J3 PAIN 05 Matan Field DAOs, decentralized Governance

so my name is my tan I'll speak today about decentralized governance protocols at scale and this name or graphic consensus I'll try to also explain the origin of the name so let's start ok so again I'll make like a quick definition so I'll speak about dows and about governance and I want to make a quick definition of what a dao is to me because it's used very liberal in space so I get what I said yesterday my talk and a little bit in more details so the Dao is a new form of human association that on one hand aligns the interest of actors and manage joint resources around a shared mission or goal or task like companies do but which also becomes more effective as it grows and is also decentralized and open like free markets do that's my definition of Dao which is maybe capture summarized in those two words it's an organization that is both orientable and super scalable so orientable means that you can orient that organizations towards any directed towards any mission and super scalable means that it scales super linearly so when you grow the organization's become more effective rather less effective as all organizations are and that's to compare with the film which is orientable but sub scalable it becomes less and less effective as it grows and the markets which are super scalable but non-orientable so as we said that thou is orientable and super scalable and to afford a super scalable organization you need to have a super scalable governance system and that's that's the theme of the talk so how do you form a super scalable govern system it's very counterintuitive you have more we have more and more agents and thus more and more decisions to make and you are becoming more and more effective in making those decisions very counterintuitive so let's see what we're really really like let's paint or draw super scalability in some more visuals so for example let's say we have a really translations so everything I'm saying here today is about large organizations so we have a very large organization here and let's just assume we are just doubling it so now we have twice the size of the organization right so like scalability means that now we can perform all of the say all the stations of the first and all of the station of the second which now means that it's just just like two organizations roughly just doubling the organization now in reality if we just had a capacity to perform decisions this large organization would also have interactions between those two imaginary organizations and thus you need to process even more effective decisions in the larger organization so that's what we mean by super sky bility being able to grow and then also saturate the entire space of decisions that comes about from growing now that's very counter to another way to put it is that when the number of agent grows the number of decision can grow super super linearly now why is that so counter intuitive because we are used to this picture we are used to that when we are growing organization anyhow all decision needs to go through some Center right so the limit of the processing power is in the center and the result of that is sub scalability which means when you grow the number of agents in the system you have become less and less effective so you grow the number of decision you can you can process sub linearly okay so that's a definition what we want to achieve today which is super scalable governance system and that's let's try now to see how we can afford that so firstly what is the problem again I'm repeating from yesterday the problem is a core tension between resilience and scalability right so if you if you want to have more and more collective attention attending each and every decision then clearly you are less scalable if you want to have little amount of collective attention put an each and every decision then potentially you're not resilient right so if small group of people can take decisions on behalf of the larger Dow clearly that's potentially sensitive to faulty behavior and when I say faulty behavior I mean both attacks right so a small small group can literally attack the larger fund for example well but also also just decoherence so you can just take decisions that are not in line with the greater majority so that's another maybe less risky form of faulty behavior so by definition by everything we just defined by definition the the desired outcome should be something that we call holographic consensus which means that we should allow for small group of people to take decisions on behalf of the larger thou but in a way that ensures the coherence between those decisions the local decisions and the global alignment what I call the Dow's truth so of course the dolls don't have objective truth right now just like any creature they have their mind their hive mind and the truth of the DAO is basically the outcome of decisions if all agents had a capacity to consider those decisions and you know spell output whatever whatever the output is maybe most of them will not even a you know express any opinion but at least they have the capacity to look at it and see that it makes sense or it is they don't have the you know the the mental capacity to express anything so this is the facial mograph consensus and this this will be the desired outcome of a scalable or super scalable governance system and the way that we are the the way the way it works is that we are putting a prediction game so our prediction game engine tokens the tokens of doubt stock a project that ensures the alignment between those decision by basically tripped economically incentivizing finding glitches and protecting them and we'll see how it works so