 So, hi everyone. So today we're gonna speak about arbitration of contract and decentralized quotes. So first, let's start with a simple example. You have Alice and Bob who met on the freelancing platform and they agreed that Bob will make a website for Alice against three users and Alice pays Bob. But Bob gave a shitty website and Alice is not happy and wants her money back. But since there is no charge back with crypto, she has no recourse. So to avoid that, she could put money in a smart contract if she gets a website, great, she sign and the contract pays Bob. But in case of a dispute, Bob say no, I made a website. If you want something better, you'll have to pay extra and both things are right. So they pay arbitration fees. The dispute goes to an arbitrator and the arbitrator can rule which will trigger the previous smart contract which will move the funds. The first catch is all could the arbitrator know what the smart contract was referring about? So to solve that, they first, when they create the contract, input the hash of the plain English contract. And this way, when there is a dispute, they can show to the arbitrator what was the contract agreed in plain English. And if there is no dispute, it's progressive preserving and no one has to ever know about what the contract was. Actually, we can even do better. Imagine we have a contract with four pages, pages one and two are on one topic, pages three and four on another one. And you don't want to reveal both of the topic of the contract if you only have one topic which is important for arbitration. So use a mercury and you just have to reveal the pages which are linked to what you are disputing. And the thing is that the arbitrator can be everything. It can be a centralized party, a multi-seg or even a DAO. And that's what CLEAROS is. I've given an example which is really simple, which is just transfer money. But in practice, we can have way more complex dispute and enforcement as long as it's enforceable by smart contract. So with money, with freelancing, e-commerce, crowdfunding, you will have, you have all those ICOs that raise money, but if the token buyer are not happy, they have no way to get the money back. So we will have milestone and when the milestone is reached, you unlock the money. And if the token order and the team don't agree, that's a dispute. Online gaming, cheating, social network, you submit content which is against the chart. You have a dispute, you remove the content and you penalize the party who are violating the rules. Intellectual property, we've seen a nice example in the DEF CON with ENS. You have someone registering your trademark. Well, you go to an ENS system to have a blacklist. You have a dispute. The dispute is, okay, this is my trademark. I don't want you to use this ENS and I want it to be blacklisted. And the result of the arbitrable contract is a blacklist of the domain if the one claiming the intellectual property is seen as correct by the arbitrator. Or insurance, this event occurred and does the insurer have to pay. So we can abstract that and have two kind of contracts, arbitrable smart contract which are responsible for enforcement of ruling and arbitrators which are responsible for giving the ruling. And the goal is to have every arbitrable smart contract arbitrable by every arbitrator contract. So this way if you're a contract developer you can easily switch from one arbitrator to another. So let's go back to Kleros. The idea of Kleros is to have everyone registered. So first we could say, okay, we let everyone register and everyone is a juror. Well, that would be quite impractical because you will either need to pay everyone arbitration fee or everyone will get peanuts. So in both ways we have an issue. So we can just draw some of them which we look at evidence and vote and get arbitration fee but not all. But there is still a problem. Since we don't have a way to be sure that someone had just one online on-chain identity, someone like this guy could register tons of time and be almost the only one drawn. To solve that, instead of drawing jurors, we draw tokens. So you activate your token for arbitration. We get a random number with one of the methods that I talked about yesterday. From this one we can get multiple ones. Everyone who has a token drawn is an arbitrator. So we add the incentive to arbitrate which is get the arbitration fee. But now we need an incentive to arbitrate correctly and honestly. But unfortunately we can't really measure honestly in an algorithmic manner. We could say, okay, it's another dispute. Did this arbitrator rule honestly or not? But that's some kind of chicken and egg problem. So the best thing we can have is coherence as a proxy of honesty. So in this example, Ignacio and Eurulio, they voted for Bob, while the majority voted for Alice. So those jurors, they will lose part of the token to the juror who voted coherently. That's inspired by game game theory and shading point. In this basic example, if you vote the same way as the other, you win something. Otherwise you lose something. And if we remove all the labels, we can see a matrix which is symmetric and from a mathematical starting point, there is no reason to vote for Bob or for Alice. Both of them could be seen as the same. But fortunately, we have labels. And in the absence of other factors, we have some solution which can emerge, which we call the focal point. And in this case, the focal point has to be honest here. So why would you vote honestly? Because you think that the other will vote honestly or perhaps because you think that the other thinks that the other thinks that they will vote honestly. In practice, it goes up to three and that's all we can get honest feelings. There are still some attacks. One could be, okay, I vote before everyone else. I vote Bob. Now, both of those parties, they know that if they vote Bob, they are coherent. So even if Bob is not the honest answer, they could vote Bob. So how to avoid that is to use commit and reveal. So instead of voting directly, you generate your number, a one-on-one locally, and you commit the hash of this number and your vote. And in the second phase, you will reverse the number, your vote, and on-chain we can verify that it was effectively corresponding to the hash you submitted. You could still say, okay, I'm going to reveal my number and my vote towards the party to influence them. So to avoid that, we have to penalize this behavior. So if someone reveals the number and the vote to another party, this party can show that on-chain, invalidate the vote, and steal part of the coins. So this way, you can't influence other trustlessly. You can either say, okay, I'm going to vote for Bob, but I don't prove it. I don't show you. I could be lying. Or you can say, okay, I'm going to vote for Bob. I show you the proof, but now you have the right to steal my coin and invalidate my vote. So it's not trustless one way or the other. Another attack problematic will be bribing. And that's particularly problematic with blockchain and smart contract because you can make bribing smart contract, which are trustless, which will automatically pay the bribe with a draw, vote, as a way you ask. So to avoid that, we need to have an appeal