Status: the post is a little in progress yet. I plan to add more information about how it's better than the existing systems and rethink how the decentralized prediction market (the truth component) would work so that it's trustworthy. JET (justice, equality, truth) is a a decentralized governance and coordination system. The goal of the system is to make people take the actions that maximize cumulative utility of the entire collective. JET is a strategy profile that under certain conditions is also an equilibrium in the human world. JET can be used to achieve coordination with regards to artificial intelligence risks, e.g. it can be used to implement an AI development pause, create rules that prohibit developing technologies that contribute to biological risk (and agree on what those rules should be) and ensuring that AI development will not increase inequality... The system can be facilitated through a locally-run, open-source software that communicates with the software of other users through a peer-to-peer network. The software has to be locally-run, open-source and communicate through a peer-to-peer network because if there is a central server that runs the software, then the people in charge of that server would have too much power, making the system fail at its purpose. Why don't people take actions that are the best for the collective, without any system: 1. Because they don't want - their utility function is different than maximizing cumulative utility of the entire collective. 2. Because they don't know which action is best - people have incomplete information and different people have different opinions about what is best. JET aims to solve those two problems while minimizing maintenance costs of the system. At the end, I will write why electoral democracy doesn't perfectly solve those problems and how JET addresses those problems. # Advantages of JET %% Show how those advantages would help on the example of implementing AI pause %% Generally, JET has the following big advantages, comparing to other systems: 1. The rules are decided based on aggregated knowledge of all people. It's better because all people have more knowledge cumulatively than the elected individuals alone. 2. Enforcement of the rules is based on aggregated information of all people - people contribute information (evidence) about who complied / disobeyed the rules, and people are rewarded for contributing that information. In our existing world, people can also contribute that information, but they are not rewarded for it. So, in the existing world, an employee of a big company might not whistleblow because they might face retaliation, and they don't get any reward. 3. Indirect reciprocity makes it foolproof to exploiting loopholes in rules. In our world, if the rules are not perfect, then a company can hack the rules - intentionally do something harmful and avoid accountability due to the rules being imperfect. If justice is based on reciprocity, if someone has done something intentionally harmful, then the other party can reciprocate that (analogically, if someone has done something beneficial that the rules didn't reward, they can also be rewarded later). 4. JET has a mechanism for ensuring that the participants will remain reasonably equal in terms of their wealth. # How the system works ## 3 components JET consists of 3 components: justice, equality and truth. The purpose of justice component is to motivate people to take actions that are beneficial to others and avoid taking actions that are detrimental to others. In our current world, justice is solved by monetary trade and legal system. JET aims to solve that in a better way. The purpose of equality component is to ensure that people won't end up being very poor, while others are very rich. There is a logarithmic relationship between utility and wealth (utility = log wealth). All else equal, it's better when wealth is split evenly. In our current world, equality is solved by universal basic services. JET aims to solve that in a better way. Justice and equality are in conflict. Rewarding people for good things that they did makes people unequal because the rewarded people become more wealthy. The goal is to achieve the right balance between justice and equality. The purpose of truth component is to ensure that people can know the truth about what they need to know. In our current world, truth is solved by prediction markets and justice system. JET aims to solve that in a better way (also using prediction markets but better prediction markets). ## Justice Justice component will suggest to people what to do and what not to do. It will inform people how much rewarded or penalized actions are. For example, if you accomplish some important task, you can be rewarded. If you steal or kill someone, you can be penalized. The reward/penalty is that the justice system favors/disfavors your preferences. In other words, if a person X acts in a beneficial way towards person Y, then the software of person Y or the software of the allies of person Y will recommend actions that benefit the person X. The idea is that people will simply play generous and contrite tit-for-tat strategy - you are slightly more nice to other people than they are to you. But if they are nice to your friends (the people who are nice to you), that counts as being nice to you as well. Analogically with enemies, however if all people play by that strategy it's unlikely that there will be enemies, it's more likely that some people will simply be indifferent to some other people. On default, everyone is your friend. So, what that means in practice, is that everyone is your friend, unless they have a history of harming good people. In order for justice component to work, it needs to know: 1. What are a person's preferences? What does a person want and don't want? How much do they want it or don't want it? 2. What are the consequences of an action? 