industry and have worked with various big clients such as Yahoo and Google. Agaves has currently 47 contracts that date back to 2012 with the Mexican federal government. These are listed in the Mexican Transparency Portal. Some of them are current, some of them have expired. Let's see if this is going to go online. So this is the Mexican Transparency Portal with the federal government. This one is a recent 2013. They don't give a whole lot of description. They tell the amounts what the contract was for, the dates that it was valid for, and just a brief description. This one is in effect until December 2013. And also the department. This one is, I believe this is the department of, sorry, I forget what the title is on this one. But anyway, they're all listed in there. Mexicans are very familiar with this company. And just to be clear, I'm sure there are many companies in Mexico and around the world that offer these services. And I'm sure that Agaves also offers standard media services and promotional services that any media company would offer. So, as far as I'm aware, this activity is not illegal. But this is information warfare. And it's sort of invisible. It's hard to measure the damage to Mexican society and free speech. But social media is the new public square. Mexicans are relying on these networks to get their news out to the world and to communicate with each other. So when these networks are manipulated, it is extremely damaging in Mexico. Human voices can't compete with this software. It's like playing the computer. The bots always win. In Mexico, a hashtag is not just a hashtag. It's a way to make the invisible visible. So two of the big questions that we have, who is operating the bots? And can the bots be stopped? Are there any questions? Thank you. Thanks a lot, Erin. We have about four minutes left for questions. So if there are any questions, please come up to one of the microphones that you can see standing. And I see there's someone standing on the left microphone. Please go ahead and ask. Yeah. Did you have any feedback from Twitter, whether they did anything to filter out all these bots? On the one hand, on the other hand, as you are able to visualize what is bot entries and what is real human interaction, wouldn't it be a nice idea to come up with some kind of Twitter browser where you can exclude the bot part to see the real meat of what's happening on Twitter for real? Okay. I think I understood your first question was, has Twitter done anything about this? Yeah. I don't know. They may be working on it, for all I know. They're certainly aware of it, because this activity happens in many countries. I also submitted a support ticket myself reporting an app, but I've never heard back on it. I believe they're aware of it. I hope they're working on it. The second part of the question was... Well, you seem to be able to detect the patterns which define bots and which are definitely not humans. So that should be an important first step to find a way to have a Twitter visualization where you don't have any bot interaction. That would be fantastic. I'm not sure how to do it myself, but that would be great if something like that was possible. Hello. Hi. For my experience, this kind of bot of network are controlled by Twitter application. So every book subscribes to a specific Twitter application and this application can behave in the Twitter name. This is a metadata of the tweet. So you can analyze which application is sending the tweet or controlling the retweet. And that's going to be a pattern to spot those networks. I'm sorry, I didn't quite understand the question. Sorry about that. Twitter application, you know what they are, is a feature for Twitter developers to make behave other Twitter users how do you want. Make write them tweets or make retweet, make favorite. I'm very sorry, it's the language. I apologize. Come find me afterwards. We will work that out. Sorry about that. So we have time for one last question. I see someone queuing up at the right microphone. Please go ahead. Yes, I would be interested in how easy it would be to transform this analysis to different countries. You already said that this is happening in other countries as well. And I think it would be interesting to have a global monitor, some kind of network to see where this is happening. You said you think it would be interesting to have a global network to find out where else this is being used? Yeah. Yeah, that would be amazing. That would be fantastic. I don't know how to do it myself. Again, I'm Alberto May and I think that's actually something that he would be very interested in doing as well. He's been working a lot on different ways to connect different social movements. So that would be right in that category. So thanks a lot, Erin Gallagher, for informing us about this. Thank you.