DAPPCON 2018: Holographic Consensus, Scalable DAOs - Matan Field (DAOStack)

Transcript

so hi everyone I guess if you want to be tired including myself so my name is my tan I'm from the Taoist project but I will focus today my talk on the holographic consensus and scalable dows so basically what I think is the most promising Avenue of blockchain that is doused and what I think is the most prominent problem of doubt which is how to scale their governance and what I think is a possible solution for that problem which is a low graphic consensus so this will be the focus of my talk okay so let's start slowly with what do you mean by douse and what do mean by launching

governance so these are this will be my three requirements for what I would expect from a dowel to satisfy so obviously firstly it should be decentralized right so by the centralization we mean there is a wide and open distribution of influence so it's not like a closed small set of voters or influencers or deciders and then there are two critical retreat area so firstly it has to be resistance it has to be resilient or resistance again manipulations right because otherwise there is no real governance system this is reflecting the influencers and secondly it has to be scalable so if

it's not scalable if it's small organization why to bother from the beginning like why to have a decentralized so the whole power of the centrist organizations that can grow and ideally grow indefinitely and I add here in parentheses I add super super growing or super effectively growing super effectively which means that in contrary to current organizations when they grow up the reflectivity goes down the effectivity per person so indefinitely scalable organization can grow up forever but superscalar organization become more effective as they grow up per person and we are familiar with those kind of structures we call the network effect because on the internet

it dinners itself and so on so can we construct an organization that behaved like a network effect such that becomes more and more effective per person as its scaled up so that will be the ideal scenario okay so what do you mean by blockchain governance so blockchain governance is a sort of like a function so given a proposal given an idea that we need to decide about it takes in the inputs of agents so there are n agents in the system and influencing age in the system it takes in those inputs and transfer order transform them into an outcome into an output okay that's true for each and every proposal so that's that will be

the definition for blockchain governance and then that's how it looks in terms of implementation in actual smart contract implementation we call this the arc a governance framework in Dow stock but going very briefly on that architecture given input in terms of proposals and then input in terms of the like the opinions on those proposals then there is the logics that trigger actions so we go call them governance schemes each governance scheme is a bit of logic that says if such-and-such happens then the organization does such-and-such then there is a these are the dues then there's other kind of bit of logic of governance we call in the global

constraints which basically says no matter what happens the organization will not do such and such so then there's the inputs of the agents the inputs go trigger the governance schemes the govern scheme trigger the controller the controller verifies with the global constraints that everything is okay and if there is there's action triggered it triggers the actors so there are two internal actors the token contract of that specific agency and the reputation contract of that specific agency I will say a few things about reputation and then there is also the Avatar contract which is basically the face of that organization the avatar is the contract that can do anything else in the blockchain so outcomes will be like will

look like printing tokens of that agency this reputation of that agency or basically doing anything inside the blockchain okay so this is this is just like a kind of like a side remark so that the centralized governance is a consensus protocol between subjective agents with their subjective inputs which agreement about subjective realities so it's kind of like generalization of what we know of consensus protocols in like the blockchain itself but the difference is that the the inputs that are coming from agents the trigger decisions are subjective inputs are not inputs that can be evaluated by a machine by a program and that's why we need human

agents to evaluate those inputs so what kind of outputs what kind of decisions or realities we want to agree about so for example allocation of resources such as tokens to agents in the system between sent device actions may be made metadata so we want to decide about curation of objects and then finally we would like to decide about the governance system itself such as the protocol and perhaps the reputation distribution so maybe a side remark about reputation because it's I know it's kind of like a charged issue or charge matter in the community so what firstly let me just say what do I mean by reputation because there's a lot of

different people are speaking about reputation in different contexts or just setting up the language that I'm using so reputation is a numerical attribution given to each agent it's a non-transferable attribution so it's not a token and if you see there is an asterisk over here to say that you might be able actually to transfer reputation or you might like to have the transferability of reputation in a dao but not an ordinary transferability of a token so maybe transfer transfer ability under some restricted conditions and then finally that's the important line it simply amounts to influence or voting power in the governance system so in every governed system you have agents

