09 May 2024 10:21

Yet Another Inadequate Placeholder

Considered morally, the fundamental case for democracy is that governments constantly need to change policies and personnel, and democracy is the way of arranging this peacefully. This is a basic conviction which I imbibed from Popper as a teenager, and which I realize is neither sophisticated nor inspiring, but also one from which I have never waivered. I would however add the claim --- which I think Popper would have endorsed --- that it is good for rulers to be accountable to the ruled, and democracy, again, is the way of arranging this. (At least, accountability of rulers to those whose power they wield, if not always accountability to those over whom the power is wielded.)

Considered cybernetically, democratic decision-making and debate are a series of feedback mechanisms for ensuring such accountability. (I can just imagine how hard I'd roll my eyes at anyone else writing that sentence, but I can't think of another way to put it.) Like any set of feedback mechanisms, they can fail to work properly, and it's worth thinking about democratic institutions in those terms, to see how to improve them.

(My dislike of "nudge" policies comes from this: if you don't realize you're being subjected to a policy intervention, you have no way to complain or object or even demand improvements. But, in a democracy, complaints, objections and demands from those on the receiving end of a policy are feedback signals.)

Democratic deliberation as a mechanism for collective cognition is yet another, and very important, topic.

Rambling: Is representative democracy really necessary for large groups?

(The following is basic enough that I'm sure the argument isn't original, but if I've read it before, I've forgotten where. It's not just Oscar Wilde's quip that the problem with socialism is that it'd take up too many evenings, though it is that too. Everything down to the horizontal rule was first drafted in the fall of 2022, after I [finally] read Michel's Political Parties.)

A finger exercise on participation, deliberation, representation etc.

Suppose a group of 40 people is prepared to spend 4 hours deciding what to do about some topic. (This is a bit larger than the number of voting faculty in my department, and a bit longer than our faculty meetings.) This time goes into hearing proposals, deliberating, bargaining, discussing, and finally voting (or reaching consensus or whatever other decision-procedure you like). On average each member can speak for 4 hr / 40 = 0.1 hr = 6 minutes. This is not that much for a complex or contentious issue, but let's take it for a baseline.

If we now imagine a group of 400, if each member still contributes for six minutes each, the group needs to spend 40 hours on deliberation, i.e., a full work-week. If the members of the group do nothing except deliberate full time, they can make only 50 decisions a year. (Clearly I am assuming American work-weeks and lack of holidays and vacations.) Alternatively, keeping the time per decision to 4 hours reduces the average participant to a contribution of 0.01 hr = 36 seconds. (Or of course some combination of a longer process and shorter contributions.) At 4000 members, we're looking at 10 weeks per decision --- while still only allowing 6 minutes per group member! --- or an average time of 3.6 seconds. A city of 400,000 --- about 1/3 bigger than Pittsburgh, as I write --- being 10,000 times larger than our initial group of 40 would either need to spend 20 working years (=40,000 hours) to reach a decision, or reduce the average contribution to 0.036 seconds (about the duration of a single frame of video). For a country with hundreds of millions of inhabitants, of course, this math is even worse.

In a phrase, the time needed for a group of size \( n \) to make a directly-democratic decision has to be at least proportional to \( n \), purely to allow everyone to contribute to the discussion. (Or the average contribution must be \( \propto 1/n \). ) The time-complexity is linear in \( n \), which is "feasible" if you're a theoretical computer scientist, but practically unacceptable.

At this point, it becomes clear that in any substantial group, most members cannot participate in the deliberation on most matters of common concern; they can at most vote. In a large democratic group, the set of active deliberators on a given topic must be sparse. Now, nothing in the arithmetic insists that the same people be active on every topic. The active set could be different for every decision, perhaps chosen by lot, by volunteering based on interest and knowledge, by appointment by the assembly-as-a-whole, etc. (Though in the last case, isn't that another decision?)