3. What actions have a person taken? How will the justice component know all of that? It's simple - it will ask all those questions to the truth component and it will get the needed information. Truth component is responsible for telling you the truth. ### Equality Equality component will be a voluntary wealth distribution / wealth insurance system. People put money into it, and if in the future they do worse than expected, then they receive money from the system. If in the future they do better than expected, then they don't receive money from the system (that money goes to those who did worse). It's a form of agreement between people that those who ended up unexpectedly rich share part of their wealth with those who ended up unexpectedly poor. It can be done in such way that everyone benefits from the agreement because of the logarithmic relationship between utility and wealth (utility = log wealth). You can read more about this here: [[How to stop inequality from growing (detailed)]] In order for equality component to work, it needs to know: 1. How much wealth does a person have? 2. How much wealth has a person shared with others? How will the equality component know all of that? It's simple - it will ask all those questions to the truth component and it will get the needed information. ### Truth Truth component is an advanced decentralized prediction market that is different from existing prediction markets in the following ways: 1. Existing prediction markets focus on predicting future. Truth component focuses on predicting past, present and future. 2. Existing prediction markets focus on predicting YES or NO. Truth component allows also to predict a numerical value, and possibly even an object (a mapping of certain properties to values). 3. Existing prediction markets focus mainly on predicting events. Truth component focuses on predicting events and other statements, for example: "is Earth flat?". 4. Existing prediction markets store knowledge as natural language statements. Truth component stores statements using a declarative[^1] programming language (and possibly also imperative and other paradigms) which make it easier to derive new knowledge programmatically, using algorithms. People can contribute information in an automated, programmatic way. 5. In the future, truth component can also accept other form of data like videos or images. Since the truth component will derive new knowledge programmatically based on the existing information, at some point it will become a decentralized artificial intelligence. 6. Existing prediction markets are centralized (they have a centralized database). Truth component is supposed to be decentralized. Users can contribute information to truth component. If their information turns out to be correct, they get rewarded. If their information turns out to be false, they get penalized. If the information is right or wrong is decided with the contributed information. In other words, a person X is rewarded if the people who contributed information after person X, contribute information that confirms the information contributed by the person X. [^1] When people think about declarative programming languages, they usually think about Prolog. But I have in mind a programming language that can represent also other knowledge that purely symbolic/logical - e.g. knowledge that is probabilistic. %% #### How can a decentralized prediction market work We want to create a service that allows people to know the truth, based on what other people know (in other words, we want to create a prediction market). We want that to be decentralized so that there is no group of people who are in charge of the system (which would risk concentration of power). Here's how the system can work. The service is implemented as a software. The software is open-source and runs locally so that everyone can see what the software is doing, as opposed to running on a server where only people in charge of that server can verify that the software is doing what the owners claim that it's doing. The software communicates with the software of other users through a peer-to-peer network, without reliance of a central server. What is the main difficulty with designing how such system could work? The main difficulty is that: how can we trust people. The people might provide some incorrect information to the system because they want others to believe in that information. The system is intended to be used for things like justice, i.e. deciding who committed, what the consequences of certain actions are (to know which actions should be rewarded or penalized). In those situations, people will have conflicts of interests, and they will want others to believe certain things. In order to know how the system can work, firstly let's model the problem as a game before we propose a solution. But I won't define that game completely precisely because it would take too much time. In the game, there are players - those are the participants of the system. There is some set of possible questions that the participants might want to know. For simplicity, let's suppose that all questions are yes/no questions. The players (participants): 1. have certain information (answers to certain questions), 2. and they also need other information (answers to other questions), 3. and they also want certain other players to believe that something is true (they want them to be convinced that the answer to a question X is Y). The players can communicate with each other. They can for example send an information to another player that the answer to X is Y. They can also send information that player A told them that the answer to X is Y. And they can also send information that the player A told them that the player B