DeSoc’s Dark Heart
DeSoc, for all its talk, has a very dark element as proposed
There has been much talk recently of a new paper by Weyl, Ohlhaver and Buterin on soulbound tokens and their use to create Decentralised Society or DeSoc. The paper provides a potential way of powering web3 by virtual of tokens that are attached and used to establish unique identities on the web. That sort of thing is critically important for web3 as I discussed here. But read the paper carefully and there is worrying element with regard how it might use those identities in practice.
It starts with the common issue that many who push for blockchain solutions to things worry about … decentralisation.
When analyzing real-world ecosystems, it is desirable to measure how decentralized the ecosystem actually is. To what extent is the ecosystem truly decentralized, and to what extent is the decentralization “fake” and the ecosystem de-facto dominated by one or a small set of coordinating entities?
The reason this is a concern is that the whole point of establishing unique identities is so that governance is not subject to coordination. In blockchain circles one such concern is the 51% attack where a single entity controls the majority of the ‘votes.’ But there is more than that:
For example, nominally independent firms may have many major shareholders in common, have directors who are friends with each other, or be regulated by the same government. In the context of token protocols, measuring decentralization of token holdings by looking at on-chain wallets is wildly inaccurate because many people have multiple wallets, and some wallets (e.g., exchanges) represent many people. Moreover, even if addresses could be traced back to unique individuals, those individuals could be socially correlated groups prone to accidental coordination (at best) or intentional collusion (at worst). A better way of measuring decentralization would capture social dependencies, weak affiliations, and strong solidarities. (emphasis in original).
To do this the authors propose “a protocol [that] could examine the correlations between SBTs held by different Souls and discount votes by Souls (pooling them as only partially separate) if they share a large number of SBTs.” What do they mean by this? There isn’t much discussion but what they mean is plain for all to see in the mathematics left in an appendix to the paper.
Which votes count for what?
Suppose you have three people who are part of a DAO (decentralised autonomous organisation). They are deciding on something and they could vote on what to do based on quadratic voting that has some nice properties (but that isn’t critical to the story). They show that three voters — Abdu, Shou and Belle — can vote for good outcomes if they do so independently.
But what if some of the voters have something in common, such as where they work?
Now suppose a simplified model where Abdu, Shou and Belle are differentiated by a single membership — workplace — and matching funds are available for startups, companies, and open-source projects (again, in the spirit of Gitcoin). Because people from the same workplace have a strong incentive to contribute to their own workplace to maximize matching funds to their company, we should expect them to coordinate. An extreme approach would be to assume that workers fully share goals and fully coordinate their behavior.
They don’t take the extreme approach. Instead, the authors argue that a “simple approach, which we call “clustering,” would put two co-workers “under the same square root” in the quadratic formula to offset their tendency to already coordinate.”
There is a lot going on there. Basically, the move is away from one person, one vote to taking the votes of people who are presumed to have some common interest — in this case, where they work — and then putting them together to weaken their force. In other words, their votes count for less!
And it is not just one dimension but potentially a whole lot of dimensions that could determine your power in voting before getting a chance to vote.
The previous example assumes Abdu, Shou and Belle have a single membership: workplace. Yet in almost all applications this would be a vast oversimplification. People have multiple community memberships, cooperative relationships, and even informal intersections. Abdu and Belle might be extended family, Shou and Belle might have attended the same school, or Shou and Abdu might be token-holders of the same layer 1 protocol, and so on. To facilitate cooperation across differences, these correlations in memberships between individuals need to be recognized in a less binary manner.
Now I’m not going to necessarily equate the authors’ motives with voter suppression tactics that are used except that they really seem to be that. I mean what else do we call limiting the voting power of people with certain affiliations and characteristics?
A Slippery Slope
Why are the authors led down this dark path? The reason is that something like quadratic voting can really be exploited when people coordinate their actions. In that way, they can get together pre-vote, work out a plan that is good for them as a group and obtain that outcome via coordination which is better for them than what is likely to arise if they all vote independently. One way to deal with that is not to use quadratic voting. The other way is to adjust this. It is that path that the authors lean into.
But rather than treating pre-existing cooperation as a bug we ought to “write over,” the key is to acknowledge it as reflecting partial cooperation that we should harness and compensate for. After all, we are in the business of encouraging cooperation. The trick is to make quadratic mechanisms work alongside pre-existing networks of cooperation, correcting for their biases and tendencies to over-coordinate. SBTs offer a natural way by allowing us to tip the scales in favor of cooperation across differences. As Nobel Laureate Elinor Ostrom famously highlighted, the problem is less coordinating public goods per se but rather one of helping communities made up of imperfectly cooperative but socially connected individuals overcome their social differences to coordinate at scale in broader networks.
But there is a reason that we don’t like fiddling with characteristics and affiliations this way — it makes some people more equal than others. And who is to say how this should occur? The whole point of voting on stuff is to surface what we don’t know and that is more valuable precisely when we can’t have the knowledge to presume that we will know what an outcome will be. This is a slippery slope towards exclusion which would surely be anathema to the whole exercise.
The authors seem enamoured by the autonomous part of DAOs rather than the need to engage in policing explicit collusion or coordination — i.e., vote selling. Indeed, almost as an afterthought in the appendix, the authors discuss “pairwise matching” which is based on the idea of one of them that monitors the network for specific attacks. They realise its advantage in that “it has the important advantage that it does not require an extrinsic source to specify who is coordinating and who is not; instead, this information is extracted from the contribution values themselves.” There are dangers here but it is akin to policing voter fraud rather than pre-emptively discounting the votes of some people.