Parameterization of crypto-economic systems

Transcript

are you ready great so thanks for coming everybody uh my name is William George and I'm a researcher for the caleros Cooperative uh so today I'm going to be talking about parametrization of crypto economic systems uh and a lot of what I'm going to say is informed by the research I've done for the parameterizing the parameters of of Claris so I'll briefly introduce that and use it as a sort of running example uh but uh like I hope to like pull in some other examples and draw some general themes uh so claros if you're not familiar is a blockchain-based dispute resolution platform and the idea is it imagine you have two parties that have some business relationship Alice and Bob Alice is like of small business owner or Bob is a

freelancer uh and instead of going to like signing a traditional contract and going to a traditional court or traditional arbitration if they have a dispute instead what they do uh is when they start their sort of work together Alice puts the payment that would be paid to Bob into an escrow on a smart contract and if ever there is huge if there's no dispute analysis clicks a button saying release payment to Bob but if there is a dispute uh then you use crowdsourcing uh with the Claro's platform where a certain number of users of claros are randomly drawn and they will judge your dispute uh and they do so in a way that's sort of incentivized and organized using a game theory principle that's called a shelling point or focal point uh this is an idea that

was introduced by Thomas shelling he went on to win the Nobel prize in economics and the idea is that if people are trying to coordinate in the absence of communication they tend to coordinate around solutions that seem sort of special or natural somehow so a famous example that shelling had was that if you have two people that are trying to meet in a given City at a given time and they know like or a given City on a given day but they don't know exactly where they don't know exactly when uh where will they go uh well they tend to meet like a famous landmark so in Paris if I had two people that didn't you know were trying to meet somewhere but they didn't know exactly where they might go both stand under the Eiffel Tower uh and at what time well maybe like noon because noon noon is a

distinguished time the Eiffel Tower is a distinguished place anywhere else would have worked uh but people sort of gravitate toward these like natural special Solutions uh and then in cleros the idea is that sort of honest rulings of disputes should be focal points should be shelling points that people will gravitate around so the different jurors are incentivized to sort of agree with the consensus of other jurors and then if you're a given juror you think okay like I'm reading the case I'm looking at the evidence I think Alice has a better argument I think the other jurors are also going to think that Alex has a better argument so I don't know after Alice expecting everybody else okay so with that sort of introduction out of the way uh like what parameters would one choose in such a system uh and like how like what would go into your

choice and parameters uh so there's some really some some you know obvious choices uh like the there's a per juror fee like an amount of compensation the jurors need to be given to compensate them for their their dispute resolution work uh that could like vary higher or lower uh jurors can like have to put a stake that can be lost uh if they're ruled to be incoherent with sort of the community consensus on a Case uh there's a notion of appeals in claro so there are a number of parameters around that uh particularly there's a parameter that determines how many times you can appeal before the system sort of stops uh like there are lengths of periods how long one has submit evidence how long you have to vote Etc uh and what does this sort of parameterization process look like now

uh well it looks like a parameterization a lot of other doubts and there's a forum and people post things in a forum uh and then maybe people vote on snapshots I post a lot of parameters in the Forum that other people do sometimes too uh and posts that kind of look like this um and I'm going to talk about like over the course of the talk how one chooses parameters how one could choose parameters and sort of what things to keep in mind and a running theme is going to be around attack resistance to your parameters being manipulated uh so for this like governance model which you know Claris uses and many dowels use similar things for their parameter choices uh this sort of security model that keep in mind compared to the other potential models uh is that no participant should control more than 50 of the tokens that are

voting on a given for that parameter update because if you could if like one like one party could control the vote you could imagine them putting malicious parameters into place uh they like set all the periods to zero or something so that like you know when a case comes up no one has time to vote or something like that uh and it's interesting to note that in Claro's uh that security model is also sort of the security model of the system itself the dispute resolution game uh and that the way in Claro's the the sort of people are choosing to be active dispute resolvers uh is that they're uh they stick a token uh and then a random token is selected and if you have that token you can vote on a dispute uh and this is Dylan so the system is civil resistant if I had

someone that would try to create like a lot of fake identities to get disproportionate voting weights uh they can split their tokens over as many dresses as they want they're still going to have the same voting weight uh so by having this kind of token weighted voting um you you can't you don't allow for someone to get disproportionate weight by having like fake fake addresses uh and note that because there's this notion of appeal that you control larger and larger panels of jurors uh but if you get like a really large panel of jurors it's very sort of statistically representative of the community uh your odds of getting more than half of those on that panel is um are are basically like you will get more than half of the votes if and only have if you have more than 50 of the stake sort of by the law

