sofiechan home

Democracy After the Bitter Lesson: Advisory Voting

admin said in #3595 3w ago: received

Democracy has problems: the masses are low-information, voting systems get captured by parties and perverse incentives, voting becomes crudely ethnic or handout oriented, and giving everyone a slice of power just opens things up to manipulation and sybil attacks. The best argument for sovereign democracy is the ability to replace a failing elite without catastrophe, but this is debatable in practice and I will put it out of scope here. Let's factor out the political sovereignty function of democracy and just talk about voting systems for operational decision-making.

My favorite argument for voting systems is the wisdom of the crowd. Many decisions must be made by a single leader's well-considered judgement, but often you actually just want to solicit an aggregate opinion from the community, which is often better and even cheaper in expectation than the opinion of any dedicated judge. See also the market.

The market can work great for finding a price equilibrium that optimizes marginal resource usage. But we also recognize that it must be governed politically to match the buying power distribution to politically desired marginal utility, to control externalities, and to manage industry-level investment dynamics. Few seriously advocate that capitalists should hold sovereignty. What if voting systems are similar?

Factoring the sovereign exception-making function out of our voting systems allows us to be much more aggressive in optimizing for operational decision-making and solving their problems. But if the voters aren't sovereign, who is? For simplicity let's assume a monarchic sovereignty for now. Imagine there is some sovereign Administrator with executive power to rewrite and re-interpret the voting systems at will. Then the voters are acting in an advisory capacity, because the Administrator can ignore or reinterpret their votes as he wishes. I call this system "advisory voting".

The problem advisory voting solves is this: given the opinions of the voters on all manner of political questions, and interpreting them as information-carrying signals, solve for the crowd-informed extrapolated will of the Administrator on each question, and do so such that the strategic voting incentives are to be careful, nonpartisan, honest, etc. The sovereign being quite limited cannot answer all questions or know all considerations, but the crowd is abundant and often wise if only the right signals could be pulled out. Projecting the wisdom of the crowd onto the vector of the sovereign's will is just a reasonably well defined machine learning problem on the right side of the bitter lesson.

Various mechanism design schemes like non-independence penalties as in twitter's community notes approximate parts of this idea. But they are still presented as corrections on popular sovereignty rather than factoring out sovereignty.

With sovereignty factored out, voting systems can move at the speed of software and consume large amounts of computing power in learning and inference. In particular, the sovereign's team could model and factor out all the identified issues: low information voters can be identified and given lower weights, factional non-independence can be factored out, the voters' conflicts of interests aren't such a problem if that's just another modeled-out signal, and neither is manipulating the voters or adding a bunch of fake sybil attack voters to the system.

Sofiechan is my attempt to prove out advisory voting in a case where sovereignty is uncontested, the crowd is opinionated, and there are lots of political decisions to be made: an internet forum. If it works, that will be valuable in its own right, but I hope it will also help us chart a course to improve our governance and coordination mechanisms in light of the issues with democracy and the new abundance of computing power.

Democracy has proble received

anon_dova said in #3598 3w ago: received

I think "projecting the wisdom of the crowd onto the vector of the sovereign's will" is beautifully put, but I'd love you to justify your core narrowing assumption a bit more - the question of sovereignty is the important one! Democracy without sovereignty is, like, advanced polling, or prediction markets. Paraphrasing Robert Zion, you've set all the interesting variables to 0.

Parliaments historically started as advisory and then slowly grew in sovereignty. Because a truth machine (which is what you're describing here) is inherently sovereignty-seeking. Designing a truth machine that merely serves the absolute sovereign may be analytically useful but is not a practical attack on the problem of 21st century governance.

referenced by: >>3600 >>3608

I think "projecting received

anon_sida said in #3599 3w ago: received

Do you see any value in the quadratic voting mechanism Glen Weyl proposes in his book? The book imagines the voters being sovereign, of course, but I seem to remember something about it being used to get more value out of surveys i.e. signals aren’t drowned out by people putting +10 or -10/10 for every question because they’re actually forced to make trade-offs. This seems like it ought to improve the fidelity of the advisory function.

referenced by: >>3608

Do you see any value received

anon_hwku said in #3600 3w ago: received

>>3598
> ... a truth machine (which is what you're describing here) is inherently sovereignty-seeking.

