Don’t Underestimate the Design of DAOs (Decentralized Autonomous Organizations)
Kensuke ITO (a researcher at Startbahn, Inc.)
This article aims to highlight the difficulties of DAOs (Decentralized Autonomous Organization), a term which we have been hearing more and more since the Fall 2021, and the challenges in their design.
The definition of a DAO is still vague, but as far as I know, it first appeared in Buterin et al. (2014) as an example of “what might be possible with Ethereum¹:”
“The general concept of a “decentralized autonomous organization” is that of a virtual entity that has a certain set of members or shareholders which, perhaps with a 67% majority, have the right to spend the entity’s funds and modify its code.”
Reading this alone, you would think “hmmm, well, that’s not so bad,” but the technological history behind it―(1) Bitcoin presented a decentralized autonomous consensus-building for the legitimacy of transaction records (among a large number of pseudonymous users), (2) Ethereum made Bitcoin’s money transfer function programmable, thereby enabling the development of applications―will show us what it means. Namely, if we focus on “what it is for” rather than “what it is”, DAO is a concept that aims to extend Bitcoin’s novelty of “decentralized autonomous consensus-building” to be able to handle not only transaction records but also various contents related to the operation of an organization².
Is such a thing really feasible? This is a very essential question concerning the significance and potential of Ethereum, since the novelty would be lost if the application is managed and operated in a centralized manner. Therefore, I believe that DAO is a concept worth pursuing for all entities developing applications on Ethereum (including Startbahn, Inc., of course).
“How to Decide” are at the Core of DAO Design
At the time of this writing, there are many different DAOs (including self-proclaimed DAOs)³, but at least the major ones are all inseparable from the “decentralized autonomous consensus-building” function.
For example, the first application that claimed to be a DAO, The DAO (Jentzsch, 2016), was intended to be a decentralized autonomous investment fund⁴. The general framework is simple: (1) anyone can send any amount of ETH to The DAO, (2) The DAO sends back tokens in proportion to the amount of ETH received, and (3) token holders can receive investment income from the ETH accumulated in The DAO. Since there will be no fund manager, the token holders (who may have different values) will have to somehow decide where to invest the accumulated ETH.
On the other hand, nounsDAO, which has been highly regarded since its public launch in 2021, is a decentralized autonomous NFT community. The general framework here is: (1) nounsDAO generates one new NFT every day, (2) nounsDAO sells the generated NFTs in an ETH-paying auction where anyone can bid on them, and (3) NFT holders can use the ETH accumulated in nounsDAO to promote the NFTs (e.g., make T-shirts with the print of nouns NFTs, give ETH to developers). Since there is no marketing person⁵, here again, the NFT holders (who may have differing value of amounts) will have to somehow decide where to utilize the accumulated ETH.
This situation suggests that the design of DAOs are really a function of “how to decide” design.
Challenges of “How to Decide” Design
However, this “how to decide” aspect is actually a very difficult thing to design.
You may think, “Why don’t we just have a majority vote?”, but it’s not that simple. This is because, as soon as we assume a decentralized autonomous environment as a DAO, we are confronted with two problems that have not been discussed much so far: “Sybil attack” and “free-riding⁶”.
Sybil attack refers to all attacks in which one person creates multiple accounts and acts as if they are multiple other people⁷. This is an important issue for DAOs because a situation in which a decision made by 100 people was actually made by one person managing 100 accounts cannot be called decentralized. We cannot rely on a centralized entity (e.g., election commission, government) to ensure 1 person = 1 vote, as in an electoral system, so we must find some way to solve this problem in a decentralized manner⁸.
Next, free-riding refers to a situation where users continue to use system resources without contributing anything to the system⁹. This is also an important issue for DAOs because a situation in which most users do not participate in the consensus-building process in the end cannot be called autonomous. To consider the election example again, free-riding may correspond to the problem of low voter turnout. This problem is complicated, as once we address free-riding by offering a fixed reward for participating in the consensus-building process, it will then attract bad actor players (e.g., bots that answer randomly) who will be looking to participate for rewards only¹⁰.
Thus, “how to decide” in DAOs suddenly becomes difficult to design due to Sybil attacks and free-riding issues that arise from the goal of achieving decentralized autonomy.
In Search of the Best Solution
Unfortunately, a practical solution that simultaneously solves both problems has not yet been found.
In terms of the Sybil attack alone, both The DAO and nounsDAO have a countermeasure in the form of voting with (fungible or non-fungible) tokens. In other words, by treating 1 token = 1 vote instead of 1 account = 1 vote, the influence over the consensus-building process remains constant even if a large number of accounts are created (although the richer the account, the more influence the structure has).
So what about free-riding?
The DAO uses a mechanism called token staking, which adds the following structure to voting with tokens:
- Accounts can stake any amount of tokens on each option.
- The option that attracts the most tokens after a certain period of time is the consensus.
- Tokens staked on options other than the consensus are forfeited and distributed among accounts that staked tokens on the consensus¹¹.
In other words, it expects that (1) the token reward (obtained by voting for the consensus option) will provide an incentive for participation, and (2) the possibility of losing one’s tokens (by inappropriate voting) will prevent bad actor (noise) players from entering the system.
