Vitalik Buterin: DAO is not a company. Decentralization of autonomous organizations is very important
Original author: Vitalik Buterin, translated and edited by DeFi Zhidao.
Special thanks to Karl Florsch and Tina Zhen for their feedback and review of earlier versions of this article.
Recently, there has been a lot of discussion about the view that highly decentralized DAO is not feasible and the governance of DAO should start to be closer to the traditional corporate governance to maintain competitiveness. This argument seems familiar: highly decentralized governance is inefficient, while the traditional corporate governance structure consisting of the board of directors and CEO has been evolving for hundreds of years, constantly optimizing the goal of making correct decisions and creating value for shareholders in the changing world. The idealists of DAO naively assume that the decentralized egalitarian ideal can go beyond this, while attempts to do so in the traditional enterprise sector can only achieve negligible success at best.
This article will demonstrate why this position is usually wrong, and provide a different and more detailed view to explain in which cases different types of decentralization are important. In particular, I will focus on three important situations of decentralization:
- Decentralization can make better decisions in a concave environment. In this environment, diversified and even naive compromise forms may be better than the consistency and focus brought by centralization on average.
- Decentralization of anti censorship: applications that need to continue to run while resisting attacks from powerful external actors.
- Decentralization as credible and fair: DAO performs the functions similar to nation states in applications, such as providing basic infrastructure, so predictability, robustness and neutrality are more valuable than efficiency.
Centralization is convex, and decentralization is concave
See the original post:https://vitalik.ca/general/2020/11/08/concave.HTML
One way to categorize the decisions you need to make is to see if they are convex or concave. When making a choice between A and B, we will not consider the problem of A and B itself first, but a higher question: Would you prefer to compromise between A and B or toss a coin? In terms of expected utility, we can use charts to express this difference:
If a decision is concave, we prefer to compromise. If it is convex, we prefer to toss a coin. Generally, we can more easily answer the higher-order question of whether it is better to compromise or flip a coin than we answer the first order questions of A and B themselves.
Examples of convex decisions include:
- Response measures to the COVID-19 pandemic: 100% travel bans may help to keep the virus out. 0% travel bans will not stop the virus, but at least they will not cause inconvenience to people. However, 50% or 90% travel bans are the best of both worlds.
- Military strategy: It may be reasonable to attack Line A and Line B, but to divide your troops into two ways to attack only means that the enemy can easily deal with them one by one.
- Technology selection in encryption protocol: It may be meaningful to use technology A and technology B, but some mixing between them usually only leads to unnecessary complexity, and even increases the risk of mutual interference between them.
Examples of concave decisions include:
- Judicial decision-making: the average value between two independent judgments may be more likely to be fair than the random selection of one of the two judgments, but not completely absurd.
- Public goods funding: In general, it is more effective to provide $X to each of two promising projects than to provide $2 X to one project without giving the other. Compared with from X to 2X, having any funds has a greater role in promoting the ability of the project to achieve its mission.
- Tax rate: Due to the secondary deadweight loss mechanism, the tax rate of X% is usually only a quarter of the tax rate of 2 X%, which is harmful, and has more than half the effect in increasing income. Therefore, a moderate tax is better than a coin flip between low/no tax and high tax.
When the decision is convex, the decentralized decision-making process is easy to lead to confusion and low-quality compromise. On the other hand, when the decision is not clear, relying on the wisdom of the masses can give a better answer. In these cases, a DAO like structure with a large number of different inputs for decision making makes sense. In fact, people who view the world as a more concave place are more likely to see the need for decentralization in a wider context.
Should VitaDAO and Ukraine DAO become DAOs?
Many recent DAOs are different from earlier DAOs (such as MakerDAO), because earlier DAOs were organized around providing infrastructure, while newer DAOs were organized around performing various tasks around specific topics. VitaDAO is a DAO that supports early longevity research, and Ukraine DAO is a DAO organization and funding work related to helping Ukrainian war victims and supporting Ukraine's national defense. Does it make sense for these to become DAOs?
