Public epistemics and futarchy by Robin Hanson | Devcon SEA

[Music] long ago I was part of a group of people who imagined the worldwide web before there was a web many people didn't think it was very plausible I left grad school to go off to join people working on that but then within a few years I started to have doubts about some of the Visions people had about the web and what it could achieve and I tried to think about other ways to solving that problem and I came up with the idea because I was hanging around a lot of rabid Libertarians that betting markets would be a powerful way to address many key questions in society and uh I started to write about that and within a few years I helped develop some exchanges and uh experiments and then about 10 years later I.E I started writing about this stuff 35 years ago 25 years ago I realized that the highest value of information aggregation is when you can inform decisions closely so I focused on what I call decision markets advising markets that advise decisions I wrote on that about 25 years ago and one of the concepts I described was futarchy and then uh in the decade after that I was working on combinatorial markets which I'm going to tell you a bit about and then in the last year we've seen a number of concrete experiments on this concept of futarchy or using conditional markets for governance I'll tell you a bit about and that's exciting and why I'm here and why you should just wait if necessary till you're old have ideas when you're young and wait till you're old because then by then something might have happened so I'm an economics professor and we social scientists have a number of lenses by which we understand society and Human Behavior Uh in economics the most straightforward standard one is just exchange of goods and services but we also have Game Theory where we understand power and strategy and political science is big on that sociology and anthropology are big on understanding things in terms of norms and Prestige which I've been focused on for the last year and catching up on things I didn't understand but another lens for human behavior is the idea that we reveal infer and aggregate information by our behaviors and we have a lot of social institutions that do that and in some sense the activity of economists in the last half century has mostly been focused on modeling this aspect of human behavior how we are signaling showing inferring and aggregating information and when you look at the many institutions we use for that one small Institution now stands out for its exceptional abilities and that is except speculative markets uh in comparison after comparison they just beat all the other institutions in terms of their ability to produce accurate estimates uh timely and uh using modest amounts of resources compared to other things so that creates this Vision I have which is could we use speculative markets a lot more elsewhere in society to produce the information aggregation that we are now producing in worse ways through other institutions like gossip news voting Academia Etc now these four lenses probably aren't everything there's a lot of social dark matter and that should make us a little cautious about any claims we make about how to change society but we should be trying this vision and seeing how far we can get with it now one way to see my vision here is just to for the moment set aside your doubts about how feasible this is and think about all the different things we could apply this sort of thing too so for example we could have markets on if x dated Y how long would their relationship last when you have a new hire what would their employee evaluation be in a few years if you have a project with a deadline will it make the deadline how would that change if you had changed resources requirements Personnel you could admit students to universities and other places based on expectation of graduating or GPA conditional on admission you could fire the CEO uh based on whether that was expected to change the stock price uh you could do security analysis based on whether red what the chance of red teams penetrating security is given that you changed uh some approach you could do various kinds of policy in terms of changing crime policy how does that change the crime rate the FED changing interest rates how does that change predict inflation uh even things like antitrust uh you could ask whether an industry will be healthy or not in many years later if you make a certain an trust move once you I think once you start to see these examples and look around you you'll see this everywhere all over your world are decisions that could be improved by speculative markets just like even this conference we could have had markets about well on what date do we have it what location what kind of speakers we could have outcomes like how many people attend or other sorts of things all of those are possible now you hopefully I presume at this point you have doubts many most of you probably think yeah but this isn't feasible you just couldn't do this there must be something in the way because we're not doing it now this is I'm a teacher most of my students are basically always assume anything that isn't the status quo just must not work and even you probably have doubts so but let me go farther with the vision a little bit before we get into how these things work the grandest vision perhaps I tried to pick what would be the most aspirational grand applications of this idea of conditional markets and that would be governance so at the moment you can think of our democracy as a system we have a bunch of citizens the elect representatives who pass bills and if the bills meet a constitutional constraints they become laws and along this process we have various information institutions that inform both the legislators and the public