Knowledge Extractable Voting for Blockchain & Distributed Governance

2. A Radical New Approach: Knowledge Extractable Voting

2.1 Assumptions

The idea of knowledge extractable voting is based on the assumption that wealth is unequally distributed — a fact today in the real world. It’s also based on the assumption that wealthy and rich people do not make choices in favor of the society, but rather rational choices in order to maximize their utility. Quoting Vitalik Buterin and Glen Weyl from their recent post on Liberation through Radical Decentralization:

After all, a system that formalizes only capital and not human individuality may inexorably serve wealth rather than humanity.

The canonical example is the industrial revolution, showcasing the conflict between employers interested in productivity/revenue and employees having been brutally exploited to this end. Basing a decision mechanism purely on wealth (expressed in tokens) is limiting, as it might rule out voters that have the desired expertise, but cannot afford the participation.

2.2 Quadratic Voting and Sybil Attacks

A radically new idea was proposed by Steven Lalley and Glen Weyl to mitigate the problem of unfair wealth distribution. The authors put forth the beautiful notion of Quadratic Voting (QV). Their idea is based on buying votes. A bit more precise, each voter can buy as many votes he wishes by paying the tokens in a fund with one caveat. The voter has to pay quadratically in the number of votes. The money is then returned to voters on a per capita basis. Suppose, for example, a voter intends to cast 10 votes. Then he pays 10^2=100 tokens to acquire the votes. On a high level, the quadratic pricing function acts like a wealth slow down mechanism. Lalley and Weyl have proven under certain assumptions QV to be a mechanism against a tyranny of majority stake holders.

While their results apply to real-world decision makings, transferring the scheme to the permission-less blockchain setting does not carry over with the expected outcome. The problem with the blockchain world are sybil attacks. The design of blockchain technologies allows a voter to cast many anonymous identities. Hence, to accumulate 10 votes, the sybil attacker simply creates 10 accounts under different identities. This way, the attacker requires 10 tokens in total to cast 10 votes. However, we would like to stress that QV may satisfy the desired outcome in the case of permission-based settings where the identities of the players are known and fixed in advance throughout the lifetime of the system(For example a proof of authority based system).

Inspired by the radically new and brilliant ideas behind QV, we present a scheme basing decisions on something we believe is sparse and better suited for blockchain applications — namely knowledge — to achieve a decision (partially) independent of wealth.

The beauty of our proposal derives from the fact that blockchains empower the tokenization of knowledge.

2.2 Mechanism Design

2.2.1 Design Goals

Every voter should be given a voice. This mechanism should not be based on what position one holds, their wealth/stake, skin color, or gender. Further, the mechanism should not discriminate, but instead incentivize voters to do a solicit analysis before making an important decision. Further the mechanism should reflect the values, norms, and ideas of a community. A community typically shares intent, belief, needs and risks in common affecting the identity of the participants and their degree of cohesiveness. The blockchain community shares, for example, the belief in and needs for decentralization. What makes a community unique is their knowledge in a particular domain.

As opposed to wealth knowledge is acquired through experience or education by perceiving, discovering, or learning. It can’t be bought on an exchange. It can’t be transferred from a knowledgable to less-knowledgable or wealthy person. Moreover, knowledge is non-fungible, as knowledge relates to a particular field of interest and expertise.

2.2.2 The Knowledge Extractable Voting

The novelty of our mechanism is to leverage a two-token model. The first, called ETH, is a staking token. The second, called KNW (for knowledge), is a non-fungible token.

Protocol Goal: Make a decision (e.g., is this a good new block).

Protocol Assumption: ETH tokens are fungible and tradable. They can be exchanged and transferred. KNW are non-fungible and non-tradable. Further they are non-transferable because they are linked to a particular ETH wallet address. (This captures the intuition of knowledge from the above discussion.)

Protocol Setup: All players in the network are in possession of ETH tokens. Suppose for ease of exposition that each players owns KNW tokens as well.

Protocol: The protocol runs between Alice who acts as a proposer and a set of community members, abbreviated as the voters, in the following way:

  • Challenge(): Alice stakes some tokens to initiate the decision.
  • Response(): The voters counter-stake the same amount of tokens proportional to the total amount of voters. Each voter casts a vote.
  • Decide(): The quorum (e.g., majority, 2/3) decides on the outcome of the election. As reward for taking part in the process, the stake of voters deviating from the quorum choice (including that of Alice in case of she made a wrong proposal) is slashed and shared among the winning voters. In addition, for the winning voters KNW tokens are minted. Vice versa, the loosing voters are punished and some of their KNW are burned.

It remains to answer to what quantity the reward and punishment in terms of minting and burning KNW occurs. While the concrete parameterisation requires more research, we suggest the following lower and upper bounds:

  • The number of minted KNW tokens should relate to the outcome of the decision and reflect the closeness of the community interest. Suppose, for example, 100% of all voters agree on the decision, then the reward is close to 1 KNW token. Vice versa, suppose 51% of all voters agree on the decision, then the reward should be close to 0 KNW token. In which case it is not clear if the result reflects really the truth of the community.
  • The number of burned KNW tokens should stand in sharp contrast to the minting of KNW tokens. We suggest to unrelate the burn rate to the outcome of the decision and fix it to the square root of KNW tokens the loosing party has, i.e. suppose the loosing party has 16 KNW tokens, then after the voting protocol it possesses sqr(16)=4 KNW tokens.