With those caveats acknowledged, though, the simple math above makes a compelling case for representative rather than direct democracy in large groups. Make deliberation and decision-making a specialized function, which only some of the group has to spend time on, but then let them spend substantial time on that. The representatives are accountable for their acts to those they represent, however, in a way which would not be achieved by (say) random choice. Of course to be held accountable actions (including deliberations) need to be public, a free and aggressive press helps, etc., through the rest of basic civics.

Representatives are also more accountable than volunteers. Non-representatives who insist on participating in a particular decision will tend to either be those who can afford the time, and/or those who are interested in the outcome, in both senses of the word "interested". Now, I think advocating for one's self-interest is not just inevitable but also legitimate. Just the same, measures which would give large benefits for a few but impose diffuse costs on many will always have dedicated supporters; stacking the group's decision-making procedures in favor of those beneficiaries seems unwise. Thus the basic argument against attempting to combine accountable representation with voluntary participation.

The bottleneck of serial attention

The reason the math above works out is that I add up the time needed for each member of the assembly to make a contribution of 6 minutes. You could imagine, instead, that they all record their contributions at once, and they're done. Behold: through the magic of parallelism, we've gone from linear time to constant time. The flaw in this, though, is that if everyone just talks into the void, and nobody listens, we don't really have anything we could call discussion, debate, or deliberation. Those require the members of the assembly to listen to each other, i.e., to pay attention to each others' contributions. And attention is necessarily serial, not parallel, at least in human beings as presently constituted. So the impossibility of equal participation in large democratic groups rests on the bottleneck of serial attention.

Evading the bottleneck without representation

That suggests some ways in which it might be possible to evade the bottleneck, or at least (to continue the metaphor) widen the opening. Three which come to mind are hierarchy, networks, and automation. All of them would work by keeping people from having to listen to most of their fellow citizens.


Take our city of 400,000 and break it down into 10,000 little groups of 40. Each of them takes 4 hours to deliberate, and then sends a delegate to a council with 39 other delegates from other little groups. Those 40 delegates are tasked with both conveying the views and decisions of their original group, and reaching a new deliberative decision. This then repeats, until at the 4th level up, we have a council of 7 delegates, which makes a decision for the entire city. A scheme like this in principle allows everyone to participate equally in the deliberation, but with a time that grows only logarithmically in \( n \), not linearly. The cost however is necessarily a lot of compression in going from each level to the next. (Remember, 4 hours of debate informing a delegate who gets to speak for at most 6 minutes.) There is also an institutional problem of how to constrain delegates to actually represent the views of their group, rather than their own. (Perhaps: decisions reached at higher levels have to be confirmed at lower ones? This'd only add a constant factor to the time complexity.) And of course the principle on which the hierarchy is organized will matter a lot! (Imagine geography of residence vs. occupation vs. sequential social security numbers.)

(If you want to say this doesn't completely evade representation, I won't quibble. If you think this sounds like the theory behind the soviets, well, I couldn't possibly comment.)


Instead of listening to everyone's contributions during deliberation, each member of the assembly gets assigned a limited set of partners, and they only have to listen to them. But instead of the strict hierarchy of groups I just sketched, everyone gets a random sample of (say) 39 other members of the assembly to discuss with. This ensures that everyone is exposed to a representative sample of initial opinions. But it's easy to arrange this network so that it's "connected", that there's a path between any two citizens, and so, again, information can propagate from anywhere to anywhere in the network. (Indeed, with 39 links per assembly member, it's actually hard to avoid having a connected network.) It's harder to be precise about the time complexity, but a guess would notice that the "diameter" of such a network, the number of steps needed to go between its two most distant members, is only \( \propto \log{n} \), so allowing \( \log{n} \) time would seem to let arguments and ideas wash back and forth, without requiring everyone to listen to everyone else.

(This is obviously inspired by Watts and Strogatz (1998), and by Ober's account of the Athenian "tribes", but also by the "philosopher cells" in Linda Nagata's Vast.)