told them that the answer to X is Y... and so on... In the game, each player need to guess the answers to the questions that they need to know. The goal of any player A in the game is to maximize: 1. the number of questions that the player A will answer correctly, 2. the number of times that any other player answers a question the way the player A wants them to answer. The game is repeated multiple times, and the players have a memory of the past games (I will refer to each game as "turn"). After each game, the players learn answers to each question in that game. What would be a pareto-optimal, subgame-perfect equilibrium in such game? I propose the following strategy profile. I don't know if the following strategy profile is pareto-optimal and/or subgame-perfect, but if it's not, then at least it's close to being pareto-optimal and subgame-perfect. Each player A assumes that there is a number from 0 to 1 that represents trustworthiness of any other player B. If trustworthiness of player B is for example 0.7, it means that player B says the truth 7 out of 10 times. At the beginning, the player assigns some prior trustworthiness to each player. Then, after each game, they update the trustworthiness of each player using Bayes theorem (or alternatively using gradient descent) based on if what they said turned out to be true or not. Each player tells the other players what they know (as long as it doesn't conflict with what they want other players to know) and tell others what they want them to believe. However, if they want some other player to believe something that is not true, then they need to balance their trustworthiness (which will go down if they lie) vs their need to make them believe into something that is not true. The incentive for a player to be trustworthy is so that their trustworthiness (their reputation) is high. They need their trustworthiness to be high so that they can lie successfully whenever they need that. That strategy profile will result with what people know being a mix of: 1. What is actually true 2. What people want others to believe is true That prediction market will not always tell people what is true (or even what people believe is true, if we sum their knowledge). For example, if someone wants people to believe that they drive a Ferrari, and they drive a Mercedes, but others don't care, then the system will tell that they drive a Ferrari. But that person will have to pay with their trustworthiness for others believing in that. Later, we can improve that system so that: 1. It allows to bet on numbers (and other data types), not just yes/no. 2. It takes into account that people might want to know something, or make people believe something to different extents (so, we assign priority to how much people want to know something, or how much they want others to know something). If people using the system decide that they want to have a truthful system that tells what is actually true, and not what is actually true + what people want others to believe, then it's possible to build a system like that on top of this system, assuming that people want that. But I will talk about this later. %% #### Will the system work? In order to know if the system is going to work, people need to use the truth component in a way that it's supposed to be used - they need to vote for what they think is true. But what is "true" (and therefore what people are rewarded for) is decided based on what the people that vote after them will do. So, in order for a person to win in that prediction market, they need to vote like the people after them will vote. So, a person will vote for what they think is true, if the person believes that the people after them will vote for what is true. Therefore, the question: will people believe that other people will vote for what is true? People to some extent know what is true, so what they think is true is close to what is true, to some extent. I also assume that the cumulative knowledge of people is close to 100%. Therefore, if a person believes that other people will vote for what they think is true, then that is sufficient for the person to vote for they think is true, and therefore for the system to work properly. Therefore, the question is: will a person believe that other people will vote for what they think is true? How the person will vote depends on what precedent has been established with regards to how people vote. If in the past, people voted for what they think was true, then people will expect that other people will vote for what they think is true. But if in the past, people voted for example in a way that benefits some group of people, then people will expect that the other people will vote in a way that benefits that group of people. So, if a bad precedent has been established in the past, then the system will not work. However, if a bad precedent has been established, then people can intentionally establish the right precedent. The question is: how to establish the right precedent, once bad precedent has been established? The problem is that in order for people to establish a precedent that they vote for what they think is true, then a person has to vote for what they think is true, but if a bad precedent has been established, then it is detrimental for that person to vote for what they think is true because they will lose money in the prediction market, if they do that. So establishing a good precedent, once a bad precedent has been established, is a collective action problem - it is beneficial for everyone that everyone votes for what they think is true, but it is detrimental for a person to vote for what they think is true, if a bad precedent has been established. That collective action problem can be solved by building the right precedent gradually. The idea of gradual precedent building is that people