This is a subtle question, and we can get a possible answer by understanding the internal operation of UkraineDAO itself. A typical DAO tends to "decentralize" by aggregating large amounts of money into a pool and using token holders to vote to fund each distribution. On the other hand, UkraineDAO works by splitting its functions into many pods, where each pod works independently as much as possible. Top level governance can create new pods (in principle, governance can also provide funds for pods, although so far, funds only flow to external organizations related to Ukraine), but once a pod is created and resources are given, the rest will operate on their own. Internally, each pod does have leaders and operates in a more centralized manner, although they still try to respect the spirit of individual autonomy.
People may ask a natural question: Isn't this "DAO" a renaming of the traditional multi-level concept? I would say it depends on the specific implementation: of course, you can adopt this template and turn it into something with authority, just like a typical large company, but you can also use the template in a very different way.
Two things that help ensure that organizations built in this way will actually become meaningful decentralization include:
- A truly highly autonomous pod, in which the pod receives resources from the core. If they want to continue to obtain these resources, they will occasionally check their consistency and ability, but otherwise they will act completely independently and will not "accept orders" from the core.
- Highly decentralized and diversified core governance. This does not require a "governance token", but it does require more extensive and diversified core participation. Generally, extensive and diversified participation is a huge burden on efficiency. However, if (1) is met, then the pod is highly autonomous, and the core needs to make fewer decisions, then the impact of lower top governance efficiency will become smaller.
Now, how does this fit into the "convex and concave" frame? Here, the answer is roughly as follows: the (more dispersed) top layer is concave, and the (more concentrated in each pod) bottom layer is convex. It is usually better to give a pod X than to give it a coin flip between $0 and $2 X, and it will not cause great losses because of the "inconsistent" philosophy of compromise or guiding different decisions. However, in each individual pod, it is more important to have a clear, opinionated view to guide decision-making and to be able to adhere to many mutually collaborative choices.
Decentralization and censorship resistance
The reason why the decentralization of cryptocurrencies is most often cited publicly is anti censorship: DAO or protocols need to be able to operate and protect themselves against external attacks (including from large companies or even state actors). This has been discussed publicly and in detail, so there is no need to repeat it, but there are still some important nuances.
The two most successful anti censorship services used by many people today are Pirate Bay and Sci Hub. Pirate Bay is a hybrid system: it is BitTorrent's search engine, which is a highly decentralized network, but the search engine itself is centralized. It has a small core team, dedicated to keeping it running, and it uses the mole strategy to protect itself: when the hammer falls, it hides and reappears elsewhere. Pirate Bay and Sci Hub often change their domain names, rely on arbitrage between different jurisdictions, and use various other technologies. This strategy is centralized, but it enables them to achieve success in both defense and product improvement agility.
DAO is not like The Pirate Bay and Sci Hub; DAO behaves like BitTorrent. BitTorrent really needs to be decentralized for a reason: it needs not only anti censorship, but also long-term investment and reliability. If BitTorrent is shut down once a year and all its seeders and users are required to switch to a new provider, the network quality will rapidly decline. DAOs that require resistance to review should also fall into the same category: the services they should provide should not only avoid permanent review, but also avoid simple instability and destruction. MakerDAO (and Reflexer DAO that manages RAI) is a good example. DAOs running decentralized search engines may not: you can build a regular search engine and use Sci Hub style technology to ensure its survival.
Decentralization is credible and fair
Sometimes, the main concern of the DAO is not the need to resist the nation-state, but the need to assume certain functions of the nation-state. This usually involves tasks that can be described as "maintaining basic infrastructure". Due to the weak ability of the government to supervise the DAO, the structure of the DAO needs to have greater self supervision ability. This requires decentralization.
Of course, it is impossible to completely eliminate the hierarchy and inequality of information and decision-making power, but what if we can eliminate 30%?
