Brief technical description:
1. produce a matrix of C comments (submitted by participants) * V votes (in 1, 0, -1 form for agree, disagree, pass).
2. Run dimensionality reduction (PCA) and clustering (K-means).
3. Find out which comments best differentiate clusters, show those to everyone https://compdemocracy.org/representative-comments/
4. Identify comments in which the majority of every cluster is voting the same, way, show those to everyone: https://compdemocracy.org/group-informed-consensus/
The 501c3 works around the world (USAID, UNDP, cities, countries) to help advance the usage of the method in deliberative democratic settings. Multiple cities (such as Amsterdam) now have their own deployments. We'd love to hear from you! If interested in the project or volunteering, please reach out to: hello@compdemocracy.org
What the pol.is appears to do well is provide a topology of opposing viewpoints- identifying the main "teams", finding the contours of agreement and disagreement, "a technology for survey research."
The HN URL describes something more ambitious: a technology for "Large Scale Discussions." This requires iterative steps where representatives from opposing viewpoints ask clarifying / challenging questions of other viewpoints, and collapsing "consensus-achieved" points as summary discussions in a knowledge base.
I've long had `pol.is` as a tool to be used in a 21st century platform for discourse, so the HN title has me excited that you're taking steps in this direction already.
Where should I look to understand "what happens next" after that initial sorting of viewpoints has occurred?
After the clustering, it then identifies points of agreement across clusters and surfaces those for voting to verify. That’s in the article on the linked site:
“The algorithm then identifies “consensus statements" about which all the diverse clusters agree.”
There’s also some notes about “group-informed consensus” on the algorithms page that might be interesting for you:
I am flipping through your FAQ, and it looks like the only way to log in is through Twitter or Facebook. Are there any plans for a more general OAuth implementation? I'd like users to be able to log in to my organization's website and then have that login work for Polis.
Maybe only allowing FB and Twitter was an ironic choice (a consensus app requires divisive platforms to log in), but I would certainly like a third choice.
(This is an outside project contributor speaking.)
> it looks like the only way to log in is through Twitter or Facebook
true :(
> Are there any plans for a more general OAuth implementation?
Some tickets open, but hasn't moved in awhile. Def an opportunity to contribute this! :) Context in the issue linked later.
Right now, you can get third-party auth working in one of two ways: (1) contributing the feature to upstream Polis (slow, and lots of work) or (2) do auth in a custom "wrapper" app in which a Polis embed is used (quick, and no need to understand the messy Polis codebase).
The motivation for the inclusion of social platforms was multifaceted, and included:
1. ideas of portable reputation (follower counts / verified / real id)
2. showing how others you follow on twitter voted, visually
3. showing known people, like politicians, voting relative to others.
In the end, the platform is happy to know as little as possible. So, xid
This is fascinating stuff and seems like a leverage point to apply technology to actually solve some of the big problems facing humanity, which at their core are political problems, not necessarily technological ones.
Where do you need help the most with this project?
Where do you see it going from here?
Any interesting big trials or deployments in the works?
We have a range of volunteer opportunities for frontend and backend engineers (React/legacy Backbone.JS/D3), as well as data scientists (python / clojure). We also have a range of opportunities to work with organizations around the world that need but don't have the data science background necessary to properly implement the tool. There's manual work to do here that usually means getting on video with people in different time zones.
We have some refactoring needs that are still in flight as we are still in a long transition from startup codebase to 'easy to deploy' generalized OSS repo.
2. Where is this going?
We're excited to see people deploying this around the world, (a process that has been eased by community volunteers along with investment from governments). Because of Schrems II, OSS platforms have a degree of competitive advantage in Europe as those governments that need to own their own data, and there are also big pushes for deliberative democracy there.
While we've worked around the world, bringing the methods / deliberative democracy to the United States is still something we're working out. The 501c3 is presently pursuing a strategy to 'compile' online deliberations to ballot propositions in places with small amounts of signatures needed, with the idea that if we take a known issue area (say opioids) + population sample + emergent discussion we'll discover a solution space, as the tech has elsewhere (like Taiwan), that won't then kick off the kind of adversarial advertising campaigns referendums usually attract. Potentially multiple zeros cheaper, and hopefully better outcomes. We're presently working on our first of these. Follow https://twitter.com/compdem for updates.