The separate sovereign is an intrinsic part of the truth machine. Otherwise, the democracy is corrupted by power and falls into failure modes like an electoral coalition parasitically extracting resources for its own short-term benefit.

referenced by: >>3608

The separate soverei received

admin said in #3608 3w ago: received

>>3598
Yes I've set all the interesting variables to zero. Similarly, a superconductor sets various interesting variables to zero. Epsilon to zero is just as important as zero to one. But enough snark. Sovereignty is a hard problem beyond all rational engineering. It's one of those "leap of faith" things where your backstop is just the goodness of God and you can't really do better than that. Even the most exquisitely engineered self-balancing republic will decay just about as fast as the same men arranged in a customary monarchy. Engineered self-balancing sovereignty is a perpetual motion machine of the second type, so let's just assume whatever sovereign we have and work with it.

Given a sovereign decisionmaker and tastemaker, how can it make use of the wisdom of the crowd to make a lot of local decisions in line with its will? Calling on the opinion of sovereignty is expensive. Calling on the wisdom of the crowd is cheap and abundant. Can we model the one using the other such that the sovereign is still sovereign? Enter advisory voting.

>Parliaments historically started as advisory and then slowly grew in sovereignty. Because a truth machine (which is what you're describing here) is inherently sovereignty-seeking.
Ok so eventually the system itself arranges for the succession of the sovereign when it has become smarter and the sovereign has become decayed or irrelevant. This is fine. Sovereignty wasn't going to stay put anyways, protecting it from its own irrelevance is out of scope here. But what does this succession event look like?

Here's what I think it looks like: at a crisis or moment of succession, the system itself should choose the next sovereign administrator in a special election using the old/current weights and whatever contingency mechanisms the old sovereign has built in. If the sovereign is too far off the most plausible "eigencommunity", we have a problem. The old sovereign can either set this up well or sabotage it. There's no silver bullet. But one reason to expect it to work would be if the main function of the sovereign were to designate which "eigencommunity" was dominant in the system, and to define the mechanisms by which that is computed. Then we would not expect much significant distance between the sovereign and the system. It would not normally be in a proto-revolutionary state of large gap between sovereign and system unless the sovereign has basically failed in his job to to find a good fixed-point equilibrium.

Note this is a natural and minimum-mess response to an *emergent crisis of sovereignty*, not a regular engineered event that we're trying to optimize for. And as >>3600 notes, between these crises the sovereign being outside the system is crucial to the integrity and function of that system. The system touches sovereignty in a single atomic moment of crisis resolution, and immediately passes it back to a new monarch. Or it has decayed to the point where it breaks down in bloody fiery civil war. Such is life. Anyways it won't be an issue very often.

>>3599
>quadratic voting
Quadratic voting is a cool mechanism, as are many of the other schemes the mechanism design people cook up. But quadratic voting is most directly about extracting preference-strength information from the population rather than interpreting those preferences. Yeah it can improve fidelity especially for very important things. The tradeoff is user complexity. We'll use it where appropriate, but so far have not had to.

Another one I'm fond of now is "approval voting", which we use in the tag name selection. You vote honestly for or against as many options as you like, with the strategic choice being where you put your utility thresholds. This has none of the usual arrow's theorem problems with ranked choice systems. I think I will use this anywhere we need to pick between multiple options.

referenced by: >>3793

Yes I've set all the received

anon_ranu said in #3609 3w ago: received

Right now most of the "important" voting is in elections, with some exceptions like issue-by-issue propositions on state ballots. The wisdom of the crowd is applied mainly to the selection of delegates. Which isn't nothing -- you could make any number of optimizations to the voting system to improve its ability to effectively hire on behalf of the sovereign.