of large numbers okay so similarly the the security model the dispute resolution game is no participant should control more than 50 of the stake so it's convenient that like this sort of governance in the game itself have the same security model because you've kind of reduced your attack like surface you only have like one thing you need to assume okay but like now zooming back out and being a little more philosophical and comparing with his mother systems you might ask okay like it's great that the governance attacks sort of security levels the same as the system but why have parameters at all like if I could just like avoid having parameters I wouldn't have to worry about attacking people Melissa to change the parameters uh and uh so this sort of gets to like a principle of governance minimization uh that could be approached in several

different ways uh and so like a first idea would be to say okay well maybe I'll just redeploy the contract with each set of new parameters that users would want to use if somebody thinks that the fee for a task should be X they deploy a version of the contract somebody else thinks a fee should be why they just deploy a different version of the contract and then parties can decide which ones they want to use yours decide which contracts they want to work on which have the most appealing fees and the problem with this uh is that your each Court each contract wouldn't be very resistant 51 attacks uh someone could come along and you know like on a contract that only has like a few people stacked on it it doesn't take very much to take to like stake more than the people who would be there and take

control of that contract so your system is kind of fragmented and the weakest link isn't very strong uh and like in a similar kind of you know analogous situation you wouldn't want a lot of copies of a D5 protocol because you would split up liquidity uh so in both cases you have kind of network effects uh in the sort of Claris example you have this network effect that if everyone is staked in the same place uh they're each that that place is stronger than lots of different you know separate courts would be uh because in order to 51 attack like a single chord you'd need more resources and a lot of fragmented chords okay so that idea is not very useful you know it's like a very powerful and like ideal situation if you have a crypto academic system that has like no network effects but it's not very useful for

most of the systems we actually have so like idea two uh maybe I'll just have parameters that auto adjust to the market uh my like fees the drugs receive will just like auto adjust uh maybe for example each user can set a price at which they're willing to be a juror for a task and then I'll just take the cheapest juror for each task it's a very free market solution we believe in free markets uh so like you know you just sort of like economics determine what the values in your system should be uh and a problem with this at least you know for this particular approach in the context of Clarence would be that you can imagine that like if I have an outsider like someone who has a vested interest in one of the cases uh that you know maybe they're one of the parties to

dispute uh if they see their cases going to come up soon maybe they just set their like threshold for what they're willing to work really really low so that they will be drawn and other people won't be drawn uh and like this you wind up drawing disproportionately people who are hostile uh that we're willing to set sort of artificially low prices uh and they can take over the course on a given task okay so that's not great so maybe like a less naive approach would be that each user has a price where they're willing to work but then each the parties to the dispute set a threshold price that they're willing to pay uh and then you randomly draw from among the users uh whose price is below the threshold uh so this works better and it works pretty well if the parties set good thresholds

if the threshold set by the parties is high enough that you draw at a wide pool of jurors uh then you know like you'll get you know maybe there'll be a few honest jurors to a set on realistically low prices uh but you know if your pool of people has a good percentage of monitors in it you're more likely to get them anyway uh when it's a problem is maybe the parties think okay I'm gonna set like a low price because I'm going to try to like drive a hard bargain with Clarence I'm trying to get like affordable super resolution by sitting like a very you know aggressive price and by setting such an aggressive price they could price out all of the honest jurors so that they're only left with the dishonest people who are trying to like maliciously attack the system uh so

that's not amazing particularly because you would hope like in designing like doing research on clearos we want clarus to sort of act like a black box uh where people with disputes can just sort of plug this into their system without thinking too much about how it works you don't want them to have to like worry that like about what threshold they're going to set and moreover you can imagine having a situation where like the two parties you would dispute have sort of asymmetrical knowledge about Colorado's works and maybe one of them sets like like proposes setting like a low threshold knowing that it will drive out most honest honest food resolvers that want to put this on as people who could favor them and then the other party could be taken advantage of so that's that's not very good either uh and then ultimately what's sort of