that are expressing opinions and then you are collecting the some of those inputs and you are making them into a decision and then you need to wait different inputs differently so you have of course token based voting which way by tokens or by token staked and then if you have a reputation system you are weighing the inputs of agent by that number and then a lot of people are asking okay but how reputation flows in the system arbitration checks in the systems or patient flow amounts to either awarding of conservation of reputation for contribution of value so let's say I'm proposing a proposal and that proposal is eventually become successful so I will be awarded with rotation for making that successful

proposal or maybe I will submit code to the development community and again I will get reputation and the awarding of that reputation is part of the part of the protocol itself so it's part of the decision-making itself and then the second kind of reputation flow is that if I'm voting with my reputation so as far as I'm in alignment with the Dow my rotation increases when I'm in miss Lang with a Dow my repetition decreases and this is my well this is a side argument I'm not going to dwell into that but I think it's important so the argument is that being non-tradable reputation based voting systems are much more resilient to manipulation than token based voting

system so I will stick to reputation from now on and I will not explain that any further so let's now ask what is the biggest problem of doubt that we have or why why don't we see doubts already so let's let's go back to the to the two requirements from a decentralization it's sorry the centralized organization we want to have it scalable and we want to have it resilient but let's define those terms in a bit more details so firstly scalability is simply the ability to scale up decision-making capacity by scaling up the number of participating agents and more so the number of effective decision that that

the DAO can make in the unit of time so let's say that you know in a in an average Dow one person will make a proposal once a month so if you have a doubt of 100 people it means that you have a stack of 100 proposal in months to decide about okay that makes sense that's probably we don't need much technology for that but what about a dowel of 10,000 people that that will have 10,000 proposals in the stock that needs to decide about so the scalability of the Dow is the ability to process that 10,000 decisions in that months and then protocol resilience so let me let me make an intermediary or

sorry middle step definition so let's find the global opinion or the truth of that Dow so the global pinion Sadao is the decision that the Dow would have made if it would have a complete input set so by complete input set I mean if all agents in the system have the sufficient capacities and sufficient attention to look and consider each and every decision and make their opinion okay even if that opinion is I don't have an opinion that is sufficient to so assuming we have a complete input set assuming we have sufficient capacity sufficient attention in the system then the decision that they will make we call

them the global opinions now we'll define resiliency to be that decisions are in the sea in the system are made in line with the global opinion but now we are hitting a direct head-on collision and direct tension or a principled tension which is that if we require from the system to have too much attention on each and every decision so global decision if we actually require each and every agent to look at each and every decision then clearly then it makes the governance system not scalable right if we have 10,000 agents and 10,000 decisions to make in a month clearly we cannot expect all of them look at at each and every decision but

then if we have too little tension on each decision so we call them local decisions so maybe we are allowing for decision to be made by small amount of reputation or small amount of people that really makes the governesses and not resilient because those people are not any more representative or reflecting the opinion Saddam and in particular costs of manipulate the decisions so there is a strong tension between scalability and resilience and in fact any consensus consensus system including locking itself has the same principle tension between skill building resilience okay so a decentralized resilient and scalable governance system that's like the only by definition that's the

solution and centralized resilient and scalable governance system has to allow for local decisions right for sake of scalability that are guaranteed to be alive with a global opinion so you need to have some sort of magic you need to allow somehow for local decisions for scalability but those decisions have to be aligned with the global opinion for sake of resilience and that's exactly what we will define to be holographic consensus so graphic consensus is a decentralized governance system that allows for local decisions that are guaranteed to be in line with the global opinion so for the rest rest of my talk I will try to sketch how holographic consensus looks like both in principle

and in some details okay so firstly in summary Soho graphical cells as I said allows for local a decision that are guaranteed to be in line with the does global opinion let me define a little bit further what I mean by local decisions so local decisions are decisions that are made with relative majority so absolute majority is that we are asking all of the reputation holders in the system what they think and then the majority of all reputation holders say yes then that's that's an absolute majority a relative majority is for example when I ask a question and then I wait for one week I see who votes in that one week and I only consider the