before diving into the protocol let me give let me just go through a few insights from option computation so I'll just go very briefly through I think was the first option completion protocol in 2015 it's the option competition marketplace so let's say that I need to compute one thousand dollars worth of computation on the internal blockchain and it's just for now assume there is no glass limit so then anyone can runs it locally it's pretty cheap and I test the results so I can write it on my machine I can see the result of that computation is five maybe that population is part of a larger program and then I I attest to the blockchain to the contract that the result is five and I'm willing to stake two thousand dollars on that at the station so I put that on the contract so if after some time there is no complaints against it we'll just use that at the station as the right result and we'll print it into the larger computation however anyone can verify locally on their machine the computation it's very cheap and perhaps within the confirmation period someone says no the result is four and I'm willing to stake two thousand dollars for that and then of course once we have conflict then we do the computation on chain the wrong a tester lose their stake cover for the cost of the computation the remaining goes to the right tester right so this is the very simple protocol and that's also the insight for the centrist governance at scale so the bottom line rather than having so we always thought that consensus means that everyone needs to looks at every edge at everything so rather than that we can have everyone can look at everything everyone can look at everything plus reward a right at the stations put high penalty for wrong at the stations and then high gain for signaling wrong at stations okay so with that let's let's let's try to build up the protocols in in steps so firstly the the immediate suggestion would be to start with a relative majority rather than absolute majority so absolute majority means that each decisions needs to be considered by roughly all voters and then you need to have majority of reputation holders of all reputation holders to support a decision in order to execute it in a relative majority scenario you open a decision for a certain period of time let's say a week then whoever votes on that decision within that time votes and then whatever those people who voted decides that's what we execute if majority of those who voted not of all reputation holders but just majority of those who voters say yes we're executing the proposal if they say no we reject the proposals this is relative majority now the problem with relative majority is that it's resilient to attack right I can open now 1 million proposals clearly most of them will be unnoticed I'm completely diluting the collective attention and then one of them or maybe in some of them I will have a deadly transaction maybe I'll just transfer all of the funds to my wallet so then the the the the first or the early trials with that to protect this problem were coming from putting a quorum so you need a certain level of voting power on answer on a proposal to be considered a valid but the problem with quorum and that's roughly the problem with any static parameters is that if quorum is said too high then you're just going back to the same problem of absolute majority if quorums is set too low you're getting the same problem of resilience of relative majority and then likely in many situation you actually have both resilience and scale problem at the same time so chrome doesn't solve the problem it definitely doesn't lead to super scalable organizations so one simple solution let's make that Billiton step so one simple solution is to funnel the collective attention to certain number of proposals and we'll call that boosting proposal or boosted proposals so how it works so you have a queue anyone can open a proposal into the queue and by default a proposal in the queue can only be executed with the approval of absolute majority now queues get some rank and we'll discuss what is the rank method later and under some conditions proposals can transfer can be transferred from Q to the boosted list the boosted list has right now n proposals and the boosted period the proposals are open for finite time at the end of that time they are being decided about by relative majority so that method method basically are funneling on the entire the collective attention is guaranteed to be looking at very you know very small and protected number of certain proposals and by that increasing the effectivity and the effectiveness of the collective attention so this is the funneling method so this is nice I'll let me let me show you a little bit how it works specifically and that's the demo - showed yesterday so these are mock-ups from this demo so anyone can this is it this is the alchemy that's the application for the centrist budgeting of open source projects so anyone can open any proposals and and everything that almost everything that I'll show here is completely configurable so anyone it's also configurable like you can put conditions on who can open proposal right now the default is at any one so we don't even need to be a member of the DAO to make a proposal but you can also configure that you need to hold the certain threshold of reputation to open a proposal and then you can open any proposal and that's also configurable whether you can transact funds or whether you can change the rule of the protocol as well and then you can also add a possible mini fee so that you need to stake some tokens in order to proposals and close to put possible lifetime configurable so for example even absolute majority