You Can Do Futarchy Yourself

Long ago, in the year 2000, Robin Hanson raised the question of whether we should vote on values and bet on beliefs. He proposes a form of government called futarchy, in which people vote on values and bet on beliefs.

Under futarchy, values experts propose measures of global welfare, policy-writing experts write up policies, and policy-evaluation experts bet on the outcome of those policies (in principle, anyone could do any of these things—but politics is valuable enough that I imagine experts will occupy these three roles). People vote on the measure of global welfare used, and the policies that have the best expected outcome by inputting the results of the betting markets into the currently-elected global welfare measure become law.

Sounds great! But as with all good things, one might suspect that the reason we don’t already have futarchy is because of some sticky Nash equilibrium.

Imagine trying to go full futarchy, straight away. There are tons of different stakeholders in politics; some of them gain power when you implement futarchy, some of them lose power. The ones who lose power attempt to veto this. Given that tearing everything down and implementing futarchy requires a similar set of moves to tearing everything down and implementing dictatorship, the historical record suggests that they can usually veto this; you are not living under a dictatorship of the first person to ever try becoming a dictator. Depending on how powerful you are when you try, this looks like everyone ignoring the futarchy of you and your less-enthusiastic high school friends, a SWAT team raiding your futarchy compound, or your nation’s constitution not having an amendment that lets you do futarchy.

So, forget systemic change for now. Imagine trying to become a personal futarch, as an elected politician. You hire a software engineer to write a prediction market app (like mine), or make a little thing on the Augur blockchain or something, and then from then on you vote on legislation according to the results of the markets rather than your own guesses. Just kidding! No you don’t! You’re an elected politician—the part of you that would be weird enough to do something as terribly unconventional as futarchy was ground away in your climb to the top, or you preserved it and got selected out of politics and you’re not actually an elected politician. And that’s if the party whip lets you make decisions like this, and if the betting commission lets you make a market in the first place.

So, forget working within the system for now. Imagine trying to, as a person sitting at your laptop reading LessWrong, do a futarchy using only the tools available to you. Think about what parts of futarchy you can do and what might need to be changed to fit. Is the Nash equilibrium really so sticky?

The values experts can remain almost as-is—it’s just a matter of getting some philosophers and economists to look at the world and map it to a single number that represents how good the world is. Our World In Data collects lots of objective metrics ready for you to multiply by utilities, and the Candidate Scoring System is an existing effort to create a weighted sum of various measures of how good things are in the US to score candidates. Some slight modifications to these to make them ready for forecasting are perfectly doable from where you’re sitting now.

However, if you’re just a values expert, people are absolutely free to ignore those values, so you’re going to want to gain some power over which policies are enacted, too.

What scope do you have to decide which policies are enacted? You can’t pick policies unilaterally, nor can you even vote on them, usually; representative democracy instead means that you vote for politicians who then vote on policies. So this personal futarchy is going to be about which politician you vote for, and the policy-proposal experts are just going to be preexisting politicians which are mostly outside of your control. Under plurality voting, tactical voting considerations mean you should be checking the value of voting for one of the top two politicians.

Now for what you can do with the final piece, the active ingredient of futarchy: the policy-evaluation experts forecasting the actual outcomes of the policies that will be enacted.

Hanson proposes prediction markets, but I’m not so sure. Certainly, they’re good, but they do face issues; one big one is not enough liquidity. It’s tempting to try to handwaving this away with “if prediction markets get bigger, there’ll be enough liquidity”, but firstly you’re a rationalist at your laptop and you can’t get everyone to go along with your hare-brained prediction market scheme, and secondly, it might be the case that there will never be enough liquidity. Even at its most pared-down, you have 2N markets for every election (where N is the number of terms in your global welfare metric), and some of these elections are going to be state (or, heavens forbid, local) elections that won’t gather the enormous amounts of attention that the big presidential elections would gather.

Furthermore, Jason Dana, Pavel Atanasov, Philip Tetlock and Barbara Mellers find that you can do better than prediction markets with teams of superforecasters—Dana et al.’s paper “Are markets more accurate than polls? The surprising informational value of “just asking”” perfectly satisfies Betteridge’s law with the conclusion that polls can do better than markets… with the catch that the polls should be weighted by track record.

Metaculus does exactly this—in addition to the community prediction which is just a straightforward median, there’s also a Metaculus prediction which weights predictions by the track record of the predictor. It even lets you call off bets via resolving the question as “ambiguous”, which awards no points and doesn’t influence anyone’s track records, which gives you everything you need for a futarchic prediction platform.

So, were you some virtuous person trying to save the world by switching the one 7.8-billionth of politics you were responsible for to the best system and acting accordingly, you might end up doing so by collecting a bunch of metrics, and then asking a battery of questions of Metaculus about these metrics. It’s even a bit better than just you doing futarchy, because the Metaculus predictions are semi-public so anyone voting according to a global welfare function that has a term for any question that has already been asked can use those forecasts.

We probably live in the kind of adequate world where someone probably already has this covered, because it’s such a low-hanging fruit with a great upside (it switches your voting pattern from “best person at tricking you” to “politician with best outcome according to our most accurate forecasts”, which at least avoids the sub-superhuman forms of trickery), so let’s do a Metaculus search for some keywords that would yield the obviously-extant attempt at doing a personal futarchy. Say, questions that are conditional on Biden becoming president—let’s search “if Biden”:

https://www.metaculus.com/questions/?order_by=-activity&search=if%20Biden

Well, there’s something there: four questions as of the time of writing. They all seem to be by the same person, an anonymous Metaculus user who goes by the handle holomanga (holomanga informs me in private communication that they go by Tetraspace Grouping on LessWrong).

These questions are somewhat sporadic, and don’t look to be a very systematic effort—were I to randomly guess, I’d say that they made those questions to get some “objective” results to cite in Twitter arguments, rather than in an actual attempt to personally implement futarchy-lite (I mean, for a start, I don’t even think holomanga lives in the US, so they couldn’t do anything with the answers to questions about Biden or Trump; though I do notice that they’ve asked some questions about broadband speeds wrt to the 2019 UK general election). But still, they’re a nice start.

So, after this nice start, what are the next steps? Specific action points for those interested (and for my future self, of course):

1: Come up with some good metrics to forecast on. These will be things that different politicians will effect differently, and that correspond to the amount of good in the world. Then ask about these on Metaculus—two questions, one for each politician, or more if you’re lucky enough to live under a voting system where you have more choice. Asking Metaculus questions is quite easy—I can click “create question”, type up a bit of background on why the question matters, and done.

2: Convert these metrics into a number that represents how good things are. The previously-mentioned Candidate Scoring System goes through the effort of weighing US presidential political issues by the total amount of good or bad the president could do, and more generally multiplying things by QALY numbers is a rich part of our cultural heritage.

3: Vote, and encourage your friends and Twitter friends to vote, according to whatever the output of this is!

Acknowledgements: Thank you to Daniel Kokotajlo for organising Blog Post Day III, without which I wouldn’t have had the motivation to write this.