This final possibility is more speculative even than the other two. For better or worse, lots of members of a democratic group will have very similar values, interests and ideas. Their contributions to the deliberation will be very similar. My calculations above assumed that if five members of assembly will all say (almost) the same thing, everyone will have to sit there and listen to all five, one after the other, because there wasn't any way of grouping them together, so one of them addresses the assembly and the other four make noises and gestures of approval. Now this might admit of a technological fix. Go back to the scenario where everyone speaks at once, in parallel. Now imagine natural-language-processing software running over these contributions and clustering them. (This'd probably work better for written contributions than for spoken ones, but then, since this is much better than anything we have today, let's assuming speech processing is solved as well!) Then, instead of having to listen to each individual, members of the assembly only have to listen to each cluster. We wouldn't even need to assume an automatically-generated summary of the cluster's contribution, just pick a random member of the cluster to listen to. Technologically, of course, we're nowhere near this, but it's not unimaginably out of reach. There would, naturally, be all sorts of higher-order wrangling, since everything from the ability of the system to process non-standard dialects to the exact clustering procedure could be a point of contention.

Speaking of compression: I need to think more about the idea that what gets debated on is a kind of lossy compression of the actual policy or action, with the de-coding into actions being done by the state apparatus. (Or even: by the interaction of the state apparatus and society.) The lossiness would be an issue for any system of government, but it might be a special problem for democratic deliberation. (Mightn't Lindblom's "disjointed incrementalism" be in part a way to compensate?) I touched on this a decade ago in "In Soviet Union, Optimization Problem Solves You", but it needs much fuller treatment, which'd have to bring in Ashby's "law of requisite variety".

Some Inspirational Quotes

(Which may not have been intended inspirationally...)
The democratic city is the city in which every one of its inhabitants is unrestrained and left to himself to do what he likes. Its inhabitants are equal to one another, and their traditional law is that no human being is superior to another in anything at all. Its inhabitants are free to do what they like. One [inhabitant] has authority over another or over someone else only insofar as he does what heightens that person's freedom.
Thus there arise among them many moral habits, many endeavors, many desires, and taking pleasure in countless things. Its inhabitants consist of countless similar and dissimilar groups. In this city are brought together those [associations] that were kept separate in all those [other] cities --- the vile and the venerable ones. Rulerships come about through any chance one of the rest of those things we have mentioned [when describing other cities]. The public, which does not have what the rulers have, has authority over those who are said to be their rulers. The one who rules them does so only by the will of the ruled, and their rulers are subject to the passions of the ruled. If their situation is examined closely, it turns out that in truth there is no ruler among them and no ruled.
Yet those who are praised and honored among them are [a] those who bring the inhabitants of the city to freedom and to everything encompassing their passions and desires and [b] those who preserve their freedom and their diverging, differing desires from [infringement] by one another and by their external enemies while restricting their own desires only to what is necessary. These are the ones among them who are honored, [deemed] most excellent, and obeyed...
Of [all] their cities, this is the marvelous and happy city. On the surface, it is like an embroidered garment replete with colored figures and dyes. Everyone loves it and loves to dwell in it, because every human being who has a passion or desire for anything is able to gain it in this city. The nations repair to it and dwell in it, so it becomes great beyond measure. People of every tribe are procreated in it by every sort of pairing off and sexual intercourse. The children generated in it are of very different innate characters and of very different education and upbringing. Thus this city comes to be many cities, not distinguished from one another but interwoven with one another, the parts of one interspersed among the parts of another. Nor is the foreigner distinguished from the native resident. All of the passions and ways of life come together in it. --- al-Farabi, The Political Regime, 113--115

The "diffusion of democracy" literature probably needs an extra notebook. My impression is that it all runs smack in to the homophily-vs-contagion issue, but I should really investigate it carefully before pronouncing judgment.