pol.is in Taiwan
better public discourse through artificial intelligence → more informed public servants → better laws
This video, and subsequent transcript, were presented by @colinmegill at the 2016 g0v conference in Taipei, Taiwan on the 14th of May, 2016. g0v (gov zero) is akin to Code for America. Many slides (obscured in the video) are presented below inline. The full deck can be found here: g0v2016_final.surge.sh
My name is Colin Megill. I am a founder of a startup based in Seattle, in the United States, called pol.is. We are focused on creating a tool that leverages machine learning to scale up online discussion.
A couple years ago, Minister Jaclyn Tsai, who is a Minister in the Executive Branch here in Taiwan, went to a g0v hackathon and issued the following challenge. She said, “We need a platform to allow the entire society to engage in rational discussion.”
This is a really big question, and I had, to be honest, a number of slides explaining the problem that she’s talking about here, and also which we have been working on at pol.is for four years.
I thought instead I would just show you this, because all of us together have gone through the process of trying to get an idea across to people online; and I think … I could explain, but … you probably know what the robot does, I will add that not only does it bang its head against the keyboard, it also rolls its head side to side.
I think that the fact that it has 34,000 retweets speaks to the fact that the problem that I’m addressing in this talk, and the problem that Minister Tsai brought to the g0v hackathon, is not a mystery to any of us. We’re all familiar, intimately familiar, with arguing on the internet. That’s what we’re going to talk about today.
The way that we argue doesn’t scale. It’s inefficient; it’s slow; it has a rather boring plodding feel to it. All across the United States, and also across the world, with notable exceptions like Podemos in Spain — most of the world doesn’t do democratic decision making, democratic discussion at scale, very well.
Let us take a look back for a moment, at a time when we did democratic decision making really well, which is classical antiquity. This is the polis, classical Athens. At the peak of Athenian democracy, classical Athens looked something like this. If you look at the middle of this city, you’ll see that there’s a triangle there, and that triangle is a marketplace. It is the agora. There was somewhere between at the time 30–60,000 eligible citizens who could participate in the democratic process, which only included people who owned land and were male.
With that notable flaw, the people were able to move about freely through the city, and they were able to congregate, to move through, buying goods, talking with each other, discussing; and out of this extraordinary time in human history, which we count as part of the foundations of the democracy and the ideas of democracy, we get Aristotle and Plato and all of the ideas of public discourse and rhetoric that we still study today.
Once every week, at the peak of Athenian democracy, somewhere around 6,000 people, which was about 10–20% of the eligible citizens, would walk up the hill from that marketplace, up to that little half moon you can see on the right there, and that’s Pnyx, which is a hill above the city.
When they got there, these 6,000 people would have had a week of discussion to think about, to talk about, to contemplate each other’s ideas. Whether you agree with them or not, whether you agree with your fellow citizens or not, you at least heard the ideas somewhere in your daily life. Conversation was woven into the fabric of daily life, and there’s something in that which I think we can learn from.
Let’s talk about the time in which, now, sometimes, we get the entire society showing up now, as happens there, right. We have the Sunflower Revolution, Occupy Parliament, and we know that the government would love to not have conversations like this. They would love to not do qualitative research in this particular way, right? We’d love to have some kind of a process that isn’t a huge spike in activity that deals a massive disruption to society. Though these can be productive, as we’ve heard from other speakers, they’re often unpredictable … They’re always unpredictable, and often, things can go badly.
There’s one interesting note, as well. If we’re talking about large-scale conversations in modern democracy, can’t we just analyze social media? Isn’t this already happening? Aren’t people talking about what they want to talk about online, all the time?
There are a few problems here. The following images were produced by Emma Pierson, who is a Stanford researcher and Rhodes scholar; and she’s created an image here which gives us a conversation on Twitter. It is about abortion, which is a very controversial issue in the United States, very politically charged; and you can see that someone in the upper right-hand corner right there, in the feminist community, has sent out a comment with the hashtag #shoutyourabortion. Someone in the conservative community has started a different hashtag, which was #shoutyouradoption.