majority of those voters so it's a relevant relative majority rather than the absolute majority of all reputation holders so that's of course very scalable approach but by itself is very unruly and or non resilient secondly how do we how we achieve that alignment how do you achieve that alignment between local decisions we just probe a slice of the system right that just a slice that has the capacity of attention to react to our question how do you make that process that local decision process to be guaranteed in alignment with the global decision so the answer is that we are creating a correct economic game so a

line is acquired by cryptic anomaly incentivizing the discovery off and the protection from misalignment and and I'll show you how it works this by the way if you're familiar with object computation protocols such as troubied or the precursor of a theorem compute computation market this is just a subjective analogue of that objective that's the objective limit of that protocol why we call it holographic well it's kind of like another goes to the hologram in the hologram you can look at any slice of the picture and you actually have the information for the entire three-dimensional in your image so in that sense it's kind of like analogous in the sense that you can

probe a slice of the system of the governing system and still get a faithful answer that reflects the entire system ok so let me quickly go with a prescription and then we'll dive into more details so firstly we start with default proposal so anyone can come up with any proposal and by default those who puzzles in order to be executed need to be approved by an absolute majority so that's of course by definition in line with the global opinion then as said before we have to have an extra stage like a extra phase we have to allow for booster proposal recall the booster proposals and those booster

proposals are open for a finite time let's say a week and after that time they are decided by relative majority so we have to have that stage or that phase however we need to protect that phase we need to protect which decisions can be boosted and how so the way it works is that predictors there is a predictor network and those predictors place predictions about the outcome of proposals and the way that they place those predictions is my place mistake basically staking Jen tokens they place a stake about the prediction about the predictions of our body outcome proposals then second state second step proposal are boosted only

upon sufficient positive signal so there are two kind of signals in the system and we'll again we'll get into it later there are two kind of signals the first signal is that predictors are looking at an ocean of a puzzle so there are like 10,000 proposals in the queue and then they're looking at the ocean of those proposals and flagging or signaling for those proposals which are interesting and they are willing to place a bet on that mistake on that so I'm I'm looking at 10,000 proposals maybe I'm looking at hundred of them and I see one proposal that I think fits perfectly that ow I'm willing to stake 1,000 gentle cans that this proposal is going to pass in that Dow in any in any of those dolls there is a second kind of signaling which is

the opposite so once there is a sufficient positive signal so once there is sufficient sufficient stake of predictors that is predicting that a certain proposal is going to pass then there is the opposite signal and maybe I think that this is overhyped or over-promoted and I'm willing to stake against that I'm going to stake 1,000 talk Jen tokens that this proposal will not pass only if there is sufficient positive signal for a proposal only then it can be boosted so in that sense the voting system outsources the navigation of the creative attention to an economically driven ridiculous network then by that it can be indefinitely scaled up there is a cost there is economic cost to

scaling up but given that economic cost you can show that you can indefinitely scale up well given the economic cost and given that you have a network effect of predictors you can show they can indefinitely scale up decision-making capacity in a resilient manner and note that decision are still made only by reputation holders so this is different from filter key in few turkey you have predictions and then you have also decisions but the predictions and decisions are made by the same network here you're separating the signal major predictors and a decision which maybe better reputation holders and I would argue that that's much more stable and resilient strategy so let's go a bit more details so that's how it looks like

in a diagram so you have a key of proposal over here this Q operates with an absolute majority so voters can vote with the reputation on each and every proposal and if a proposal reaches an absolute majority of yes is being executed now there is incoming flow proposal anyone can bring in a new proposal maybe there's some proposal fee that needs to be paid and then predictors there's a second action predicters can stake gentle cans about the outcome of those proposals and there is an order class so firstly there's the fishermen that they are predicting that a certain puzzle outfit a notion of puzzle is interesting and will pass and after there is a sufficient positive

