The Wisdom of Crowds Only Appears Under Conditions, and the First Pillar Meets Them

Supporters of direct democracy like to cite the wisdom of crowds, and opponents like to cite the madness of them. Both are right about something, and the argument goes nowhere because neither side says under what conditions a crowd is wise. James Surowiecki's book of that title did say, and the conditions are specific enough to be checked against a proposed system. I want to check them against mine.

The Two Pillars

The model of direct democracy I have argued for elsewhere on this blog rests on two pillars.

The first is the voting pillar: the public decides the questions at the highest level directly. Anyone may put a proposal forward, anyone may argue for or against it in the open, anyone may take an existing proposal and put forward an amended version, and every citizen contributes a vote on the result. It runs continuously rather than once every four years, and it is confined to the large questions, not the technical administration beneath them.

The second is the balancing pillar: a mechanism that can check a decision of the first. It might be a veto held by an office, a right of appeal, a delay before a decision takes effect, or a jury that inspects how a decision was carried out. Its purpose is to catch the cases where the first pillar has been rushed or misled.

The wisdom-of-crowds argument applies to the first pillar, and I have never seen the connection drawn properly.

Surowiecki's Conditions

Surowiecki's claim is not that crowds are wise. It is that a crowd's aggregate judgement beats the judgement of its best members when four conditions hold, and fails badly when they do not.

Diversity of opinion. Each person brings some private information or a private way of reading the shared information, even if it is eccentric. A crowd of people who all learned the same view from the same source is not a crowd; it is one opinion counted many times.

Independence. People form their judgement without being steered by the judgements of those around them. The failure mode is the cascade: everyone defers to the early movers, and the aggregate reflects a handful of people plus imitation.

Decentralisation. People draw on local knowledge, the specifics of their own situation, rather than on a single central account of the facts.

Aggregation. There is some mechanism that turns the private judgements into a collective decision, and it counts everyone rather than letting a subset speak for the rest.

When these hold, errors in individual judgement are largely uncorrelated and cancel out, and what remains is the shared signal. When they fail, errors are correlated and reinforce each other, and you get the crowd the opponents are afraid of.

Mapping Them Onto the Pillar

Now put the first pillar next to the list.

Anyone may propose. That is diversity of input at the source: proposals are not filtered through a party that has already decided what is on the agenda. A proposal that would embarrass every party can still reach the vote.

Anyone may argue in the open, and anyone may copy a proposal and improve it. That is how a crowd refines rather than merely counts. It is also where decentralisation lives: the improvements come from people who know the specific corner of the problem the original proposer did not.

Everyone votes, and every vote counts the same. That is aggregation with no subset speaking for the rest, which is precisely the condition representative government violates by design, since the whole point of representation is that a subset speaks.

Independence is the hard one, and I want to be honest that the pillar does not supply it automatically. A public that votes after weeks of coordinated campaigning is exposed to cascades. What the pillar does offer is the structural remedy: frequent decisions on separate questions, so that no single cascade can carry a bundle, and open argument, so that the cascade meets its counter-argument in the same place. That is not a guarantee. It is a better position than one vote every four years on everything at once, which is a cascade by construction.

What Representation Gets Wrong on Every Count

It is worth noticing that the current system fails not one of the conditions but all four.

Diversity is destroyed by bundling: the party assembles a package and you take it or leave it. Independence is destroyed by the campaign, which exists to manufacture a cascade. Decentralisation is destroyed by the fact that a few hundred people in one building decide everything, drawing on the same briefings. And aggregation is replaced by delegation, which counts the public once, coarsely, and then counts a subset for four years.

So the objection that the public cannot be trusted to decide is aimed at the wrong system. The arrangement that most resembles a mad crowd, a mass steered by a few voices into a single all-or-nothing choice, is the one we have.

What This Does Not Prove

None of this shows that the public will decide well. Surowiecki's conditions make the aggregate better than its members; they do not make it right. A diverse, independent, decentralised, properly aggregated crowd can still be collectively wrong about a question where nobody has the relevant information.

What the mapping shows is narrower and, I think, sufficient. If you believe the wisdom of crowds is real under stated conditions, the first pillar is a deliberate attempt to build a decision procedure that meets them, and representative government is a procedure that meets none. That is a design argument, and it is the one I would want to be answered.