We can see here that there are multiple groups that emerge, and this is an analysis that is replicable on other Twitter conversations as well; and the reason that this happens is called homophily. There is a plain English definition for this, which is, “Birds of a feather flock together.” The idea is that people follow each other. That’s what social media is all about. It’s about discovery.
What this means at a practical standpoint, later on, is that we are broadcasting ideas from people we agree with. We know we agree with them, and we know we have something in common with them because we follow them; and we’re then broadcasting the idea that we get from them to people that agree with us. This is the model. We don’t have to guess about this; we can see it. This is the filter bubble quantified. Social media is not a place where we’re talking to each other; we’re talking at each other, and we’re talking around each other, and sometimes we’re not talking to each other at all. We’re just ignoring each other, because it’s easier.
I want to tell you about a time that arguing on the internet went well. I want to talk to you about the implications for our democracies, and I want to talk to you about how it’s inspired our team to build a better tool. This is the story of pol.is in Taiwan.
The first conversation that ever happened was started by clkao, who’s sitting right there. It was started about a JavaScript conference called JSDC, right? Okay. I’m going to try to get this right.
A Chinese company … Speakers from a Chinese company had been invited to speak at a JavaScript conference in Taiwan. This was very controversial, and I think it was a little more controversial, I’m not sure, but I believe that TonyQ is a bit of an instigator. TonyQ started a thread, and they were about 200 comments in the thread. It was a bit of a flame war. At the end of this 200 comment flame war, clkao posted a pol.is conversation, and was like, “Hey, let’s try this thing out. Let’s see if this helps.”
What did everybody do? Everybody could come in and write comments, just like they could on social media. This part of pol.is is not different, but in pol.is, you cannot reply. The only way to engage with other people, what their ideas were, is to agree and disagree and pass.
If you think about it, sitting in a stadium of 100,000 people if you can talk directly to someone across, and then someone else can talk to you, this information structure just breaks immediately. I think trolls … It kind of helped us realize how broken this information structure is. Basically, replies, at scale, don’t work. We did away with replies. That’s part of the core, that’s part of the foundation.
Doing away with replies gets you something very special. It gets you a matrix. You can have a look at what the data inside pol.is is like; it’s not complicated. It’s every participant, and what they thought about every comment. You can look for a participant and see how they agree, they agree, if they disagree, they disagree, they pass; or for a comment, you can see, this person agreed, this person disagreed, this person agreed, this person disagreed.
Humans aren’t very good at analyzing this; but this scales up, and machines are really good at analyzing this. You use this all the time. You may not realize this, but you do. This is what Netflix would see — every movie, and every viewer. Every time you rate a movie, every time you buy a product, you’re creating data; and we do machine learning on that data in pol.is like Netflix would do on movies. Netflix identifies clusters, things like people who love comedy, people who love horror, people who love comedy and documentaries but hate horror, people who love comedies and horror but hate documentaries.
The number of permutations are massive when you’re dealing with a matrix; however, it’s surprising that there’s only a few ways through in the end, because the issues typically boil down to a few core comments.
The machine learning that’s done, in pol.is, is done in real-time, and we do clustering, just like you would have in a recommender engine, except that pol.is visualizes the groups.
Here is this JavaScript conference. On one hand, we say … The group on the left said that inviting speakers to share technology is not an endorsement. Right? It’s not endorsement of that company. Don’t hate the player hate the game. I don’t know if that expression translates. They’re subjects, right, to a system which they don’t have control over; so don’t punish them for it. The group on the right disagreed with that. Then, here’s a statement that the group on the right agreed with, but the group on the left disagreed with.
We see real divisiveness. We see borne out here the elements of that flame war that we saw, but here’s the magic:
The magic is that pol.is also finds and visualizes consensus; and so here is this beautiful comment which someone wrote anonymously. “I think JSCD organizers have the freedom to organize the agenda.” The fact that this exists, the fact that this comment exists, colors every other comment in this conversation; because suddenly, it’s respectful disagreement. It’s disagreement, but we have a core consensus about who we are. Our identity is solid, and this is important to all of us, and we love each other.