signal only then there is the the challengers who then willing to stake tokens that those positive signal are actually fake and then only if there is a sufficient positive signal and I will get back to it in a minute only then the proposal can be boosted which means that it goes to another queue and this is the boosting queue in that queue proposals are being open to a finite time shorter time maybe a week maybe three days depending on the dowel in that queue propose are being executed or decided about with relative majority so if it's majority of voters who vote in that tankers say yes it's being

executed if they say no it's being rejected and finally there is also quite ending figure that if you are familiar with finite time attacks is critical and still I again I'm emphasizing again because it's a question that is key I keep getting that the decision is only made by the voters and not by the predictors finally every like two important things one is that there is a predictor bounty that I haven't told you about but basically there is an economic incentive put here by the Dow or by the fees from the incoming proposals either depending on the use case but that bounty is critical that bounty makes the installation for the positive predictors to find out

those interesting puzzles once there is a positive signal that the stake from the positive signal becomes the bounty for the negative the challengers okay but in order to start that process you have to have that positive bounty so that's and you can think of it as the cost of management because of scaling up your management or decision-making capacity the second important point is that you see I told you that there is a staring like boosting threshold a sufficient positive signal which basically means the sufficient stake is staked the dream snake is staked on predicting that a certain boss is going to pass but now I'm also adding that this threshold is actually dynamic and

it's exponential in the number of repose that are already open already boosted so the more proposal are already boosted the more and in fact nonlinearly more attention is already collective attention is already diluted and does you need to have a higher threshold for boosting yet another proposal and that's important okay almost almost towards the end let me try to argue quickly without getting into rigorous details let me try to argue that we achieved alignment of the local decisions with the global opinion so let's assume that the proposal is not in line with the Dow global opinion so now there are two

options either it was not boosted or it was boosted so if it was not boosted right then it's in the queue if it's in the queue it can only be approved by an absolute majority absolute majority by the finish and the global opinion right so by definition it's in line with the global appeal or not approved on the other hand if it has been boosted it means there is an economic incentive to put a stake if it has been boosted it's only because there was a sufficient stake put on the proposal predicting it's going to pass but that dense stake becomes the boundary for the Challenger so there there is there's an economic incentive to put a stake in a test that this browser is not in line with the global

opinion so if you think that it's it's wrong you have economic incentive to say that to express that and signal that on the network and more so once once you did that you also yours you have a skin in the game and you have the economic incentives but then to call the right voters you think are missing from that vote process that that will change the result and notice that the boosting threshold as I mentioned before and that's the economic incentive to stake against the booster proposal is exponential in the number of already boosted proposal which then amounts the delusion of collective attention or if you wish the extent that the global opinion is approximated why is it approximated you see here you have a final number of proposals so let's say

there are five proposals open for one month you would clearly agree that the entire collective attention is focused just about five proposals and thus there is a global opinion even even if people have not voted it's not because they lack attention right but then what's approximating that fact the size of that boost because if that boost is now twenty in a week if that boost is now hundred in a week is that still true that everyone has the capacity to vote on each and every decision the answer is no the answer is that is continuous and the higher the the larger that stack is the more and more approximated is the fact that we are operating under a global opinion but on the other hand as

shown here the more proposal we have in the stack the more the dilution of the collective attention and thus the more the extent the global opinions that push summated but in order to protect that there is exponential more incentive for predictors to signal for that situation so finally predictors have three roles the signal for interesting proposal out of a notion they warn from bad promoted proposals and finally once staked and have scheme the game they call for voters to reflect the global opinion and to summarize the talk there is a principled tension between scalability and resilience dowels

her graphical senses suggested solution that makes local or relative majority decisions to be guaranteed in line with the global opinion predictors are incentivized in order to achieve that predictors are incentivized to find good unnoticed proposals and to warn from bad overly promoted ones boosted or relative majority decisions are only possible with a sufficient prediction predictors take the positive stake and thus the warning incentive correlates with the extent that the collective attention is already diluted and that makes decided proposals are guaranteed to be in line with global opinion and the resilience scale tension is resolved or translated

into economic cost and the outcome is indefinitely scalable organization so let me end the talk with that thank you [Applause]