SIEVE - a Better Sortition

This post on Best Of A Great Lot is a part of a series on the subject of designing a new form of governance. Each piece aims to stand alone, but fits together on the Table of Contents.

An earlier version of this referred to this as 2xR, which is a worse name.

This is my current version of this, and I stand by the idea that it’s better than many things we currently do. But I am working on a new version that’s focused more on allowing people to reject candidates, and hope to post it and link to it from here when it’s ready.

For the moment let's set aside the other problems of voting: how poorly it scales; the arguments over the best voting method; bundled representation being a poor fit for our actual preferences; the challenge of people whose perspectives aren't represented by any system that chooses a majoritarian winner and similar. Even if we were to fix all of these, we'd still be left with a fundamental incentive problem.

Those who seek power are not worthy of that power. - Plato

Not everyone who seeks power seeks it for their own ends, but enough do. Great institutions can channel and limit that power, but, as they say, garbage in, garbage out. An ideal system would pick our best and our brightest, not our best of those who want power. Yet we must have some way to choose.

Unfortunately, this problem affects more than just voting - Presidential appointments and in-agency promotions, among other systems are also subject to significant behind-the-scenes jockeying that privileges the politically astute. The last two decades have seen frequent commentary about "building your personal brand" because we could describe our society as marketing-driven as easily as we could say that it's a meritocracy. It's been a fortunate thing that it's harder to market things with zero merit than things that have some, but we all know that the best choice isn't always the winner over the best marketed.

This is why there's so frequently the temptation to sortition (selection by random). The Ancient Greeks did it, choosing their government by lot, why can't we? Of course they carefully curated who could be chosen, a fact that's often conveniently glossed over. Regardless, we have enough distance from their society that it's hard to tell how well it worked in the day to day.

There's also something to be said for the simplicity of governance in a society that understands vastly less about how reality works and is more willing to accept explanations involving implacable divine forces. Reality has a surprising amount of detail, and these days we actually have to reckon with it. We've also made great use of that detail to create additional levels of complexity beyond what nature provided us. They often say of Mussolini that whatever his downsides, he made the trains run on time. The Ancient Greeks didn't have to worry about making trains function, much less to a schedule.

Only 35% of Americans have a college degree. If we were to choose some fairly technical governmental role, whether Member of Congress or agency head, completely by random, 2/3 of the time we'd get someone who hadn't graduated from college. I've known very capable people who never graduated from college, but they're exceptions. Most people who don't attend or graduate college do so because the kind of work required for college doesn't fit well with their abilities. The knowledge work for governance shares a lot of similar attributes and required abilities.

Another proposal often made: we might also use track records to determine who is meritorious. Using prediction markets or other tools, ask not who has the best credentials but who has proven themselves correct in the past. I both love and hate this idea, and it's worth its own post later. My shortest reason for dismissing them as a solution to this problem is to point out that we still need to choose people to evaluate the outcomes that generate those track records.

Throughout any system of governance we'll have to choose again and again who should be either responsible or involved in some governmental activity, whether that's the equivalent of Member of Congress, judge, leader of an agency, or any other important position. Are there other options?

The option that I've invented (at least, I don't know of anyone proposing it prior) is to combine sortition with a small scale advanced voting. I call this method the SIEVE (Sortition Initiated, Enhanced Voting Elected).

The method goes as follows: first, two pools of people are randomly selected, one for candidates and one for voters. The candidate pool might be fifty people, and the voter pool might be two thousand. The voter pool is small enough that voters and candidates can have a lot of direct, genuine, human-to-human interaction without a professional marketing team to manage the candidates' appearances. The randomness ensures that candidates can't spend years preparing and politicking to run but instead know that the work they've done will be reviewed if they get lucky enough to be selected as a candidate.

After spending time with the candidates and reviewing them, the voters can use an advanced voting method like STAR or ranked choice methods to choose among the candidates. It's relatively easy to train a few thousand people on even the most complex voting system without the mistakes and misunderstandings you get when you try to roll it out to millions of voters.