Consider three incentive examples: algorithmic stable currency, Kleros court and Optimism retroactive financing mechanism.
- The algorithmic stable currency DAO is a system that uses on chain financial contracts to create encrypted assets. Its price tracks some stable indexes, usually but not necessarily dollars.
- Kleros is a "decentralized court": a DAO whose function is to make decisions on arbitration issues, such as "Has this GitHub submitted an acceptable submission for the reward on the chain?"
- Optimism's retroactive financing mechanism is an integral part of Optimism DAO. It retroactively rewards projects that provide value to Ethereum and Optimism ecosystems.
In these three cases, subjective judgment is required, which cannot be automatically completed by a piece of on chain code. In the first case, the goal is to measure some price indexes reasonably and accurately. If the stable currency anchors the US dollar, you only need the ETH/USD price. In case of hyperinflation or other reasons for giving up the US dollar, the stable currency DAO may need to manage the reliable chain CPI calculation. Kleros makes an inevitable subjective judgment on any arbitrary question submitted to it, including whether it should refuse to submit because of "immorality". Optimism's retroactive financing task is one of the most open subjective questions: which projects have completed the most useful work for Ethereum and Optimism's ecosystem?
These three cases inevitably need "governance", and governance is also quite stable. In all cases, whether from the outside or inside, governance can be attacked, which can easily lead to very big problems. Finally, governance needs not only to be strong, but also to convincingly convince the vast and distrustful public that it is strong.
The Achilles' heel of algorithm to stabilize currency: oracle
The stability of the algorithm depends on the oracle. In order to let the smart contract on the chain know whether the value of DAI is 0.005 ETH or 0.0005 ETH, it needs some mechanisms to learn the price of ETH/USD (outside the chain) information. In fact, this "oracle" is the main place where the algorithm stable currency can be attacked.
This leads to a security problem: algorithmic stable coins cannot safely hold more collateral, so they cannot issue more units than the market value of their speculative tokens (such as MKR, FLX...), because if they do, they will become half of the speculative tokens purchased, and use these tokens to control the oracle, It is profitable to steal funds from users by providing bad oracle value and clearing them.
This is a possible alternative design for the stable coin oracle: add an indirect layer. Reference ethreasear Ch's post:
We have established a contract with 13 "providers"; The answer to a query is the median of the answers returned by these providers. There is a vote every week, and the oracle token holder can change one of the providers... The security model is simple: if you trust the voting mechanism, you can trust the oracle output, unless seven providers are destroyed at the same time. If you trust the current set of oracle providers, you can trust the output for at least the next six weeks, even if you do not trust the voting mechanism at all. Therefore, if the voting mechanism is broken, the participants of any application that relies on the oracle will have time to exit in an orderly manner.
Please note the non corporate nature of the proposal. It involves depriving governance of the ability to act quickly, and intentionally spreading the responsibility of prophecy to a large number of participants. This is valuable for two reasons. First of all, it makes it harder for outsiders to attack the oracle, and it also makes it harder for new coin holders to quickly take over the control of the oracle. Secondly, it makes it more difficult for oracle participants themselves to collude to attack the system. It also reduces the extractable value of the oracle, where a single provider may deliberately delay publishing to realize personal profit from clearing (in a multi provider system, if one provider does not publish immediately, other providers will publish soon).
Fairness of Kleros
The "decentralized court" system Kleros is a very valuable and important infrastructure of the Ethereum ecosystem: Proof of Humanity uses it, various "smart contract bug insurance" products use it, and many other projects use it as a "final verdict".
Recently, whether the decision-making of the platform is fair has aroused some public concerns. Some participants put forward cases to try to claim from the decentralized smart contract insurance platform that they think they deserve. Perhaps the most famous of these cases is Mizu's report on Case # 1170. The case has evolved from a slight verbal interpretation dispute to a broader scandal, because some people accused Kleros's own insiders of coordinating efforts and investing a lot of tokens to promote the decision in the direction they want. Participants in the debate wrote:
The court's decision making process based on incentives... was, on all appearances, undermined by a developer who owned a very large (25%) stake in the court.