The rationality behind the square root (i.e. inverse quadratic function) burn rate is to prevent wealthy voters from gaining KNW tokens simply through repetitive participation in the votings. Note, a wealthy voter may not fear to loose ETH token stakes or game the perceptual outcome of the decision by colluding with many voters.

2.3 Utility Function

Until now, we described the mechanics of the knowledge extractable voting scheme. It remains to utilize the gained knowledge in order to improve the voting outcome. We make three proposals:

  • Weighted Voting: the KNW token is used to weight votes. Ideally votes of more knowledgable voters shall have a higher impact on the voting outcome than those of less knowledgable voters. One example of such model may use KNW token as a multiplier. Let’s say that an individual has a good track record because he voted correctly on several previous occasions and helped successfully resolve disputes within the distributed network. In this case, his KNW token could be 1.5. Then, under the multiplier weighted model, his voting token will be multiplied by his KNW token which would make the vote worth more than someone who is not as trustworthy (whose KNW token is less than 1.5). Similarly, if he did not make wise decisions in previous voting rounds, his KNW token could be less than 1 (0.5 for instance). This would mean that his vote would count much less than someone who has shown to be more reliable in the past.
  • Selected Choice: the KNW token is used to identify voters and pre-classify them according to their expertise. This gives rise to a sampling function. For example, for an expert voting (i.e. a decision that asks for particular knowledge in a special domain) the sampling function would choose the top 10 voters with most KNW. For a mixed-expertise voting (i.e. a decision that asks for a fair representation of all knowledge groups), the sampling function chooses an equal amount of voters from every KNW level. Lastly, one can choose the voters at random to cast a group independently of the KNW distribution.
  • Delegated Choice: the KNW token is used to identify a delegate. As in the selected choice model, the KWN token serves the pre-classification of expert delegates. Through the activity (measured in terms of active voting participations) and the type of KWN tokens (recall, the KWN token is a non-fungible token; as such it relates to a particular field of interest, belief or value), voters can select the representative that suit best their interests.

Gaussian Distribution of Voters and some example Sampling Domains (Expert — Mixed — Non-Expert)

3. Applications

3.1. Proof-of-Stake Consensus w/Reputation

Knowledge extractable voting has applications in Proof-of-Stake based protocols (PoS). PoS is a type of protocol through which consensus can be reached in a distributed system. Ethereum is planning to use PoS for fairly determining the creator of the next block of the blockchain; however, there are numerous other uses for this algorithm to resolve different issues in decentralized enterprises.

In PoS disputes are resolved by validators. They are individuals who are willing and able to stake (lock up) some of their coins (ETH) or tokens as a collateral on the outcome of the dispute. When the dispute gets resolved, those who bet on the right outcome are rewarded proportional to their initial stake. While this protocol is a huge improvement over Proof-of-Work (which can require up to several thousand times more electricity resources over a lifetime of a system), there is still room for improvement. The current problem with PoS is that is tailored towards giving significant stake holders the ability to frequently participate in the protocol and obtain rewards. As wealth is unequally distributed, it leaves the majority of decisions to the minority of rich stake holders.

The knowledge tokens can be used to improve the staking mechanism in PoS distributed systems by rewarding those who take their role as a validator seriously and base their bets on knowledge instead of making a pure guess. It can also be set up to punish those who do not put any thought in the staking and betting process. Suppose, for example, that the selection of the block proposer happens on a mix of ETH and KNW tokens. This way, knowledge balance out wealth, and gives also the non-wealthy participants to engage in the consensus.

Therefore, KNW tokens provide an incentive system that promotes transparency and educated decisions. It is an essential piece that may improve the PoS algorithm even further resulting in a more reliable and efficient way to resolve conflict in decentralized enterprises.

3.2 Graded Token-Curated Registries

Graded or any form of token-curated registries (TCRs) is basically a list curated in a distributed way (i.e. without a central authority like google or amazon). Crypto-economic incentive mechanisms ensure that the owner of this list curates the content meticulously. TCR’s may be used for many things, such as: whitelisting websites, decentralized community-based map with points of interest, providing advertisers with websites most relevant for their content and many more. In simplest terms, they are basically community-vetted lists made of objects.

Knowledge extractable voting makes TCR’s to be even more accurate. In today’s world, an object within a TCR’s can still be compromised. A majority of people who did not have accurate knowledge on whether an object should truly be in the list (registry) could overpower several experts by pure chance. However, now with the KNW mechanisms, the probability of such scenario drops significantly. The system can make sure that there is a proportional distribution of experts, those who performed averagely, and those who performed poorly in the past voting rounds (represented in the green shaded area in the diagram above). This way token-curated lists effectively turn into knowledge-curated lists and have the advantage that for certain questions only the qualified people participate in the curation of the content.

read original article here