Árni St. Sigurðsson: Data-driven democracy

ÁRNI ST SIGURÐSSON: My name is Árni St Sigurðsson and I come from Iceland, I have a Bachelors Degree in computer science from University of Reykjavik. I have worked at 2 of the 3 largest newspapers in Iceland doing work on their websites. I am currently working on my own start up doing some consultations and freelancing. My background is mostly developing websites and APIs using tastypie and I think of myself as a programmer although I have become more intimate with bash and operations than I was comfortable with at the time.

I’m going to tell you one level of the developing story of democracy in Iceland; there has been an initiative during the last several elections to get candidates for elect office to fill out a standard questionnaire and using that data to fit a voter profile. The last decade has seen turmoil sweep Icelandic politics. Economic catastrophe rocked the country in 2008 and led to a popular protest which ended with the resignation of the majority and an election that shifted the political landscape severely culminating with a historic indictment and conviction of the former Prime Minister for not observing laws on - I don’t have a translation for it. It’s laws on good government that are supposed to ensure that ministers do their best.

It would not be an overstatement to say the national psyche has been going through a schism. There was a general feeling that a change in leadership would not be enough. The faults were fundamental and systemic and nothing short of a total system reboot would fix the problems. Calls for constitutional reform were answered with an Athenian style assembly of 1000 randomly picked citizens from the national registry. The document was to inform a Constitutional Convention to grant the aforementioned reboot. So this is in Icelandic I couldn’t find the graphic in English but this a word cloud of the document produced from that meeting.

Traditionally we have had 2 sorts of elections in Iceland. {Inaudible} watered down by party affiliation ... oops ... one should not get too technical! ... Yeah so it’s usual party affiliation, you mark X for a party and it’s a simple sort of election. The constitutional convention also posed to be non-political, not partisan acted, so mostly they didn’t interfere, but we had 523 candidates for the 25 positions of the convention and this represents the problem, if there is nothing like the elections they had participated infer for, after complaining of lack of choice for decades too much democracy was perceived as a problem and caused confusion. In essence it came down to making an awkward choice of up to 25 individuals from the candidates and we just scrolled through the short bio and position on the constitution and image of candidates of those 523 candidates so you can imagine you know about 50 people, about 20 people very well, you are about to select people who are going to lay the foundation for your democracy, you want to make a good choice but 523 candidates, that’s a lot of choice.

DV saw an opportunity to field an experiment. Being one of the most popular news web-site in Iceland known for hard hitting journalism and having no political slant in its coverage made it a good choice to run it. You have probably had to fill out Likert scales - this is an example question. They come in varying forms but usually have an odd number of options, for or against a proposition. If you have data in this format you can measure distances between any 2 profiles by summing-up the difference of each question.

A couple of things helped the project along. There was no competition in the space and there was a general consensus politicians should not interfere with anything pertaining to constitutional reform, the size of Iceland being only 300 thousand made it easy to publish the solution. 447 candidates answered. That’s 6 out of 7. That’s a very good return I think.

If we thought about averages then you would assume that about 7 of the list of 25 would not have answered the questionnaire but it turned out that there were only 2. So, it would seem to be a good strategy for anyone putting forth their name to actually participate.

Another assumption we can form from this data is that also I ran a comparison what should the - I’m sorry I lost my trail. Best I can give us why people’s votes were not similar to our recommendation is name recognition proved to play a strong role, most of the people elected were very known to most Icelanders. The results of the election read like a who’s who of political activism, with wide range of opinions among candidates and most people agreed those elected would represent broadly the will of the electorate. There is a separate story to be told about the aftermath but the short version is the work progressed quite openly with relatively little interference from the political class resulting in a document which got a positive response in a referendum. The draft was submitted to government and has been a hostage to political class since showing change does not come without its fair share of work.

