Machine learning and deliberative democracy: mathematical modelling of Pol.is

Projekt
, , , ,
1

Some years back, I was a user of an online tool for deliberative democracy called pol.is. I then noticed that I could help with the mathematics behind it, so here I am. It might be interesting for people who are into these things, so here is a short summary of my project and the tool itself.

Pol.is background

For those who haven’t heard about it, pol.is is a survey-like tool where participants are called to vote agree/disagree/pass on a series of statements. At the same time, they have the opportunity to submit their own statements, for later participants to vote on. The system

  • visualizes in real time the clusters of similar voters,
  • the topics they agree and disagree on, and
  • how they compare to other groups.

It has been shown in the past that people use this information to build “bridge statements”. This leads to increased consensus among the different groups as the conversation develops.

Here’s a breakdown of a real conversation of 2026 people in 2018, along with some impressive visualizations of the tool, such as this:

image
image1161×1039 85 KB

All this is quite different than the polarized world we’re currently living in, and this is why the project is close to my heart :heart:. We need more of these tools, and we need them to be effective and transparent.

My work

The premise sounds simple enough but there is an extensive mathematical background to this, a small part of which has been looked at so far. I jumped in to model the problem formally, aiming to connect it to state of the art research in e.g.

  • network theory,
  • graph modelling and algorithms, and
  • social science,
  • machine learning,
  • among others.

By doing that I hope to

  • shield the system from possible crowd-based attacks,
  • suggest improvements based on research,
  • make the inner workings of the system more transparent, and
  • make further development possible.

I work transparently and in the open - you can find my running notes at https://hackmd.io/@ThenWho/PolisGraph/ . Anyone can comment, edit, expand on them or just get inspired :slightly_smiling_face: . If you think this is something for you, you have questions and/or want to work together, do get in touch and let’s get the conversation going! :construction_worker_man: :building_construction: :classical_building:

Pol.is Community

In parallel to all this, I am also volunteering at the pol.is community. I’m not part of the core team, but I have a good overview of the code base, the current way of working, and how you can get involved. If you are interested in these things, drop me a line and I’ll try to help any way I can.

Where to find me

Besides here, of course, you can also find me on twitter, linkedin or at the pol.is community chat. I love talking about democracy online and offline, be it team dynamics, minority engagement, or mass participation. Naturally I also enjoy talking about the algorithms and mathematics behind civic tech tools :slightly_smiling_face:

4 gillningar
2

Thank you so much for sharing this information about Pol.is and your work around it! As we’ve agreed on, you will be giving a short demo of the tool and the community behind it during the Friday Fika this week, on 5 February 2021! :snowman_with_snow: We’ll meet on Zoom at 3 pm with Meeting ID 87834088118 and Passcode 424242. Looking forward to it!

2 gillningar
3

Super interesting! Thanks for the summary and sharing your notes! At Digidem Lab, we’ve been following pol.is since its successful use by g0v and we’ve been using it a few times (the latest being a participatory budgeting process for youth in New York City, report here) last fall.

While we really like the graphs (especially the “Wow” effect it has on civil servants and politicians), we also find it limited to do more than basic surveying. And we prefer more advanced forms of participation where citizens can actually co-construct and decide rather than just say yes-no to statements.

That being said, pol.is can successfully be combined with other tools (such as Decidim, in the case of NYC) and deeper physical deliberation.

2 gillningar
4

Thanks @Pierre , that’s very good feedback!

When a guided facilitation process is needed, Pol.is can be part of the toolchain, as in your case, but it definitely cannot be used standalone. It’s always better to talk face-to-face with as many stakeholders as possible, and a lot hangs on the facilitator.

Where the tool shines is when you want to engage thousands of people or more, as is the case e.g. in vTaiwan. There you need to care about e.g. minority voices not getting lost in the noise, or keeping engagement levels high. This is what I am working on from the mathematics side.

Even then, it needs some thinking by practitioners, on how to use it more effectively. E.g. most think of it as a poll, which is usually a one-off interaction. In the case of pol.is, coming back and re-engaging while the discussion is underway is a crucial factor.

If you drop by the fika later today, I’ll be happy to discuss these topics and more :slightly_smiling_face:

1 gillning
5

Pasting some links from my presentation yesterday. Hopefully can help someone navigate the complex landscape that is the pol.is codebase and community :slight_smile:

1 gillning