proposals can have a certain lifetime say two months after which there are just just rejected because they're not relevant anymore and then voting proposal is weighted by reputation which is just an attribute number for each address and I'll say just a couple of words about reputation flow later in the end of the talk and then regular proposals again are approved by absolute majority threshold and that's also configurable to be 50 percent or 60 percent or whatever chosen then you can boost it and this is a condition if the rank is higher than some threshold or are not and then in the genesis protocol that we are launching with which shows are not to be exponential in the number of already boosted proposals so it's a very protective scheme so you can have it less than a exponential you can have it polynomial for example but exponential is highly protective as you'll see where n is the number of booster proposal already in the boosted list and what happens with the booster proposals they are open for certain boost period which is also configurable and then at the end of that period that are decided by relative majority and one last point is that there is a quiet ending period which is also configurable so let's say the last day no change of decision should be made in last day and if there was a change of decision in the last day you just increase the boosting period with one another day until you have one last quiet day that's that's critical for boosting ending attacks against boosting and expanding attacks so now we're adding the prediction game anyone can predict the Dow approve of a certain proposal by simply staking Jen on the proposal so I can come and say look III look at this proposal maybe I also have an interest to promote that proposal maybe that proposal is to fund my own project and then I'm willing to stake two hundred dollars or maybe two thousand dollars because I believe that this this Dow will approve that proposal so it's it's pared down it's not an objective truth I'm not saying something objective about this repose is good or bad I'm just saying and I'm predicting this that will approve the proposal now anyone can also stake against that prediction of course wrong predictors lose it all and the right predictors gained some more Jen these are the rules of the game a little bit in more details I can ask you it very fast but wrong predictors lose all of their Jen tokens then also some of the all of the stakes of both sides goes to incentivize the voters the pre boost voters or pre post voters share B fraction which is also configure configurable of all stakes and we call it predict fee prediction fee and that is split between all pre boost voters in proportions of the reputation then there is one one last piece which is the Dao bounty which is basically added to the stake that is then gained by the successful predictor it's critical to incentivize a positive predictors in case that no one predicts against them so you can think of it some is sort of like the management cost of the Dow and then all successful predictors gained in proportion there's take the loss take of the losers plus del bounty - the prediction fee knows that regular proposals in the queue that are hitting lifetime are considered rejected and accordingly positively positive stakes them are lost now how do we fix the rank so generally you can have many different formula and it can depend on for example the amount of vote that votes that voted yes and the amount of all that voted no the amounts of stakes that put yes and the amount of stakes that predicted no and which is a very very simple formula just s plasmus n - s - which is the sum of Jen predicted that the proposal will pass - the sum of Jen's predicted that the proposal will not pass now why don't we make it also dependent on the on the voting because that's already implicitly taking into account with the with the staking right like if if I see that Vitalik just approve the proposal in theorem I will run to put a stake the tool it will pass because I know that he's a very influencer so what is the outcome of that what is the outcome of that that is protecting the collective attention so you see that when you have more boosted proposals number n it means that there is potentially higher boosting threshold because ours are not goes like exponential of N and that's exponentially more supporting stake needed for yet another proposal to be boosted but once that amount of stake was put in favor of that proposal there is potentially more interest of those who stated to make sure that the right votes shows up like if I if I vote if I stake for proposal I put five thousand dollars and then eventually I see that when it goes into the boosting period the proposal seemed not to not to be passing they can then I know that something is wrong or I think at least I think that something is wrong then I called I can call the voters that I think there are influencers and should be in favor of that proposals maybe they didn't show up yet to vote it so basically I'm acting right now as a single signalling agent basically we are outsourcing signaling or navigation of collective attention to a greater audience because the audience of stickers that's or the predictors and that's important to Express is not within the DAO so it's a much much greater audience that can look at each and every proposal so the bottom line of that is that if there is exist a gap between a proposal status and its belief majority outcome then this gap turns into a potential reward which then generates an economic incentive to fish those glitches second point is that predictors like any investors and traders in the market in fact like any money market economic market