This Parlamerica's podcast is part of a three-episode series on raising awareness to address the fake news challenge. Nicholas Monaco is a digital propaganda researcher and computational linguist at the Institute for the Future's Digital Intelligence Lab. In this presentation, Mr. Monaco provides insight on the challenges that bots and computational propaganda present for democratic processes and political representation, providing examples of their use during the Brexit referendum and the last presidential campaign in the United States. He also shares actions and initiatives that can curb these negative impacts.
This presentation was recorded during the 14th Plenary Assembly of Parlamericas, Parliamentary Actions to Promote Responsible Political Discourse, held in Medellin, Colombia on November 16th and 17th of 2017. Gracias por haberme invitado. I'm happy to be here. It's my first time in Colombia. And yeah, I'm here today to talk to you about computational propaganda.
So I'm representing ComProp, the Computational Propaganda Project at the Oxford Internet Institute. We've been working on bots, specifically in computational propaganda at large for a number of years now. And I'm also representing the Digital Intelligence Lab, the Digintel Lab Institute for the Future in Palo Alto, California, where I'm also a researcher. And at the Digintel Lab, what we're interested in doing is kind of building a bridge between all these elements of society that are interested in these topics. So civil society, but also academics and researchers, private sector people, and legislators like yourselves. We're interested in working out concrete solutions to these problems, as well as turning out quality information, quality research to try to help inform the public, but also people from all parties, journalists, legislators.
on the problem so that we can work towards some concrete solutions. And 10 minutes isn't a lot of time to get into everything, so I'm happy to talk to you all after the talk. And my contact info is here as well, if you want to get in contact with me. I'd welcome that. So just a quick roadmap of where we're going to go. First, I want to talk about what computational propaganda is. Then I want to talk about some brief examples from the last year, since it's been really a hot spot of news information. And it's been a peak period of activity for bots and computational propaganda. And then, of course, since you're all legislators, you're not only concerned that there is a problem, but you're concerned with solutions. I wanted to talk about a few ideas I have about solutions.
So this is a very sensitive clicker. What is computational propaganda? I think it can simply be defined as the malicious use of software to megaphone or dampen political messages online. And the goal, of course, is manipulation of public opinion. And one of the great takeaways I'd like you to take away today is that this is not only a computational problem, but it's a very human problem as well. And correspondingly, solutions must be both computational and social in nature.
In addition to that, it's a very international problem. ComProp has been looking into this again for at least four or five years, and we've seen the presence of the same problem in a real diversity of countries. So Azerbaijan, China, Taiwan, the US, UK, all around South and Central America, Kenya. Really, governance and polity are secondary to the story. Both autocracies and democracies are experiencing these same computational propaganda problems. And it's also cross-platform. You hear a lot about Twitter and Facebook, and I think rightfully so. Twitter especially is kind of a hotbed for this kind of activity. But there are bots as well on YouTube, on Instagram, and all the major platforms.
I think I'm here today mainly to talk about bots, social media bots, and talk a bit about what they are and what they can do. So social media bots are simply code that control profiles and poses humans online. So they're computer programs that poses humans online. And I know it's not extremely transparent to understand how that works, but it's not all that different from how humans access the internet. So as a human on your computer to access the internet, you open a browser, maybe Chrome, Firefox, Internet Explorer, you type in a URL, and you communicate with another computer. That computer sends you a file, essentially, which your browser renders and reads. You interact with it, and if you want to send some info back through the website, you use your browser to do the same thing. Essentially, the same thing happens with computer programs. They use something called an API on these big platforms. An API stands for application.
programming interface. And a simple way to think about it is like a browser for a computer code. Once your computer code is connected to the website, it can do all the things that a human can do on the website. If we're talking Twitter, it can follow, it can like, it can retweet, it can even interact with users in advanced cases. So that's kind of the backbone, the technological backbone of how these things work. And I'm happy to get more into that if anyone's interested.
Another quick nuance point I want to make is just about misinformation versus disinformation. In English, they're kind of used interchangeably in the media. I don't know if it's the same in Spanish, but Data and Society, a really great NGO out of New York, released this paper called Lexicon of Lies over the summer, which digs into kind of the granular details, the differences between all these terms. And essentially, the idea behind misinformation is unknowingly sharing false information. So if I retweet a news story that's fake, but I don't know it's fake and I have no malevolent intent, that's misinformation. If I knowingly make a fake story with intent to spread it and disseminate it with a political goal in mind, that's disinformation. And again, I definitely refer you to...
to Lexicon of Lies if you're interested in that kind of analysis. So what can bots do? I think one of the main things is the idea of megaphoning. And you've probably heard about this when you've read about it. I'm sure you're all familiar with Twitter and the fact that Twitter has hashtags. And it also has trending topics, which are essentially the most popular trending hashtags in any given area at any given time. What bots can do is tweet a certain hashtag.
10,000 times or however many times they want and essentially manufacture their own trends. We call this manufacturing consensus. It's basically making false grassroots support for a hashtag. But this gets really interesting when it's a political hashtag. And this goes both ways. You can manufacture consensus for, say, make America great again for a cause that you're promoting for your own benefit. But you can also dampen other hashtags. So if I'm Donald Trump and I see a Bernie Sanders hashtag,
trending and I don't like that. I can make my boss tweet something about me or something completely irrelevant just to push Bernie out of the news feed so people won't see it. And again, we call this megaphone or we call this dampening. And hashtag hijacking and poisoning are similar. A great example of this is Syria. The Free Syria Electronic Army was tweeting and organizing under hashtag Syria, quite logically, in 2011. And quite effectively so. Pro-Assad forces got onto this fact and started tweeting cat pictures and soccer matches and irrelevant links with hashtag Syria, which really undermined the effectiveness of the protest hashtag. So these are just a few of the things you can use bots to do and that we've seen bots use to do.