system. So if someone managed to bribe, draw at some level, the other party can appeal and get more and more draws. And if you are in the honest majority assumption, it will be highly more expensive for the dishonest party because most of the round he will lose and even in the round he will win. He will also have to pay bribe while the defender just have to pay appeal fees. So that gives a huge advantage to the defender. Another problem which can arise is, okay, perhaps I don't like freelancing dispute. I don't want to arbitrate those. I want to arbitrate insurance. So for that, we need an arborescence of course, which are increasingly and increasingly specialized. So when you register to be a juror, you don't just choose a court, you choose a pass to one court. So for chief, it's a general e-commerce and freelancing. So it can be drawn in all of these. But we can ask, okay, how do we make all the court? If you are aware to let everyone make subcourt, that will rapidly end in a mess of subcourt with no clearance. So that's why we need governance. And we've seen in previous governance system, like the DAO that we have a huge problem which is most of the token holder, they just don't vote. So how can we count the vote of people who don't want to vote? That's sort of through liquid voting. So in liquid voting, either you vote and everything works as if it was in direct voting, or if you don't and you've designed before and you show them before an advisor, you vote the same way as your advisor. And if this advisor doesn't vote, but also as an advisor, you vote as the way the advisor votes and it can go to the point that you have one party who is effectively voting. And now we've seen the website dispute and we see we have tons of website disputes. So why not make a court specifically for those? So we can also have some other kind of problem which is it's habituation, it's not just right or wrong. That could be the case for Oracle. If it's something with the president, yeah, you as a light or you tell the truth. That's easy. But for disputes, it can be more shady. So where do we draw the line? If you just let Jorah vote like that, some Jorah would vote honestly but be still considered as incoherent. They will lose token. They will be frustrated. So to reduce this problem, we'll have policies. A policy is something which looks like a law. It's basically a guideline of how to arbitrate. One example of policy will be if you sell some good and you sell it at distance and it's more than one user and you don't use track shipment and the buyer claims he has not received it. We assume that the buyer has not received it. So the one user, some party will say no, we could do that for 0.1, 1.410 and it may not be consensual between the Jorah. So for them to know all they should rule, they should have a guideline which can be voted through governance. Another problem is the 51% attack. What if someone buys more than half of the coin and keep giving the next rolling, accepting bribe that will pose a huge problem? That's also why we need some specific coin because buying half of the coin of Kleros will be way more expensive as buying half of the user which will be using Kleros because of market depth. If you want to buy half of the coin of a wine coin type, you don't just have to pay for half of the market cap because okay, the first seller, you pay them market price but you rapidly run out of seller at market price as the price increases. So perhaps even with a value which is higher than the market cap, you may not even be able to buy half of those. Moreover, if you buy those coins, you do some attack and you accept bribe, you are dishonest, well, you end up with a dishonest court and who won't is dispute to be arbitrated by your dishonest court? No one. So your token will rapidly become worthless. You can fork it even and make another court with only the honest party but even in that there is still a catch which is a contract which were created before they don't know the court become dishonest, they don't know that they have to update to a new court and a new token and they will still be under the arbitration of the dishonest court. But one could argue that it's the same with ethereum and connected object, blockchain aware. If someone buy half of the ethereum and features an attack, all the objects relying on blockchain they don't know there is an attack and they will still behave the way the attack chain is behaving. And in the first step I've shown you some really simplified version of what the incentive could look like but in practice it's way more complicated. So you have some part which is the token redistribution voting the same way as the other. You also have the honesty of the decision because if the court give honest decision your tokens are token in honest court and honest court is valuable so the value of your token increase a bit and if the court give dishonest ruling your value decrease a bit and since the token are logged during the cases you have no way to sell them before. But we also catch in the other sense which is you're not a computer you can't just like get evidence process them in one microsecond and give a ruling. It actually takes time you have to look at paperwork, at images, at argument of parties and yeah time is money so it has a cost for you. So you could just say okay now that may not be worth it I will just vote randomly to get the fees. So depending to the value of all these variables in the incentives we can have multiple points where a party could always vote this way and each one would be as coherent as the other but some would lead to dishonest court some would lead to honest one and some would lead to court just giving random results. So what we call this is Nash Equilibrium. It's a Nash Equilibrium when everyone's strategy is in such a way that after everyone knows everyone has strategy no one regret his own strategy and changing from one strategy to another does not have a lot of sense if you don't think that all the parties are also going to change your strategy. So that's why it's really important that at the beginning the court give honest ruling because this way no one would have an interest in changing just for him a strategy to a dishonest one. So how can we verify that it works how can we verify that it's with experimental crypto economics. So at the beginning you don't link that to real contract which we are smart contract enforcement but you still have draw and still have incentive in term of token redistribution. You take a set of dispute that you know are already solved but that you don't prefer to draw and you look at what the result will be compared to what you call your own green truth which is the result of the dispute you had before. And you can use metrics which are inspired by machine learning to evaluate all the ruling of the court is actually related to the ruling of your green truth. So basically you do machine learning with human as classifier. So thanks everyone for listening if you want you can contribute we have a basic version of the court some habitable smart contract you can interact with contract on the testnet and we are still willing the Web3 interface to act with you in a nice and user friendly manner. Thanks.