go on their economic evolution and improve over time so if I'm a good predictor I make nice profits which means that they have more money to put even more larger stakes and make even more money and if I'm bad predictor I'm losing my money and I start making predictions so you can see the token holders can boost or filter proposals but they cannot make decisions so you can you cannot buy a decision so it's protected but you can bite into consideration but of course if you are buying into consideration a bad proposal you're going to lose your money also note the logic similarity to often competition that I've just mentioned beginning but also to fishermen protocol of parity or polkadot if you look at it so almost at the end of the talk just a quick comparison with future key bits because last question that comes about a often so future key is also a prediction market game for governance now the difference is that and that's also my critics for future key is that future is purely economic so the prediction market not only signals but also makes decision and the outcome is twofold or the problem is twofold so firstly this kind of system cannot account for wide variety of value systems so it's pure economic and buy that is very narrowed in the value system it can reflect and the second that is highly manoeuvrable just as markets are highly manoeuvrable as we see every day and that's why regulation basically came about so the second part of that comment is just to say that we notice that in governance there are really two different functions so one function is a decision and another function is the is the signaling so if you want the choice making and sense making are and decisions are well made by by governance bodies but then signaling is well made by markets so we just disentangle those two functionalities and let the market do the signaling the sense making and decision body to do the decision the choice making I won't say much about reputation flow what I mean by repetition flow is that how okay how reputation changes in the system beyond the simple way which is just manual reputation flow so we can make a proposal in the system to add or subtract reputation to an agent but there is also a great big rotation flow so for example you can have algorithms where those who vote in accordance with a majority early on gain reputation and those who vote not different from the majority early on a lose reputation you can make algorithms that those who make successful proposals that are eventually passes again reputation zones of force you have you can have a great reputation flow which is yet another topic and my last slide is just mentioning that these protocols or graphic consensus and I'm also covering here funneling collective attention and in some sense also monetization of tension is just a part of a greater toolkit for scaleable governance system in particular it should go together with delegation the ability to they get your votes and compositionality so the ability to have structures so like a company of sub companies of sub sub companies and so on so forth so all these together I believe by now we have a full solution for a fully super scalable governing system thank you hi thanks so much for your talk is so good to hear fresh ideas on governance I have two questions number one is how what are the different incentives or if you Cindy Chen for like posting proposals rather than requiring someone who wants to propose to stake some tokens in the native token of the Tao and if the proposal doesn't pass like chip or endorse tokens and number two is like can you repeat the process rather than using the other token for like posting proposals just in order to create a proposal in Tao you need to stake a certain amount of the town native topo you asking to white gem talk is not that wise like different incentives and the second one is like what are the ways to like prevent the prediction market being completely gamed by the people that are going to make prediction and kind of like betting on whatever they're gonna do and then because that could that could lead to market manipulation that way okay so the first question let me just repeated so the free quest just to make sure that I hear correctly so the first question is why using Jen tokens in all of those inside down and not the Dao tokens for energy tower tokens and just stake the token yeah in order to create a proposal right so the thing is that by using so firstly you could have asked the same thing on bio on ne DUP and specifically theorem right you can you could have allowed for a gas to be paid in Bitcoin right that's true for any DUP now there are two you really need to ask two different questions so firstly you're asking whether a utility model for token is defensible so that's the first question and then the second question is why using that utility model universally right so the first question whether it's defensible so my answer yes this is defensible because if you change the token from Jen tokens to a smaller and store it with different tokens sort of a local Dao talking or whatever you want basically what you what will happen is you will be used you will lose all of the greater network that it's because it's all depends on distribution right when you distribute your app you define the contract when someone else distribute your app someone else defines the contract so you control your distribution but you can't don't control application of other distributors right so but the thing is that there is a Nash equilibrium here that most people will not change the con tract and those who will change the contract basic will lose the greater equilibrium the greater network that will simply not see the predictors