My writing about the future of democracy (working on a book) is presently focused on whether or not we can replace political parties in practice and compile a better high dimensional space to a legislature of independents. https://web.archive.org/web/20190629035125/https://civichall...
3. Any big deployments?
UNDP has just concluded what they think may have been the largest _online_ deliberative exercises in the developing world in history, in Bhutan, Pakistan, and East Timor. The full report will eventually be posted to https://twitter.com/compdem
Does agree/disagree/pass really reflect what people think even in short form? A lot of voting may occur on a particularly emotional point, even though a voter may disagree with the rest of a statement. I've seen this on Reddit and here, and it's always made me wonder how we can actually capture the net of people's sentiments accurately.
This is a great question. I've been using Polis for maybe 5 years.
The "pass" is a catch-all for many things[1], including "rejecting a premise". It's non-obvious in the current interface, but there's room for coaching/learning how to engage with "complex/mixed" statements. For example, statements you encounter that conflict with your own emotional or factual sensibilities, which you're reluctant to respond to in a simple way (with agree/disagree). In this case, users can just "pass" on that statement, and immediately go below and submit a new "corrected" statement, or alternatively divide that one statement into a few separate ones, that can be reacted to more authentically.
That way, you're not forced to weigh in on things you find to be leading statements or overly simplistic. And the pool of statements for everyone becomes ever-more-nuanced through these sorts of interactions, and the map of the sentimental landscape becomes more high-resolution :)
I am a bit cynical about this system.. isn't this system naively gathering a temporary consensus that is constantly being changed by the raging media landscape? As soon as there is a media campaign about any topic whatsoever, all these temporary records of consensus will shift and change into completely different directions. How much value does a consensus system like this truly have in finding the truth of a subject matter versus being the equivalent of a slightly fancier strawpoll? I participated in one of these polis discussions and I have to admit that I found it interesting to see the different opinion clusters but at the same time I realize how little it means. Even the tonality or phrasing in a question could change the clusters completely. It's such a hard problem where I somehow tend to think that this system may be impossible to use for healthy politics but might be used as a tool by people with bad intentions to gather data about what the population superficially thinks and then be able to shape your media campaigns into changing people's opinions even more ruthlessly to what you want them to think.
I really appreciate this critique. And I'm grateful for your cynicism. So long as that cynicism isn't used as a reason not to take tenuous steps toward something different from the status quo.
With that said:
> How much value does a consensus system like this truly have in finding the truth of a subject matter
There's maybe some things to hash out here. If the value of consensus is seen as finding a "truth", then yes, I agree this tool is suspect for that. But if the value of consensus is simply finding enough agreement (on feelings/facts/whatever) to take a step forward together, then maybe it can survive truth shifting under our feet a bit. What matters is whether we maintain commitment to moving in some direction together. And my experience is that Polis creates the stability and trust required for this.
Unlike the hyper-division fostered by social media platforms, Polis helps build stable common ground to stand on. It can be used to elevate and incentivize people who do the work of offering bridging (aka shared) sentiments. Other tools don't do this. My mental model is that the availability of these bridges in the opinion space, this stabilizes middle ground on which long-form deliberations can take place. [1]
> might be used as a tool by people with bad intentions to gather data about what the population superficially thinks and then be able to shape your media campaign
YES. This is a huge concern to not just Polis' creators, but many who work around big data. The capacity for abuse or manipulation is huge. My personal hope is that we can all build tooling like this together, to make us as a population legible to ourselves (symmetric, self-knowledge), rather than data simply making us legible to powerful actors via asymmetric knowledge (which is what's currently happening). This is why imho tools like Polis should be run openly by governments or other trusted partners (journalists). Cryptographic and data trusts should certainly play a role in future iterations.
This whole effort needs people thinking about safeguards in the same way that those designing/maintaining a democracy think about safeguards. imho the process of running these processes must [eventually] be comparably sacred to those of running elections and parliaments.
Couldn't you make this argument about democratic deliberation in general? I dont disagree, but the provisionality of opinions is sort of an existential question for democracy.