gradually start taking the cooperative action more and more (in this case, it's voting for what they think is true). It might not be in the interest of a person to completely start to take the cooperative action right away. But taking cooperative action carries certain benefit because if you always cooperate a little bit more than others, then you gradually get things closer to being optimal. That benefit might not outweigh the cost of the cooperative action, but taking the cooperative action to a certain small extent also has the benefit that at some point the cooperation will be optimal, and the cost if very small if the action is only a little bit cooperative. So, it is in the interest of people to gradually build the precedent of voting for what they think is true, after a bad precedent has been built. Once the right precedent has been built, the system should reward those who help to build the right precedent the most (those people will make money because in the end people will vote for they have voted). That reward creates an incentive to vote for what they think is true from the very start of building the right precedent. Therefore, people have incentive to vote for what they think is true at any point. That is not necessarily true, when there is not too much future ahead. For example, if people expect that the world will end soon, then they don't have interest in using the system in the way that it's supposed to be used. Because they won't live long enough to experience negative consequences of that. I also have another idea of how the decentralized prediction market could work. %% # What is wrong with Electoral Democracy The current alternative is electoral democracy. Electoral democracy has the following problems: 1. The elected officials make choices solely based on their own knowledge, not aggregated information from all people. The elected officials don't have knowledge about everything. Better decisions could be made if the decisions were a result of combined knowledge of all people. 2. People don't know who they should elect - when people vote in elections, they don't have incentive to invest a lot of their time to make a good choice, and they also are not competent to make the choice of electing government officials. 3. Corruption - rich people can corrupt the government officials. Directly giving money to government officials to make some choices is illegal, but there are probably some indirect ways to do that. 4. In our existing electoral democracy, there is some things wrong with how election happens, for example you can create a false candidate that will steal votes from another candidate with similar ideology, helping a third candidate to win. This is not a problem with electoral democracy per se but with the existing electoral democracy. 5. A problem with our economical system is that people can screw other people over. For example, people who have better negotiation ability can negotiate better deals than those who have worse negotiating ability. People can also make other people sign agreements that are detrimental to them. There's nothing in our system that can retroactively hold people accountable for screwing others up, once we learn more information about effects of people actions and their intentions. 6. Too much power between elections - the elected officials don't always have strong enough incentive to act in the interest of citizens. Their only incentive is to be re-elected next time, but it's not a sufficient incentive. Futarchy is another alternative, but I won't compare my system to it, because I don't know Futarchy in detail yet, so I won't make judgments for now. As far as I know, Futarchy is also a good system, although probably there are some problems with, and probably those problems can be addressed. # How does JET solve the problems of electoral democracy 1. Decisions and rules are made based on aggregated information, not solely based on what the elected officials know. 2. There are no elected officials, so there is no problem that people don't know who to elect - that they don't spend enough time thinking through it and are not competent to make that decision. 3. There is no corruption, although it is still the case that some people will have some degree of unearned privilege. 4. There is no election, so there are no problems that come with elections, like creating false candidates to steal votes from another candidate with similar ideology so that a third candidate can win. 5. With JET, it's harder to screw people over because if someone harms another person and people initially won't notice that, then if in the future people will notice that someone intentionally screwed someone else, then the prediction market will update its beliefs and the harm that has been made will be reciprocated. It also works in the other way - if someone does something good that initially goes unnoticed, then once people update their beliefs, it will be retroactively rewarded. 6. There are no elected officials, so there is also no problem with incentives of the elected officials. %% # How to implement that system in our world Phases: 1. Relying on weaker incentives from previous strategies for building incentives (theoretical justification why it's in the interest of people to act ethically and moving towards cooperative equilibrium). 2. Someone implementing software motivated by incentives from point 1. 3. In order to solve chicken and egg problem, people signal their interest in the project (I want to be informed, when the project has a lot of users) and there are also 3 other solutions for solving that collective action problem. I will describe them later, but they are analogical to what I have described in [[Alignment between humans/Archived/Cooperation norm/How to establish Cooperation norm|How to establish Cooperation norm]].