like I think the philosophical point you take away from this is you don't only have these kind of free market effects you know there's like econ 101 diagram where I have like a supply curve a speed resolution and a demand curve for dispute resolution um and I I want this sort of like ideal equilibrium where the curve is intersect uh that's going to like maximize sort of global utility uh you also have this like supply of dishonest dispute resolution uh these people that are like willing to attack the system uh and you can imagine that like if you just naively Implement your your sort of free market system you'll wind up at the lower right equilibrium uh where most of your resolvers are attackers uh so like we've added constraints to the the sort

of world of free market you know kind of solutions where it's not just a matter of like I don't think the markets move freely you also have to worry about being attack resistant uh that's something you need to like you know build in uh so that you have good outcomes uh so what are some tools we have to parameterize crypto economic systems claros and others how to make sure that we're at the like upper you know like intersection not one of the ones intersections in the lower right uh so a tool that we've used that uh you know for Claris parameters uh is it like we built a calculator that I sort of wrote in this Google collab sheet uh it takes in a few inputs uh and outputs parameters that satisfy a number of constraints uh so what kinds of constraints so just like an example uh

you want jurors who vote randomly to lose money on average that that's one of the constraints in the calculator uh so jurors you know like they could vote randomly maybe they noticed that the Freelancers were winning 80 of the freelancer small businesses disputes uh so they just always vote for the freelancer because that wins more often than not uh you know like just strategies for for the jurors that don't take into account like the individual facts of an individual case we consider lazy strategies and they should all lose money uh and our sort of main tools to disincentivize Legacy strategies are our parameters uh the arbitration fees that are paid to the jurors uh and these deposits that they can lose if they they evoke poorly uh and then you want to tune them somehow in such a way so that like

honest participants uh gain money on average if I'm like looking at the case and really making an effort to review the evidence I should gain money on average but like a lazy participant who's voting randomly or something should lose money on average uh and how does the calculator do that well it has some inputs uh that uh have to for example like the rate of success of lazy voting uh so like how might one determine what that is well like it's not you know it seems like relatively straightforward to pick out if I just picked a random option well then my odd success would be one over the number of options this is the like more sophisticated lazy strategy where I always hope for the most common winning option uh well that would give me the rate of success of that option so if

like the freelancer always wins with 80 of the time you know then my rate of success is always winning professors 80. that's pretty objective pretty straightforward I can sort of like read that off the chain if I can't just reason it out theoretically uh like similarly you want the rate of success if you vote honestly uh so you can argue that should be greater than the overall rate of agreement of like users with the sort of with the collective answers uh the right at which people win like the dispute resolution game well why well not everyone is necessarily participating honestly but you've you've heuristically at least that the people who are participating dishonestly are going to have a lower rate of agreeing with the community than the people who are participating honestly so if you like look at sort of like the statistics

we have on overall rate of agreement with the community and you take out the dishonest people like the people who are actually participating honestly should be even higher than the overall rate of agreement with the community uh and that's something that you can like really straightforwardly read off of like a blockchain Analytics tool uh and then you're like okay uh what other influence do I need well I need some notion of how hard the task is uh if my test takes hours of effort by some specialized person that should pay differently than if it's something that someone could do in five minutes while riding a bus uh and so how can we get these estimates of effort well unlike the sort of previous two slides where you can really readily read off information on Jane about like what your input to the calculator should be here

this is something that's really subjective like you really need input from the jurors themselves uh how hard is this task uh and in practice in Claro's what we wind up doing is you know like when parameters are put to The Forum you know people debate whether there are the estimates of reasonable uh you know and get feedback on telegram sometimes I propose things or other people propose things in the Forum you know when I get people you know yelling on telegram that the you know the estimates are unreasonable uh and even note that this like diagram here which is implicitly being used by the calculator and its sort of constraints uh where people like make a certain effort and then they have a probability P of being successful that corresponds to like an honest strategy uh and they may if they make less of that effort they have a probability of T

which is sort of like if they voted randomly is already like a really real simplification of how one model's user's effort and like in reality they have this like Smooth curve where they like there's not like a fixed cutoff for how hard something is they make a little bit of effort their odds improve they make a little more effort their odds improve until eventually they spend hours pouring over a test they're sort of odds of getting the white panther to kind of level off uh and this you know in reality the sort of information you'd want about the like efforts that your users are making uh is a curve like this for every single juror in the system uh so that's like a lot of really nuanced information uh that is is hard to pick out even from you know people actively engaging in governance proposals and