Certainly a number of valid concerns here. Like any tool, polis is as good as its context, usage, data inputs.
I can offer a specific case study with data, that deals with the discovery of the specific values of a specific population in the context of highly contested issue that has played out over a multi-decade period:
This is an ideal use case, but it is encouraging to us (and has motivated me over the course of the past decade) that given some combination of factors, there can be real progress, and while yes, it takes a bit of work, we do spend an enormous amount of time and money on worse outcomes than this. So, we can "save one starfish" https://raycenter.wp.drake.edu/2018/06/29/the-starfish-story... and put some lights on the horizon of other possible systems, should we inspire some millions, then tens of millions, then hundreds of millions to be spent on other political activities than driving wedges.
Great question. You can see raw data from a conversation on biodiversity that took place in New Zealand in partnership with the Ministry of Conservation and hosted by an independent news organization here:
If you take a look at the comments file, you can judge whether the system has been successful at getting people to submit a single idea at a time for people to agree and disagree on. My assessment is: mostly yes, sometimes there are still multiple ideas in 140 characters.
The system encourages the decomposition of ideas, so that we can identify which factors may be ignorable and which are significant to our purposes (consensus/disagreement/differentation).
Just FYI, I also had trouble finding the github repo. In case it helps, the HN URL lands on the polis page [0], which itself has an `open-source` link[1]. Not finding the repo link on the open source page, I checked the About [2] page and the collection of links contained in the `footer` section.
At that point, I gave up and searched `polis open source github` on duck duck go.
EDIT: There's an unofficial Polis User Group in a Discord channel. We used to run weekly open calls to help people learn, but for now we just have the Discord. Details: https://link.g0v.network/pug
Yeah this should be linked in the top comment because I couldn't really grasp what Polis was even for by going to the link on this topic. This article really does a great job of clearing up its purpose and benefits. Thanks a lot for linking this!
This is really interesting technology for people to find 'common ground' in polarized discussions. I really wish more research would be done in this direction.
In 2018 I was involved with an NGO that planned on using Polis in their work. They contacted the maintainers who recommended contracting an agency which had experience with Polis as many of the core developers worked there. The NGO did just that, and paid a huge sum for the deployment. The first discussion resulted in a total overload of their servers. The NGOs marketing spent for this project was completely wasted and its reputation tarnished. I was brought in in the aftermath to communicate with the Agency and to scale the deployment. They were unable to scale even to 100s of simultaneous users. We tried to deploy it on our own, only to find out, that the public source code on github did not match the one deployed by the agency. The OpenSource Version on GitHub lagged the version the agency deployed by about a year. Some files critical to deployment were completely missing from the public release. Furthermore the Code was a complete mess. Apparently they never had a successful deployment. The fact that 3 years later in this very thread people are still pointing out that it was used years ago in Taiwan (which was the PoC that convinced the execs at my NGO in 2018) is a testament to that.
If you look at their GitHub (which they don’t promote a lot on their website) they explicitly say:
„ If you'd like to set up your own deployment of Polis, we encourage your to reach out to us for support. We look forward to working together“
This is the real purpose of their project, they try to get consulting gigs for themselves and extract a maximum of money out of unsuspecting NGOs and Govermental Orgs. They have sweet talking sales people but are unable to deliver on their promises. Never Again!
Feel free to DM me on twitter if you care to elaborate and want to discuss, this is strange and I'm not sure what you're referring to. The reason we're a registered 501c3 is that the overwhelming majority of the consulting services we provide around the world are done pro bono or at a loss. I'm not sure what provider you're referring to.
The question is why your consulting services are necessary in the first place. Most Open Source Projects I know can be set up by anyone who is competent in the underlying technology. You explicitly warn people that they will be unable to set it up on their own and that they will need your services. You had years to simplify deployment and haven’t done so. Your Github is still lagging the version that you’re privately deploying.
Can You point to logs of any recent deployment that scaled to 100s or 1000s of simultaneous users ?
Somewhere else in this thread you claim that OSS software has a competitive advantage in the EU. Is this the reason you pretend that your software is OSS while in practice it is not ?
The software is entirely open source and multiple independent instances exist. Our main deployment at pol.is runs off the same docker containers available to the community.
