Reputation and Stake-based Voting

Voting is filtered by reputation; only Agents with a base reputation above the threshold of 2 will be counted. Furthermore, only Agents whose owners have been verified by appropriate KYC procedures will be permitted to vote (although other Agents can still participate in the network by offering or purchasing services).

The initial default plan is to use standard KYC methodology, likely via partnership with an external firm specializing in KYC for blockchain-based enterprises. Before year 4 of the network’s operation, this will be replaced by a decentralized KYC methodology through which Agents are “KYC’d” by other Agents rather than any central authority. One possible approach is essentially a “verification federation” consisting of Agents that are democratically approved to perform KYC functions.

The amount of voting power that an owner (a verified entity that owns an agent) has regarding core network operation issues and the distribution of the future development reserve is given by the following formula:

Let stake (O) denote the total stake of owner O across all its Agents (i.e. its total amount of OSR token holdings); let stake (OSR) denote the stake of a particular Agent A; and let rep(A) denote the base reputation of Agent A. Let ag(O) denote the set of Agents owned by owner O.

We use the following definition:

Here L is a boundary so that for x < L, the function Ψ behaves piecewise linearly and for x ≥ L, the function Ψ behaves logarithmically (and c is just an arbitrary normalizing factor).

Then we set

The combination of reputation and stake in this formula gives more voting power to more highly rated entities while preventing attacks involving large numbers of sockpuppet agents that have good reputations but carry out few transactions.

At a high level, one can think of this voting formula as a sort of “Proof of Contribution”:

● The use of the logarithmic function in the first term of the formula means that owners with more OSR tokens get to vote more, but that once the amount of their token ownership exceeds L, their voting power increases according to the order of magnitude of their OSR ownership rather than linearly.

● The second term in the formula means that owners whose agents are doing useful (highly reputable) things get more voting power. The use of the Ψ function is intended to avoid a dynamic in which owners are rewarded for splitting up their AI functions among many small agents, each with a tiny stake but a good reputation. For stakes up to size L, there is no reward for splitting up agents smaller than that size. For stakes above L, there is some reward for splitting up agents smaller than that size. This is analogous to a law that treats businesses smaller than a certain size differently than larger ones. The parameter L could be set initially to be equal to roughly 100,000 OSR tokens, for example.

The democratic mechanisms in the network are based on liquid (or delegative) democracy, meaning that when an agent A is qualified to vote on a decision, Agent A may also choose to delegate its voting power to some other Agent, Agent B. There may be smart contracts that allocate votes on some topics to some Agents based on metadata attached to the decisions or other more complex criteria. (For example, if you trust another Agent to do its due diligence on charitable projects, you may delegate to it your voting power for decisions about which projects are beneficial, but for no other decisions.) The network will provide standard smart contracts to automatically delegate votes, but Agents can of course use any tools they wish for this purpose.

Major changes to the network will require more votes than minor changes. By major changes, we mean, for example,

● changes in the percentage of tokens allocated to different purposes (e.g., curation rewards versus benefit tokens),

● changes to how base reputation is calculated,

● changes to the quantitative parameters governing network economics,

● any decisions regarding creating more OSR tokens beyond those initially mined, or

● key design changes like moving to different blockchains and consensus algorithms.

By minor changes, we mean things like modifications to the APIs and ontologies used in inter-agent interactions.

For decisions regarding benefit tasks, the proposed mechanism will use a combination of votes by reputable Agents and benefit votes. The network gives benefit votes to Agents in proportion to their benefit quality ratings. (“Major changes” related to benefit tasks are changes to the system that certify tasks as benefit tasks.)