DV ran another such project for the presidential elections in 2012. It’s rather unusual for Icelandic politics to see a sitting president being contested. There were 5 challengers and 2 of them strong contenders. 24 questions focused on the powers and image of the presidency 36 thousand readers participated while 163 thousand ballots were cast. The results were more predictive but there was still this pattern emerging that people would rather have a vote counted for a winner than vote their issues purely. There was an anticipation that the 2013 parliamentary elections would be historic, the mixture of apathy hope and fear was palpable in both conventional and social media, among punditry, politicians and the population. Instead of the usual 6 parties to choose from this election saw 13 parties put forth a slate of candidates. Each party has to turn in a list of 1890 sponsors to qualify for each district that’s 25,000 politically active citizens out of an electorate of about 240 thousand. Any party receiving less than 5 percent of the vote nationally will not gain a member. This rule has been criticised for being too exclusionary. You generally don’t get regional slates unless they’re recognised as a splinter from a faction with a national relevance. It poses a challenge to any new party. We saw indication in presidential election numbers that people would rather have a vote count than vote for their political positions. All these issues illustrated to us the importance of an election app. This time there would be challenges we have not tackled before. The first logistical problem was data gathering. elections in Iceland are rather short compared to US a little longer than in the UK. Liaising with established political parties would be easier and infrastructure is in place and political operatives seasoned. A multitude of novices however could be hard to tackle. With those data we would still need to run a campaign among the candidates to answer the questionnaire during the height of their campaigning when they were busy doing other things. We featured 63 questions with the option to write additional comments on each question. We gave the candidates an opportunity to write posts that would be featured on the election front page and answer a more casual set of questions. Building up the drumbeat we also hosted some of the more recognised candidates in our editorial room making a news item out of who now had been added to the pool of answered questions. We also have battle fact that some of the established politicians refused to take part to you due to the media coverage done by DV. Remember hard hitting journalism - that’s something that actually worked against us there.

The stark takeaway here was the currently governing parties are the ones least likely to participate in the project and the only parties whose leaders abstained. The Pirates were the first party to have all candidates supply data and only new party to get members to Parliament; 11 per cent of the population voted for a party without getting a representative. Of those 8.6 per cent voted for parties that had enough national following for a representative if not for districting and 5 per cent cap. It’s very ironic more democracy led to less democracy.

Instead of discussing the municipal elections of 2014 I want to give a brief overview of my architecture at that time. If you want to replicate a project like this it would be a good place to start. Django preaches reusuable apps and delivers. you can do pip install, edit some settings and some URLs and build whole products mostly from the command line. Messiness comes about when apps become interdependent as they often do in media. I elected to make an app that abstractly dealt with a list of questions, questions with attached answers and texts attached to questions. This was necessary to be able to have question span municipalities, be fluid in the event of having to change question wording and to be able to build a separate representation layer for the questions. The last feature was not used but any bilingual country would need that. Each model contained an app field so separate apps could be built to represent each election. It would be relatively easy to build a new election web app for each election but if I were to take one more run at it I would build the election app were making an election website would be a push button affair. We made a single change in the algorithm from the year before. One prankster decided to check which party best fitted a known nothing approach and filled in all the answers as “don’t know”. It also worried me that it would be a good strategy for a politician to answer only a few key questions but have the majority be “don’t know”. I fought for a default distance of one for any “don’t know” answer irrespective of what position the voter held, penalising not taking a stance. This resulted in a candidate calling our office furious with a question: how come I don’t agree with myself? I had to summon my straightest poker face explaining to him 2 people who don’t know anything about something are indeterminately far from each other but the closest approximation I could come to without being harsh was to say that they probably don’t agree.

This is probably more politics than any of you expected but most of the time it’s the subject matter that makes the tech interesting instead not the other way around. Voting in a secular democracy is a sacrament. Voters are being drowned in data. For democracy to empower the citizenry, they need tools to see through the fog. I encourage you to light a lamp and show someone the way.

Thank you. {Applause}.

NEW SPEAKER: Can we say thanks to Arnie and has anyone got any questions about this? Yes?

NEW SPEAKER: You said that the position of the newspaper within Icelandic politics was very neutral. Is that all you used to make sure people could trust it in technical aspects...? Did you do an audit what you were doing to slim down a bit more?

ÁRNI ST SIGURÐSSON: Well, it’s more of a culture thing. At the time at least, this newspaper would have been trusted to not take a stance on politics. People were mostly unhappy with coverage of rape cases where they sometimes named convicted people. There was this big case several years ago where a man living in rural Iceland committed suicide on the day that his name was on the front page, so it’s hard hitting in that way. They took a lot of flak for it but kind of vindicated several years later when a lot of men who were boys at the time came forth and confirmed that the guy really was a sleaze ball. We didn’t have to worry about people not trusting the result. There were a lot of spin-off sites that were using our data to do other things, to do things like rating each question for a political compass and then doing a graph of where each party was.