of the greater Network will not see anymore those proposals as being staked than will not react to them so it's a Nash equilibrium now once you understand it is defensible model they need to ask why use it and the reason why use is just to you know to actually build the value in to those agents are I mean gives back the value to those agents that are building that value and that close a psyche and that's that's the same for a theorem or any other it's the central application in the second question why it's not gain above so well I'm actually really inviting everyone to to post that question and actually offer a vector attack vectors so I will argue that this is not game herbal and I'm yeah I mean I cannot I mean you can try to formally prove that it's not gained about but or rather you can just offer attack vectors and then try to answer them so I I haven't found any any significant talk vectors that model I also spoke with a lot of people that it couldn't have fun but I'm actually inviting it I will think that's like the best outcome of that talk is if you know some attack vectors are being very naive and that maybe doesn't work at all so I basically swordy the proposal is going to pass tokens are liked by a lot of no shares I am a big token holder in the tower of which though of at our local though yeah yeah so then I have a lot of tokens and I buy a lot of shares cheap shares of like just tokens and then I bought with all my stake yes no but the voting that's this is exactly the reason here that voting so reputation is separate from tokens so we can buy tokens that's exactly my claim so if any economic governance system is highly when you put but I agree with you completely and that's exactly why government system should not be economic but then the problem with non economic government system is that are highly non scalable and then you need the market economic market to actually make them scale but again that's exactly the outcome of free stock thank you Oh question here hi thank you can you talk a bit more about reputation and how you mentioned reputation is used to weight the vote of people how do you establish reputation right so that's what like that's my last comment so basically reputation is simply a name for voting power or influence so if you if I have twice reputation my vote counts twice or my my attestation or my not sorry man my vote or my opinion now how reputation changes an established in particular is is what I call reputation flow so how reputation flow in the Dow between different agents so basically as I said there is the manual way to the repletion flow which means that you can make proposal to add reputation to an agent that's very slow way right and then there is our greet make which I don't cover in this talk there are charisma quai to flow reputation for example you can make it such that pre boosting voters that voted retrospectively correctly with the majority they gain reputation and pre boosted voters that voted retrospectively against the majority lose reputation you can make this a successful proposal proposals gain reputation and rejected proposals of rejected proposal loser operation and so on so forth so you can make like a great Inc reputation flow you can also have air drop you can have by the way we are launching the first dau next week and here in the desk you get a coupons with which you can actually get tokens in that Dow but also reputation so this is the way we we know that this is community is a good starting points good seed of reputations we are air-dropping not for the entirety reham ecosystem but we're a dropping for this conference no more question okay yeah just to come back on the algorithmic one that you mentioned so if I understand correctly you're gonna give more repetition to people who previously voted with the majority if you establish that basically you instead of incentivize people to always look the same and you exclude people who don't agree with the dowel so it's easier to make decision because you encourage people to always vote the same thing but doesn't that contradict the fact that you want to have a huge doubt because if you exclude people who don't agree with the majority isn't isn't there a paradox here right so as big as possible but then you exclude all the people who don't agree so you reduce the number of actors yeah yeah so this is a good question which which comes comes often so that's also one of the reason I didn't want to go to it repetition flow but basically so firstly let me just say again it's configurable so different ways will give you different protocols which will give you different philosophies if you want and I want to put the family the baseline for for scalable and resilient protocols now to this question specifically my first answer would be that if you are Miss aligned with the Dow that's fine because you're really misrepresenting that you know that the alignment and it just mean that you need to fork into a new value system whereby you are the majority of reputation then if more and more people are lying with you they will just you know go to work in that Dow and with that you'll have like a wide you know array of vectors of alignment for different value systems which is totally right so I mean the purpose the purpose of scaling up organization is not just to be coherent organizations actually we had the most the largest possible but cohere organization with coherence so that's that we like the simple answer but this is basic triggers again the triggers more philosophic question you can have different protocols that capture different behaviors collective behaviors and this is one option you can have also different updation flows okay thank you [Applause]