Honestly, it comes down to that. We don’t really have an opportunity on social media to say, “I love you. I really respect you. I see what you’re doing, and I really feel that it’s good generally, you know. There’s just little things, right?” We have a global shame laser that sits in space and just evaporates people, right? We don’t have the global, like, “Hey, could you modify your behavior just slightly in this very specific way?” Right? We don’t have that yet.
I’d like to do an experiment here. I’m going to start a pol.is conversation, and I’m going to leave the link up so you can all find it; and I’d like someone to paste it into the hackpad. We’re going to form groups, and we’re going to form groups in an unexpected way. Here’s the topic.
“What ingredients do you like cooking with?”
By the way, this is all you have to do to create a pol.is conversation, Click new, you type in a topic, and then you click see it, share it, and there you go, it exists. You can share it. Go ahead, go to this link on your phones, on your laptops, and then type in … If someone could grab this and post it in the hackfolder, that would be great.
Everyone can go ahead and go to this link, and drop in a statement. Please put in the weirdest things you know, right? Egg, that’ll bring everybody together, right? We probably all are cooking with egg, unless we’re vegan, right?
It may … My hypothesis is that we will identify vegetarians and people who eat meat very quickly. We’ll analyze this as the end result [edit: pol.is did successfully identify vegetarians in real time based on their voting pattern, see at right — and they are in high numbers just as in the real statistics]. I’m going to get back to it now.
This grew in Taiwan over the past couple years, and it grew in conversations that were up to … Here’s the conversation that was up to 3,000 people in one space. That’s the beautiful thing about the fact that it’s a matrix, and there’s no replies. It grows indefinitely; so if 100,000 people show up, well, then you’re just doing math on a matrix of a 100,000 people.
The thing you want to limit is the number of comments, if you’re going to start your own. Don’t let it get to 100,000 comments. That will not lead to a good pol.is conversation.
Here, we’ll talk a little bit about what this grew into in Taiwan. This has been an extraordinary journey for us as a team, as the pol.is team as well, to watch what has been built on top of the tools that we’ve been building.
“Talk to Taiwan” is a political talk show here in Taiwan. It focuses on raw issues, and the idea is this: There’s a Facebook group, I think 18,000 people and growing.
The conversations are sent out through that Facebook group, and then people are drawn into full conversations over the course of a week. At the end of that week, that organic feedback, it’s both quantitat- It’s qualitative data, people said what they thought; and it’s quantitative, because they agreed and disagreed with each other, and were able to layer that quantitative feedback over all the qualitative dimensions that everyone came up with.
That feedback, which is rich and organic, has been put to politicians and discussed with them in an hour-long show, live. This was recently profiled in Monocle Magazine. This is shot of what the format looks like. It’s really beautiful, and you should take a look at it; it’s called “Talk to Taiwan.”
I have one more point to make with regards to this, which is that “Talk to Taiwan” is a huge inspiration for me in my thinking about which possible media with pol.is because, if we go back to that conversation about abortion, the conversation is clearly balkanized. We are divided. People are not communicating to each other.
There are these magical nodes in the middle, that have been pulled out into the no-man’s land, by the graph algorithm. Right? BuzzFeed and Guardian are writing up the narrative. They’re writing the story, and as they write this story, there’s argumentation about it. Of course there is; it’s so natural to argue about what truth is in a place where the truth is being confirmed and written down and codified, and as that’s being then spread out to different communities.
I truly believe one of the reasons that comments are so toxic is that comments are the battleground, and publishers are the battleground, where these different sides are coming together. By using pol.is to acknowledge that there are multiple groups, and also show consensus, this is a moment where we can really come together and discuss things in a way which is much more sustainable and much more productive and much less vitriolic.
Let’s get back to Minister Tsai. What did that grow into? Minister Jaclyn Tsai asked g0v to put together a way for the entire population to have a rational discussion, which is a big ask, right? It’s something that … As you saw with the robot, right? This isn’t going particularly well. This isn’t a solved problem.