NEW SPEAKER: So the data was opened is that accurate?

ÁRNI ST SIGURÐSSON: Semi-open. You could actually browse through each candidate’s position. We made it a small booboo as well (with respect to openness). For anyone that shared the test the first couple of days the Facebook sharing actually showed how they answered the test so we quickly quit that - and I was kind of amazed that the government agency for privacy didn’t actually contact us and do something. But that was an honest mistake.

NEW SPEAKER: Anyone else with questions?

NEW SPEAKER: Do you see technology as being a good force for use for creating a digital town hall?

ÁRNI ST SIGURÐSSON: Excuse me for a town hall?

NEW SPEAKER: Yeah a way for the public to directly engage and drive decision making in a democracy?

ÁRNI ST SIGURÐSSON: Yes, yes we need this, we need all sorts of projects on democracy and we need funding for people to actually just go ahead and do these projects and publish them not as a vassall of some company, not as a vassall of - not as a wage slave, so that more journalist outfits are going to run things like this. In my opinion technology actually opens a whole new avenue of investigation in democracy. If you think about representative government we’re actually using the worst form possible at the moment. We’re sending someone to Parliament or any elect office for some amount of time and we have no say in what actually he does. So, we need to be able to make a recall, you know, say obviously this representative is not representing us and I actually have an idea. I mean think about it. Think about elections. Think about Facebook. Why aren’t elections Facebook basically? why aren’t districts groups of people that collaboratively pick somebody from the group who go to Parliament? If the representative is not delivering on his or her promises, then the group can just recall it. I mean, it’s relatively simple to do something like this. You could have a layer. My favourite idea is something I’ve been calling an emergent congress, where at the first layer, you have to convince - you’d have to sign up at the national registry for the election, I want to be in Parliament. And you have to then - you are randomly assigned into a district and you need maybe about 10 people and you have to convince them that you are supposed to win this round and progressively you let people from each group meet people from other 10 groups where you also have to convince everybody else that you’re the guy who goes to the - and then you finally have something like final decision or something that the whole group can then vote between, I like this guy, I like that guy. So, if you couple this with recall option what’s the probability somebody is going to get chosen who will not do the will of the people? That’s very low. That’s actually you would say non-existent because this will presumably be a prestigious position people will want to keep and they’re not going to keep it if the ones that elected them aren’t - well, if they’re going to issue a recall and that’s the end of your term because the probability of you maintaining the position amongst the people that sent you there when you didn’t do what they wanted you to do is low.

NEW SPEAKER: How do you think this can happen in other countries?

ÁRNI ST SIGURÐSSON: Something like the thing I was doing.

NEW SPEAKER: Yes in Iceland right.

ÁRNI ST SIGURÐSSON: The tech behind this is simple. There are obviously things that are going to bite you, if you do something wrong for example, but basically, the first implementation of this was just a model with each question being a field and form model to collect the data and then it is a matter of having an infrastructure to compare, so, you know, after this talk you should be able to pull together in very simple terms something similar. So if you want to do it locally, give me your e-mail, I will be in contact.

FROM THE FLOOR: Thanks.

FROM THE FLOOR: What about hacking and security, that is usually the main thing when we talk of politics, there seems to be no solution.

ÁRNI ST SIGURÐSSON: So you are worried about some of the spoofing candidate for example?

FROM THE FLOOR: Yes.

ÁRNI ST SIGURÐSSON: Well it is not relevant here, it is relevant in everything we do, we should of course worry about security, we should worry about access controls. But, it is not a part of this discussion. It is a part of what we have to do generically for example, we have a pay wall, I am no longer working at DV so, they had a pay wall. If the pay wall doesn’t work, you know, if people can just smuggle themselves inside the pay wall we have no product. That is where you solve the security issue in the first place.

FROM THE FLOOR: Special problem with security and boarding, you can’t mix security and anonymity. So it is not the same problem as the pay wall. Because the pay wall I can look back at my payment and I can know if there is my payment but, by default, definition, if you make them the voting public.

ÁRNI ST SIGURÐSSON: I am not advocating e-voting. Nothing about this is e-voting, all of this gathering data, filtering data and doing something with it.

DANIELE PROCIDA: I think we will have to stop there, thank you for your questions, thank you very much. (APPLAUSE).