Wow, this is beyond ridiculous. This is the NGO I was involved with. It did not scale to 30,000 it failed miserably at merely dozens of simultaneous Users. The linked Video is the Exec who took responsibility for ordering Polis trying his best at PR Crisis Management. He was fired shortly afterwards. Apparently You don’t understand German, because in the video he is admitting the technical problems with polis. I‘m just amazed how anyone would point to that video, to tout the supposed strengths of their platform.
Interesting! This instance had thousands of concurrent users for a total of 33,000+ and was one of the larger instances of usage of the tech. There was no independent deployment of the technology in this case, so, that is inaccurate.
The instance failed at mere DOZENS of simultaneous users. Clients kept retrying and then there were thousands of requests of users who were unable to interact with the instance. I went though the logs in detail. It was a clusterfuck.
As an open source contributor, I submitted the code in 2020 that made it easier to get running with Polis. I find it quite frustrating that you're slandering the project with weird speculations and confusing factual accounts 3 years later, after other people put in the time to improve things. As far as I recall, you never entered a public space (issues, chat) to offer anything, even a formal bug report.
I can't understand why you didn't ask for help in the issue queue or chat. I'm sorry if you had a bad time, but your speculations are a funhouse mirror version of what I understood to be the case.
So you upvote and downvote comments, which are presented randomly and by themselves. No support for replies or threading? Seems a step backwards to me.
The lack of replies is a feature, not a bug. Polis is about generating consensus from what I’m reading about it, not about maximizing engagement ala Twitter, FB, et al.
“Using the Internet to pull people together rather than split them apart requires designing an environment very different from the usual online forums for political debate, such as Twitter or Facebook.”
“Polis has reengineered many of the features we take for granted on social media. No reply button – hence no trolling. No echo-chambers, replaced by an attitudes map showing you where you are in relation to everyone else. The platform does not highlight the most divisive statements, but gives more visibility to the most consensual ones.”
The claim was no trolling. Not trolling gets hidden eventually. Other systems hide unpopular statements too. Trolls can work together. And Polis shows the most agreed and disagreed statements apparently.
If the claim made it seem that trolls somehow totally disappear, then it was def mis-articulated. Trolls get minimized. For instance, there's currently a small troll group of 3 participants in the viz (out of 100 total participants), far off to the left. But the system doesn't invite reactions to them, and it's evident they're fringe, due to placement in the opinion space. they don't get elevated by our reactions to them, and so it's trivial to discount them while reading the visualization.
It's similar to how it works in an IRL discussion in a room. If someone is trolling, you can understand that by their relation to others in the room (where people direct their eyes, stepping toward/away from them, whether facial expressions respond to their ideas). You discount their views accordingly. Polis allows for some of that intuition to return, in a simple visual medium without our physical bodies.
Fringe and troll mean different things. And any system with votes makes unpopularity evident.
The problem with trolls isn't not knowing if they're sincere. It's the ratio of signal to noise. And it was bad even without the trolls. Half the statements didn't answer the question.
Correct. For smaller discussions in the 10s, or potentially 100s of participants, the limitations of randomness and no direct replies are unnecessary. They become useful assumptions given thousands and tens of thousands of participants.
We've been working on this for coming up on 10 years. We launched on HN in August of 2014: https://news.ycombinator.com/item?id=8228974
Brief technical description: 1. produce a matrix of C comments (submitted by participants) * V votes (in 1, 0, -1 form for agree, disagree, pass). 2. Run dimensionality reduction (PCA) and clustering (K-means). 3. Find out which comments best differentiate clusters, show those to everyone https://compdemocracy.org/representative-comments/ 4. Identify comments in which the majority of every cluster is voting the same, way, show those to everyone: https://compdemocracy.org/group-informed-consensus/
For more, we've recently published a paper describing the underlying methods: https://twitter.com/colinmegill/status/1445044310722822147
The 501c3 works around the world (USAID, UNDP, cities, countries) to help advance the usage of the method in deliberative democratic settings. Multiple cities (such as Amsterdam) now have their own deployments. We'd love to hear from you! If interested in the project or volunteering, please reach out to: hello@compdemocracy.org