At first glance, the token staking seems to work well, but there are several problems. First, if the system rewards the option that attracts the most tokens, voters will predict which option everyone else is likely to choose and stake their tokens on it (so-called “beauty contest”). This is fatal to the “how to decide” design because it does not elicit the voter’s own beliefs. It is also questionable whether this system can solve the free-riding problem, even with the possibility of losing the staked tokens. In fact, in a simple token-staking scheme, the expected reward is zero, and even negative when the gas fee is taken into account.
And in nounsDAO, there are no specific countermeasures to address the free-riding problem. Voting with tokens is currently driven by the enthusiasm and engagement of the community, in which sustainability is unclear, rather than by economic incentives. nounsDAO adopts delegated voting (i.e., an address with an NFT can assign voting rights to another address) to avoid low turnout; however, there is no reward for those delegated¹².
Thus, if we try to solve both the Sybil attack and free-riding at the same time, it will end up failing or causing other problems¹³.
Final Thoughts
In this article, I have pointed out the difficulties of DAOs from a design perspective.
More specifically, we discussed how difficult it is to solve Sybil attacks and free-riding issues while ensuring decentralized autonomous properties in “how to decide” design, which is inseparable from DAO.
On the other hand, as I mentioned at the beginning,In fact, I am researching DAO design using a field called the peer-prediction method, and Startbahn, Inc. is continuing to study the extent to which and how the functions of its own product, Startrail, can be converted to a DAO.
In fact, I am researching DAO design using a field called the peer-prediction method, and Startbahn, Inc. is continuing to study the extent to which and how the functions of its own product, Startrail, can be converted to a DAO.
We intend to continue to publish our thoughts about these important aspects of the blockchain — stay tuned!
References
Buterin, V. (2013). Bootstrapping a decentralized autonomous corporation: part I. Bitcoin Magazine, 19.
Buterin, V. (2014). A next-generation smart contract and decentralized application platform. white paper, 3(37), 2–1.
Douceur, J. R. (2002). The sybil attack, In International workshop on peer-to-peer systems. Springer.
Jentzsch, C. (2016). Decentralized autonomous organization to automate governance. White paper, November.
Ramaswamy, L., & Liu, L. (2003). Free riding: A new challenge to peer-to-peer file sharing systems, In 36th annual hawaii international conference on system sciences, 2003. proceedings of the. IEEE.
[1] Strictly speaking, Buterin (2013) was the first to mention DAO in the form of DAC.
“But what if, with the power of modern information technology, we can encode the mission statement into code; that is, create an inviolable contract that generates revenue, pays people to perform some function, and finds hardware for itself to run on, all without any need for top-down human direction?”
Buterin et al. (2014) also classify DAOs into two types of subcontents:
“So far much of the talk around DAOs has been around the “capitalist” model of a “decentralized autonomous corporation” (DAC) with dividend-receiving shareholders and tradable shares; an alternative, perhaps described as a “decentralized autonomous community”, would have all members have an equal share in the decision making and require 67% of existing members to agree to add or remove a member.”
[2] Thus, Bitcoin and Ethereum are sometimes considered DAOs.
[3] For more information, see this article for example.
[4] The reason I use the past tense is that The DAO was eventually closed afterwards. This was due to a code flaw that led to the loss of accumulated ETH, and this incident also led to the subsequent Ethereum split.
[5] On the other hand, not all participants in a DAO have the same role. For example, there are entities with special authority called “curators” in The DAO and “nounders” in nounsDAO. Opinions vary as to whether such a situation can really be called a DAO. I think that if anyone can play this role (e.g., Bitcoin mining node), it can be considered a DAO.
[6] The subject of “how to decide” has been studied mainly in the fields of voting theory, social choice theory, and game theory (as the subject of “what rules should be used to aggregate the opinions of people with different tastes?”).
[7] “the forging of multiple identities” (Douceur, 2002, p.251). Therefore, just using Discord or Slack, for example, where one person can create multiple accounts, is not a DAO.
[8] Bitcoin and Ethereum, which have realized decentralized autonomous consensus-building regarding transaction records, have countermeasures against Sybil attacks with the rule that the more computing resources one has, the more influence one has over the consensus-building process (thus, no matter how many accounts one creates, the influence is the same).
[9] “an individual user who uses the system resources without contributing anything to the system” (Ramaswamy & Liu, 2003, p.1). Note that the usage of “free-riding” here is in the context of P2P systems, especially file-sharing systems, and differs from its usage in economics.
[10] Bitcoin and Ethereum have countermeasures against free-riding with the rule that they issue new Bitcoin/Ethereum as a reward only to the accounts that have proposed transaction records with enough supported from subsequent transaction records (thus only accounts that have diligently participated in the consensus-building process can be rewarded).
[11] The amount of reward is proportional to the amount of tokens originally staked.
[12] My personal suggestion is that nounsDAO should introduce a specification such as “the more an NFT holder participates in voting, the harder it will be for the characteristics of that NFT to appear in newly issued NFTs in the future” to prevent free-riding. Currently in nounsDAO, the characteristics of each NFT are completely random and there is no difference in rarity, but we can add a (posteriori) rarity level to each NFT according to the voting aggressiveness. I think this is an interesting way to take advantage of the characteristics of nounsDAO, which issues new NFTs every day.
[13] In addition, we need to consider another mechanism to put enough value on the tokens to provide an incentive to move people. And even if the tokens are valuable, there is even the question of whether people will work to maximize the amount of reward tokens in the first place. There is no end to these problems.