At the time, in her own words, “the society was very chaotic,” and so vTaiwan was born. vTaiwan, in short, is a consultative process, where the government agrees to have binding consultation based on feedback. I’ll tell you a little bit about the process of that, and what it’s looked like here.
The first thing that happens in vTaiwan is that the issue is qualified. For an issue to qualify, all of the stakeholders have to be digitally enabled. This is a conversation that happened for Uber, and this vTaiwan process using pol.is did affect Uber legislation and Uber regulations here in Taiwan. It’s law, right? The loop is closed with consultation if the government involved, and that has an impact on people’s willingness to participate.
We see in the Uber conversation, yes, there’s disagreement. There’s enormous disagreement on this issue between groups, enormous disagreement between the groups.
But also, there is a core consensus. Consensus being shown in the context of the division has proven to be a very powerful way to visualize and spread these insights.
All of the issues are digital issues: crowdfunding, Uber, Airbnb, online liquor sales … Everyone that is a stakeholder has a computer. The second is to identify stakeholders and to gather information, and the government does that part.
The third is to use Facebook ads to draw people into the pol.is conversation, and then after people have participated, have a face-to-face meeting with all the stakeholders that is framed by the pol.is results.
Then, the administration drafts a law, and the legislature then votes on that law. pol.is is being used in vTaiwan as that consultative mechanism. pol.is doesn’t really have anything to say about what you do after you get your feedback, but it can help you get that scalable feedback. That’s where that tool sits in the ecosystem …
Let’s look at this one use case, which is online alcohol sales. This happened earlier this year, I believe in March. The government couldn’t move forward, because the constituents could not agree. The ministers were not able to move forward with the issue of regulating online liquor sales, and the problems with the ministers, all had constituents who were yelling at each other across the divide, much like the United States government is basically completely deadlocked on a small number of issues that follow the same kind of profile.
450 people came in and voted, and you’ll see that this ratio is interesting, it actually holds. There are typically about 10 times the people that vote than comment. That’s actually something that we discovered from all these conversations.
There are about 10 times more people that typically just vote than comment, and I think this gets to the heart of something that is true of all online participation, which is that there are fewer people that are very loud and have a lot to say; but if the only medium is commenting, that’s all you’ll see.
The other things that we see strong disagreements coexisting with consensus; and you can see in the vTaiwan conversation that the … Here we have strong disagreement, once again, but we also have a consensus in this context.
A brief quote: “Democratic governance rests on the capacity of and opportunity for citizens to engage in enlightened debate. Although deciding public policy through argumentation has little to recommend it in terms of efficiency, the purpose of public deliberation, as Aristotle recognized in his rhetoric, is not efficient government, but educated judgment.”
It is so extraordinary to see a government willing to engage in power sharing in a binding way. However, what else do we call democracy, if it is not power sharing between the government and the people in a binding way, right? We have to have a binding mechanism. It has been extraordinary to see that evolve. In this context, Gerald Hauser’s quote on Habermas what it means to have a democracy, and why we need public deliberation to fuel that democracy.
Minister Jaclyn Tsai said that people told her that she was going to need three to five years for a regulation. She said, that’s not workable. We need to do this more efficiently. In this context of efficient, educated judgment, I would argue this is the only way forward; that if we’re going to have educated judgment in democratic decision making, that it needs to be efficient, and that by parallelizing everyone’s action, it now can be, much in the same way that everyone submitted to open street map in parallel and created this massive structure very quickly.
This is an opportunity that we have to do both, to have educated judgment on issues, and to do that very quickly. I’m very excited to say that the deadlock on online liquor sales lasted for six years, and the vTaiwan team, using pol.is was able to break it in 3 to 5 months, which was a huge accomplishment by Audrey and the vTaiwan team. I think it speaks to the extraordinary power of online deliberation in validating the people’s, the government’s ability to act.
Lastly, I’d like to just say, I’m incredibly honored to be here, and I’m incredibly inspired by all of you. I think what you’re doing is incredibly brave, and I am continually inspired by all the work that has been done in Taiwan. Thank you.
