Agreed Upon Solutions is built to exemplify a kind of constructive theory of politics. We want to answer the question "what is the best possible democracy," without worrying about the requirements of political feasibility.
If you ask yourself what the most imaginably perfect democracy does, it's something like "Talk about everyone's opinion, on every topic, in order of most to least important. Then everyone votes and reaches consensus about what to do, and that's how the decisions are made."
When the list of issues is small and there aren't too many people, it's possible to do this. But, it doesn't scale. The largest group I'm aware of that does it successfully are the Quakers, and their ability to pull it off is one of the most remarkable acts of community trust I have ever seen; not something that can reasonably expected of a crowd of strangers. So what's the best you can do with a public website?
It turns out there's an answer to this, because doing something very similar is an important problem in the theory of distributed databases. Consider the problem of trying to predict real-world majority opinion using noisy ballots. You want to output "Yes", "No" or "Unknown", for all opinions. Your goal is to say "Yes" or "No" for as many opinions as possible. You are allowed to say "Unknown" whenever you want, but you must never give a wrong answer. This is impossible without assuming an upper bound on the amount of noise, so your goal is to maximize the amount of noise you can handle while still remaining correct. For our examples, we'll assume "Yes" is the majority position.
The noise in the polls is assumed to be Byzantine, a kind of adversarial and unavoidable worst case error. Not only are some votes in your sample guaranteed to be bad, they're bad in the worst possible way, chosen with full knowledge of how your voting system works. This does not just mean always voting "No". They might, for instance, want to blend in if your voting system has some sort of reputation system, to spring their traps at the worst possible moment. In reality, the situation is probably not this dire, but by assuming the worst you can derive the most stable system.
In this system, majority voting does not work. If your vote comes down to 50.1%\49.9%, your adversary could easily change less than a percent of the votes and flip the outcome. Your margin of resistance is 0%.
Let's say you do the opposite, and require a unanimous vote. Then all you need is a single voter to defect, which brings the system to a halt. Your margin of resistance is also 0%.
The optimum threshold turns out to be twothirds.[1][2] A robot can't flip a close decision or block a unanimous decision without having a third of the vote. All you need to do is keep robots below 33% of the vote, which feels like an achievable technical goal.
We're not trying to implement a voting system based on elaborate id checks, we're trying to implement one based on extensive use of fault tolerant algorithms. This is an approach that has not been well studied, because it has some significant downsides: specifically, it does not always reach a decision. Our attitude is that doesn't matter, as long as some questions can reach agreement we'll always be able to make forward progress.
The rest of what we do is similarly involved. There are no direct replies because it makes harassment effectively impossible. We discuss every thing because it's a constructive version of "provide any comment". The "Most Important Thing" pins down a shared answer to an otherwise very vague concept. It's not just limited to voting, either, it extends all the way through the design of the website. If you have an account you can enable "solid mode" in your account preferences to make the website less bouncy on mobile. Why? Because it feels better to touch, and we care about that. Much of what we expect to feel odd about the website is intentional. [3]
This is not a traditional approach to politics. It's a set of ideas unusual enough that it's easier to build, run, and demonstrate that they work in practice; than it is to successfully argue they "would hypothetically work if it were built". That's the core of constructive politics, making arguments by building working examples. It's hard to argue something can't be done when it already exists.
[1] The concept of a "twothirds" is much more involved than simply a number, which is why we write "twothirds" instead of "two-thirds"
[2] Depending on your view of the United States government, the margin of resistance is either one person, nine people, or a few hundred people, for a total margin of resistance of <0.001%. That's why we're facing a possible civil war next week, exciting stuff. \s
[3] This does not include navigational, performance, or conceptual clarity issues. If you experience those, let us know.