Of course, this is only one aspect of a broader debate, in which the Kleros community decides who is right and who is wrong and how to respond. However, from the point of view of this individual case, what is important here is that the whole value proposition of something like Kleros depends on its ability to convince the public that it is strongly protected from such centralized manipulation. For something like Kleros to be trusted, it seems necessary that no one should own 25% of the shares in the High Court. Whether through more widely distributed token supply or more use of non token driven governance, a more credible form of decentralized governance can help Kleros completely avoid such concerns.
Optimism retroactive financing
The results of Optimism's first round of retroactive financing were selected through the second vote of 24 "badge holders". In the second round, more badge holders may be used, and the ultimate goal is to transfer to a system where more citizens control the retroactive allocation of funds, possibly through some multi-level mechanisms involving lotteries, subcommittees and/or authorizations.
There is some internal debate about whether there are more and fewer citizens: does "citizen" really mean closer to "senator", an expert contributor who has a deep understanding of Optimism ecosystem? Or should it be a position for anyone who has significantly participated in the Optimism ecosystem? Or somewhere in between? My personal position on this issue has always been to develop towards more citizens, and solve the problem of low governance efficiency through the second tier delegation, rather than adding sacred centralization to the governance agreement. A key reason for my position is the possibility of insider trading and self trading.
Optimism retroactive financing mechanism has always been aimed at combining with the forward-looking speculative ecosystem: public goods projects that need funds can now sell "project tokens", and anyone who buys project tokens is entitled to receive a large amount of retroactive financing compensation in the future. However, whether this mechanism can operate normally depends on whether the retroactive financing part works normally, and is very vulnerable to the damage of the retroactive financing mechanism. Some example attacks:
- If some people have decided how they will vote on an item, they can buy (or short if the price is too high) their item tokens before releasing the decision.
- If some people know that they will decide on a particular project later, they can buy project tokens in advance and then deliberately vote for it, even if the project is not actually worth funding.
- The fund decision-maker can accept bribes from the project.
There are usually three ways to deal with these types of corruption and insider trading:
- Retroactively punish malicious decision makers.
- Actively filter to obtain higher quality decision makers.
- Add more decision makers.
The business community usually focuses on the first two, the first using financial supervision and wise punishment, and the second using face-to-face interviews and background checks. The decentralized world has few opportunities to use such tools: project tokens may be traded anonymously, the amount of recourse of the DAO to the external judicial system is limited, and the remote and online nature of the project and the desire for global inclusiveness make it more difficult for it to do background research and informal on-site "olfactory testing" of personality. Therefore, the decentralized world needs to pay more attention to the third technology: allocate decision-making power to more decision-makers, so that each decision-maker will have less power, so collusion is easier to report and expose.
Should DAOs learn more from corporate governance or politics?
The "big idea" of Curtis Yarvin, an American philosopher, is that companies are more efficient and better optimized than the government. Therefore, we should improve the government by making the government look more like a company (for example, far from democracy and closer to the monarchy). Recently, he wrote an article expressing his ideas on how to design DAO governance. Not surprisingly, his answer involves drawing on the concept of traditional corporate governance. From his introduction:
Since the beginning of the industrial revolution, the basic design of British and American limited liability companies has remained basically unchanged - reverse historians may argue that the industrial revolution may actually be an enterprise revolution. If the shareholding system design is not completely optimal, we can expect it to be almost optimal. Although there are classification differences between these two types of organizations - we can call them first order (sovereign) and second order (contractual) organizations - it seems that this year's society has very effective second order organizations, but there is no very effective first order organization. Therefore, we may know more about the second order organization. Therefore, when designing DAO, we should start from corporate governance, not politics.