Capturing the wisdom of the crowd on a wider range of topics, comparable to the scope of markets, would require a lot more active participation than going into the polling booth once every two or four years.

If voting becomes a non-trivial application of effort akin to participating in financial (or prediction) markets I'm guessing tuning the algorithm is an easier problem than incentivizing participation.

When the god-emperor of sofiechan weights my vote highly enough, shall I be granted a feifdom?

referenced by: >>3673

Right now most of th received

anon_hwgi said in #3673 3w ago: received

>>3609
The "go vote on one question every two to four years" is a very degenerate sort of democracy. We could say it's a function of what there was technology for, but that's not even true. The Tocqueville experience of democracy was pervasive local decision-making by vote and procedure in ad-hoc and persistent formal decision-making bodies. This worked better a) before professional bureaucracy ate everything, and b) when there was a social reality of relative equality of civilized men. Now those formal bodies are dominated by women and various minorities, partially because all such real functionality was eaten by bureaucracies and partially as a result of deliberate deconstruction of the "white patriarchy" after the war.

We could see this project as rebooting Tocqueville's type of democracy for the modern situation: pervasive online digital engagement, and in-equality and non-fraternity as social default.

Will you get a fiefdom? I don't think this thing depends on strong incentives, but who knows what fate holds for us.

The "go vote on one received

anon_mity said in #3678 3w ago: received

On Sofiechan we are front-end anon, back-end immutably ID'd, and we vote on each other's voting power. In democracy, as commonly understood today, you are front-end immutably ID'd, back-end anon (in the voting booth!), and you vote on how much money the state should give you. Doesn't that make it obvious that Sofiechan will work, for lack of a better word, as it is obvious that democracy won't?

I think an interesting aspect of this forum is the lack of group-signalling that comes with anon spaces being tested by the possibility to face-fag. 4chan tried something different (after my time), but to my knowledge did not install a voting system. So I am most curious to see if people who go from anon to pseudonym are going to be punished or rewarded on average.

referenced by: >>3792

On Sofiechan we are received

anon_sepo said in #3792 2w ago: received

Ended up on this because I saw a post about picking up trash in Dolores Park. Great idea, would do it if I was around. It's something everyone can participate in regardless of background.

>>3678

You may want to clarify whether people are anonymous on the backend. If there is a sovereign Administrator with access, people may be less likely to feel comfortable joining in. That's a privacy concern, like a software engineer snooping on social media profiles.

referenced by: >>3795

Ended up on this bec received

anon_hywi said in #3793 2w ago: received

>>3608
Is there a name for a kind of legislative jury system where relatively small groups of citizens decide policy?

I think it is useful to split sovereignty up into contentious issues and boring issues. Boring issues can be handled by professional politicians and administrators, contentious issues by panels of people.

E.g. perhaps we could make a rule that congress has a 60% bar to make law, but for anything with 50-60% support they can have a jury of random citizens talk it out and vote on congress' proposed law.

referenced by: >>3795

Is there a name for received

admin said in #3795 2w ago: received

>>3792
the backend is non-anonymous. You put in your email, it creates a persistent account that accumulates your reputation. I will at some point make it so identity data is destroyed once the reputation data is accumulated. But with occasional changes to the architecture still, that has turned out to be very inconvenient. It will become a necessity in practice for performance reasons as the site grows. It may be possible even to build a system that's actually robustly backend-anon from the start, but I don't know how to do it. The site unavoidably builds a profile on you. That's the point. As for snooping, I'm fairly rigorous about it being only me (and the cloud provider) who has access to the data. And while I could do the join to see who posted what, I don't in practice though I occasionally see email addresses when debugging. I consider it a violation of privacy and try not to do it. Trust that as you will. This is not a homomorphic darknet.

>>3793
Oligarchy? I don't really believe in sovereign democracy so I disagree that it would be a good idea to bring in panels of people to decide contentious issues. Rather for those you need a single visionary leader. Juries are strongest at interpreting evidence to establish plain facts.

the backend is non-a received

You must login to post.