You can also use bots to promote transparency and protest. You can use them to harass people. This is a particularly pernicious problem. We're seeing a lot of state-sponsored trolling going on around the world. I'm writing a paper on that with Carly Nist, one of my colleagues that should be coming out shortly.
Just a quick statistic to give you an idea. There was a journalist in Turkey who covered the Gezi Park protests in 2013, which were quite a hot issue. And within minutes, she started receiving thousands of rape and death threats on Twitter. Obviously, for a human, that would be very, very hard to do. If you use bots, it's very trivial to program them to send out these messages. So that's a very concerning thing, but we don't have a lot of time to talk about it today. Another quick point is crawlers. You can use bots to gather intelligence or...
Operate chat rooms, use Google, use Wikipedia. They're an infrastructural part of the Internet. So a blanket ban on bots, for example, as a policy would not be effective because it would kind of destroy the Internet as we know it right now. There are good bots and infrastructural bots that are used for maintenance purposes. And I just wanted to talk very quickly about some quick examples of computational propaganda. This comes straight out of ComProp. Dr. Phil Howard and Bensei Colonyi wrote a paper on Brexit on Twitter.
Basically, they gathered data for a month leading up to the referendum, and they found that bots played a small but strategic role in the referendum. Bots were deployed on both sides critically, which is something that's seen a lot. Dr. Howard also published a paper on Venezuela in 2015 that actually showed opposition bots were more frequent than pro-government bots. But normally, both sides use bots. It's not a one-sided issue. But interestingly, the pro-Brexit bots dominated.
And I think one of the most salient interesting findings is that less than 1% of the sample accounts in all of the data were generating almost a third of all the messages. So you can see this is a really disproportionate amount of conversation coming from less than 1% of accounts. And what this does is it really challenges the idea of a marketplace of ideas. The best ideas can't rise to the top if bots are tweeting more loudly and more frequently than organic human users. Onto the U.S. presidential election. Before the election got even heated or there were nominees on both sides in the Nevada primary already, there were Latino bots tweeting support for Trump. And again, it's this idea of manufacturing consensus. Trump obviously has said very racist, very incendiary remarks about the Latino demographic many, many times and did during the primary. But he pointed to this and said, well, look, people really like me. Latinos love me. Look at all these Latinos on Twitter tweeting for me when in fact they were just bots. Another interesting statistic is during the debates, there were bots on both sides again.
Trump bots outnumbered Clinton bots by ratio of 4 to 1. Then by election day, it widened to 5 to 1. So there were five times as many pro-Trump bots as pro-Clinton bots. And some interesting analysis here from another researcher, Emilio Ferrara. He found that there were several pro-Trump bot accounts that went dark right after the election on November 9th and showed back up again tweeting for Marine Le Pen, the right-wing candidate, in the French election six months later.
So moving on to solutions, I just wrote a piece with my colleague Sam Woolley, who's the director of the Digital Intelligence Lab, and we dig deeper into these ideas, so I'd encourage you to read it. It's in the RIP-ON forum. But basically, blue-ribbon commissions of academic experts, private sector executives, and legislators are one of our top ideas. I think the idea of getting everyone in a room, getting everyone having long-term conversations, can help us all see the problem from different angles. We as researchers don't have the knowledge to codify these things into law or proper legislation as you do, but by the same token, the private sector doesn't have insights into how pernicious social media can be for democracy, and every side has their shortcomings.
Data sharing among private companies is another thing that would help. If we know that propaganda campaigns have taken place in the past, and we have that data, we should encourage it to be shared between big companies, but also between researchers. And this can help inform efforts to counteract this in the future and be in a better position to fight against it. These propositions are kind of being proposed right now in the U.S., more transparency in political advertising, which I think will help quite a bit, but also automation, making bots announce that they are bots on Twitter. Announce that they are bots, excuse me. I think that would go a long way to solving the problem, contributing to helping the problem, without impinging on free speech at all.
And again, I'd like to encourage you to remember that solutions should be both human and computational in nature. I think a hybrid of these two things would be most effective. I just wanted to close quick with a quote. This is a famous Colombian hacker who came out in 2016 and talked about working on digital campaigns, nefarious campaigns behind the scenes all over South and Central America. His name is Andres Sepulver, and he said, when I realized that people believe what the internet says more than reality, I discovered I had the power to make people believe almost anything. I think this has changed to an extent. People are aware that there's computational propaganda, but not a lot of people know what we should do about it, or what they should do about it personally. I think this quote encapsulates the problems, but also the solutions. So what can we do? We can change what people see, and we can change how they react to it. By what people see, content...
You can obviously limit content online. I don't think that's a preferable option. I think you'd all probably not refer to do that. But making transparent political ads, making transparent bots, those are ways to change things that people see without infringing on rights. And changing how they react to it, promoting digital literacy, promoting media literacy, those are more long-term goals. But I do think they make a difference. And for instance, in Taiwan, the digital minister, Audrey Tang, is currently implementing an education program that is encouraging digital literacy in elementary schools and schools as a solution to this problem. So thank you.