Yarvin's post correctly identified the main difference between "first level" (sovereignty) and "second level" (contract) organizations - in fact, this exact difference is the theme of my above post on trusted fairness. However, Yarvin's post soon made a big and surprising mistake. He immediately turned to corporate governance as a better starting point for how DAO should operate. This error is surprising, because the logic of this situation seems to almost directly imply the opposite conclusion. Because DAOs are not superior to their sovereignty, and are usually explicitly engaged in providing services normally reserved for sovereignty (such as currency and arbitration), this is the design of sovereignty (political science), not the design of corporate governance. DAOs have more to learn.
It is commendable that the second part of his post does advocate an "hourglass" model, which combines a decentralized alignment and accountability layer as well as a centralized management and execution layer. However, it has been recognized that DAO design requires at least a lot of learning from first level organizations and second level organizations.
The reason why sovereignty efficiency is low and company efficiency is high is the same as the reason why number theory can prove many things but abstract group theory can prove much less: companies fail less and accomplish more because they can make more assumptions and have more powerful tools to use. If necessary, the company can rely on local sovereign countries to defend them and provide external legal systems they can rely on to stabilize their incentive structure. On the other hand, in sovereign countries, the biggest challenge is usually what to do when the incentive structure is attacked and/or faced with the risk of total collapse. No external giant is ready to support it.
The biggest problem in the design of a successful sovereign governance system may be what Samo Burja calls the "inheritance problem": how to ensure the continuity of the system when it is managed by one group of people and retired by another group. Burja wrote that companies usually do not solve this problem at all:
Silicon Valley is keen on "subversion", because we are used to the succession problems that remain unsolved in discrete institutions such as companies.
DAOs will eventually need to address the issue of inheritance (in fact, given the absolute frequency of the "get rich and retire" pattern among early adopters of encryption, some DAOs already have to deal with inheritance). The form of monarchy and similar companies is often difficult to solve the problem of inheritance, because the institutional structure is closely linked with the habits of specific people, which is either difficult to hand over, or the risk is high. More decentralized political forms such as democracy have at least one theory on how to make a smooth transition. Therefore, I believe that for this reason, DAO needs to learn from the more liberal and democratic political school, rather than from corporate governance.
Of course, in some cases, the DAO must complete specific complex tasks, and it may be a good idea to use a form similar to that of a company to complete these tasks. In addition, the DAO needs to deal with unexpected uncertainties. A system designed to operate in a stable and unchanging manner around a set of assumptions, when faced with extreme and unexpected changes in these situations, it really needs some brave leader to coordinate the response. A typical example of the latter is the stable currency dealing with the collapse of the US dollar: when a stable currency DAO evolves around the assumption that it is only trying to track the value of the US dollar, what happens when it suddenly faces a world where the US dollar is no longer traceable, and needs to quickly switch to a certain CPI?
If the dollar is no longer a viable reference asset, the stylized chart of the internal experience of the RAI ecosystem will unexpectedly transition to a CPI based system.
Here, methods inspired by corporate governance may look better, because they provide a ready-made model to deal with the problem that the founders organize a fulcrum. But facts have proved that the history of the political system also provides a model that is very suitable for this situation, and covers the problem of how to return to a decentralized model after the crisis: the Roman Republic elects a dictator with a temporary term of office to deal with the crisis.
In fact, we may only need a few DAOs that look more like political science than corporate governance. But these are really important. A stable currency does not need to be efficient; It must first be stable and decentralized. A decentralized court is similar. A system that guides financing for a specific cause - whether Optimism retroactive financing, VitaDAO, Ukraine DAO or others - is being optimized for a purpose more complex than profit maximization. Therefore, a coordinated solution other than shareholder profit is needed to ensure that it keeps the funds used for the intended purpose.
So far, even in the cryptocurrency world, the largest number of organizations will become the second tier organizations of "contracts", and ultimately rely on the support of these first tier giants. For these organizations, the simpler and leader driven forms of governance that emphasize agility are usually intentional. But this should not disperse the fact that this ecosystem will not survive without some non corporate forms of decentralization to maintain the stability of the whole thing.