like on Telegram and it sort of like highlights what what loses from being able to go buy a pure free market approach because like in like the sort of like the magic of free markets is that people were submitting the prices they're willing to dispute resolution work at like all of this information about the curves of different users is captured by that uh in a way that like this sort of collective voting on I on a on a given price uh like captures like in a less finely grained fashion okay so just like some other tools people use to like propose crypto economic parameters uh a lot of people a lot of other platforms use something called agent-based modeling uh so particularly in D5 protocols Gauntlet is a group that proposes parameters for verticals such as Ave and compound and

then the idea here uh is that you have like a bunch of personas uh that have some like coded way they act so like maybe my like Risk averse borrower uh so you know this isn't a computer simulation uh and like that the user in that simulation is code in such a way that if their risk ever goes above some threshold maybe they just closes their position if it's like some moderate level of risk and they add collateral you know rules for like how they interact with the world around them the simulated world around them uh and then when you're running simulations you pick some values of sort of global global values ether prices gas costs maybe those like like evolved by some kind of brownie emotion maybe there's some Trend or tendency uh but they can also be affected by what

simulators actors are doing uh so if my simulated actor is like all have their positions liquidated and sell a bunch of these maybe they'll like eat the price of my model plunges uh and like the like population of the simulated actors can evolve over time uh maybe all the risk-averse people leave because there's too much risk and then the like population of actors is all like the risk loving people uh and these tools are really useful for picking out like Global outcomes uh they can pick out feedback loops uh like maybe it's like okay if all of the like people are selling each because they're being liquidated you know because like the price of ease went down so people are being liquidated so like more than being sold so the price of piece goes down this is a feedback loop it's maybe not so hard to like reason out

theoretically uh but if you have like more complicated feedback loops you know might be able to pick them out from the model without having to like come up with a theological argument to be like oh there's this danger here okay so just to conclude uh like you want to have Prime when you want to parameterize your crypto economic systems the ideal is to not have parameters at all that would be the goal uh however that can be problematic uh in that if you have if you just want to redeploy a new contract every time you have some new choice of values in your system that doesn't work very well if you have Network effects and most platforms that we have have Network effects of some kind uh if you want to have things that automatically adjust to the free market you have to like worry

about constraints related to attack resistance that aren't present in a lot of sort of traditional systems where you just have some some like external actor who can punish people uh you know like the attackers doesn't have to be built into the system uh and then in um like you have to you know if you have a governance proposed puzzle system where people update the parameters based on some kind of voting like how does the model for a hijacking the governance compared to this is the model for attacking the system itself uh and then we have these different tools that we can use to try to come up with good parameters like that like pushes towards the attack resistant Elite equilibria uh so these sort of like calculators that I you know that we use with Claro's where you can program a bunch of rigid constraints uh are like

useful tools for you know members of the community that are trying to like input information about like you know the price that they're willing to work at maybe they don't understand like all the constraints in the parameter but they can still express some like price information uh by like putting in estimates uh into it into a calculator and outputting like a set of white uh attack resistant parameters that were compatible with their estimates so this gives you some way of people having like some being able to you know get accessible inputs into what parameters should be that would agree with what they think they're like level of work is or you know whatever other subjective estimates you have and then you also have you know like a agent-based modeling uh you can use both uh like approaches sort of simultaneously and

this is particularly important like useful whenever you have any kind of system that's like chaotic that interacts with itself in like really complicating chaotic ways uh the you know like in the chaos theory sense because you can pick out fat feedback loops in sort of Black Swan events okay so that's all uh I'll be happy to take any questions if there's time [Applause] great

or not hey thank you very interesting um is it possible that some voters are given more weight than others so like you could have um experts in the field who I could say 10 times more powerful than a regular voter in a governance proposal or in clarovs in cleros yeah so uh currently uh we don't wake voters like that however that's an active research that's a like a subject of research and that we're considering like expertise as indicated by sellbound

tokens so if you can somehow prove that the centralized system that you're an expert maybe you like submit yourself to a curated list maybe Claro's itself like judges whether you have the expertise you claim to have maybe you have a credential from some some like external system uh that could be used in such a way that you're like people are weighted differently uh and have like for example a higher chance of being drawn maybe you need this like expertise badge to be drawn in certain kinds of disputes so like approaches like that are becoming more possible as people sort of innovate on kind of like social layers decentralized Society kind of ideas okay yeah so thanks everybody [Music]

thank you