This method gains us the benefits of sortition and the best of small-scale voting - candidates don't get to self-select in, so we shouldn't expect to be picking from the unusually interested in power, but we still pick the best of the candidate pool. The voters know that they're not wasting their time, because the voter pool is small enough that their votes really matter, yet can be made large enough to be a reasonable sample of the population.

This gets us far enough to be called "the best of an average lot", but not far enough for this method to reflect the name of this blog. For "the best of a great lot" we have to go farther and add an eligibility requirement to the candidate pool.

Eligibility requirements do not have a great history in recent times. The main examples that come to people's minds are elements of the Jim Crow regime, such as poll taxes and literacy tests, which were used to enforce institutional racism after slavery was formally ended. On the candidate side, the American Constitution allows only the barest limits on eligibility because the Founders did not want to recreate the caste systems of Europe that set the nobility apart as leaders and declared the commoners unfit for power.

There are good reasons to be skeptical of candidate eligibility requirements: they serve as a tempting target for politicians with power to use to prevent their opponents from running. They can also be a tool used to prevent minorities and other outgroups from participating in the political process.

Yet in some circumstances they make a lot of sense. We have an effective eligibility requirement on who can become a Supreme Court Justice today - the role is (by cultural expectation) limited to people who have been judges, law professors at prestigious universities or politicians from the highest level positions in the law. For some roles, there is an obvious pool of people who are most qualified, and pairing that pool directly to a 2xR allows for the best of both worlds - a high quality, randomly selected candidate pool that we can choose the best from.

An eligibility requirement is less likely to be a political target when the underlying selection dynamics have been randomized. If I can't guarantee that I'll be chosen as a possible candidate, my incentive to prevent someone else from being chosen is less personal and can be more directly aligned with the idea of who would be best in the role.

Eligibility requirements can be further weakened but still useful by turning them into probability gradients. For example, if we are selecting for a role that calls for a deep understanding of the law, we could require a judicial appointment, or, for a weaker eligibility requirement and broader possible selection, we could simply advantage former judges by making them twice as likely to be selected into the candidate pool.

Eligibility requirements are best paired with strong laws against invidious discrimination. The word invidious does a lot of work there, just as it does in our current nondiscrimination law. It means that the discrimination must not be arbitrary or unrelated to a legitimate purpose. Here, the legitimate purpose is narrowing the pool to ensure competition among likely candidates. This means that an eligibility requirement must be achievable by a dedicated person.

Imagine for a moment that we were to select people using SIEVE for a role like Member of Congress, where they will be deciding on laws. We want a smart and capable candidate field to choose from, so we might set up an eligibility requirement that allows anyone who has demonstrated in some way that they are (at least minimally) smart and capable. A broad eligibility requirement that is easily satisfiable is a perfect complement to a 2xR because it filters out the majority of people and ensures the candidate pool is likely to have a good competition of choices. Those who are interested and capable can easily become eligible either through activities that they will perform in the normal course of their life, or through a little bit of work.

For this kind of general role, we might design it that anyone with a college degree is eligible, as well as anyone with a general or honorable discharge from the military, and anyone serving in a leadership role within their workplace. For good measure we also allow anyone who can round up 100 signatures in their support.

Our associated nondiscrimination statement might read that this eligibility requirement must not place a heavy burden on eligibility on any member of society who is recognizable to their community as a leader.

I hope the example of Member of Congress can serve as a useful intuition pump about what an eligibility requirement for a SIEVE could look like. I don't intend to propose that we replace district or statewide voting for Congress with SIEVE. Instead, I'm presenting it as a tool that will be useful in a number of circumstances, especially in cases where we can build an obvious eligibility pool. SIEVE with eligibility gets us an opportunity to reduce the incentive for those who want power to seek it, and lets us choose the best of a great lot.