about their beliefs about what bills have what consequences my alternative which I've called futari long before I knew the other Association of the word sorry is that we would still have legislators to decide our values to decide what outcomes we want and they would basically vote on and create a measure of national welfare but in addition uh we they would no longer decide what bills passed instead people could propose new bills say through an auction agenda mechanism and then we'd ask the markets is National welfare going to be higher if we adopt this proposed Bill versus not and we just adopt the bills that do so that would be a way to do governance full governance using uh betting markets so markets just have a lot of advantages over other institutions they're numerically precise they're consistent across many issues they're frequently updated turns out to be hard to manipulate actually uh they're and the key thing is when we do experimental ver tests comparing their accuracy to other institutions are either about the same or substantially better that's the key thing that excites me and should excite you about using this in more places and the vision isn't to just make betting markets and see who has fun betting there and only have the markets that people find fun the idea is that if you want to know the answer to a question you subsidize a market on your question and then that attracts people to trade because of the subsidies you put there and then you get the answer to your question but you paid for it so the idea is this is a market for information you offer to pay for the information you want people who have or could get the information they generate it and then you get better estimates uh so we know lots of Tricks about how to make this stuff work better market makers allow a lot of uh advantages you can limit sabotage it actually works better to have winners from one market only Only Winners allowed on the next Market that's more accuracy uh and you can have problems that are hard to judge uh be judged only randomly and bet on what would happen if they were judged we can do things like just having any sort of distribution of a curve and basically estimate the curve and you can make an edit and move the curve and then that's the new thing everybody else sees or you could have a big hierarchy like who's what the field of a of a Nobel Prize is going to be in and you could edit anywhere in that hierarchy at a very fine grain level or at a high level this is all quite feasible and to show you just how far we can go with the technology I'm going to tell you about combinatorial markets which have been implemented I'll tell you to the IDE is not only do we have a set of claims but we let people bet on all possible combinations of those claims and the straightforward way is you have an edit based interface that is you see a bunch of Ence you're allowed to go pick in any estimate you want and change it that becomes a bet and then the next person who sees it they'll see the number you put there but in addition you can assume any number you see assume any particular value and now everything else you see is conditional on the Assumption you made and you can make several assumptions and that's a general way to browse and edit a full common tutorial distribution and it turns out that the subsidy you require to subsidize a common tutorial distribution is actually no more than subsidy for the each of the individual claims now of course a full commentor distribution has a vast number of parameters too many to be explicit but there's a standard way to deal with that calls A B net and aasian network uh cuts the explosion by making key Independence assumptions and I was actually part of a project 10 years ago where we had a thousand questions we could have had more where we had a full Baye Network connecting all these questions so that you could edit any part of this network and your updates would propagate through the whole network and we did the exact updates and exact management of people's assets so this is actually possible yeah it might be a little harder than the blockchain I haven't worked on that um so do people actually do this stuff well for a long time we've had conditional markets on elections because we've had markets on who who's going to be nominated who's going to win so the ratio of those two is in fact the chance of winning if you're nominated and markets have long been giving the parties advice about who to nominate that mostly they've ignored but it's there but in this year we actually had two websites metaculus and manifold which are not full money prediction markets but they have some incentives offering conditional estimates on the election and number of policy things so this is a real thing that's now implemented and live for people to trade uh this election no doubt will be even more popular next time around uh many years ago actually on this combinatorial site I talked about we actually had conditional markets on changing the blockchain size for Bitcoin at least in order to estimate the consequence of policy there so this happened uh in the last few years as I mentioned we've had a number of efforts to apply this concept of conditional markets to governance first there is a secretive india-based Med government medical organization that's done roughly a half a dozen uh markets where they've asked hundreds of thousands of their uh customers to bet on various health polic IES that they've adopted or not and I think they adopted five out of six uh a little bit later we have the metad Dow which is apparently now done 23 proposals for governance of the metad uh and only six of those were rejected they have some customers now 14 proposals three rejected uh you know a lot of volume in trade metad is now apparently focusing more on using governance to advise say metad Dow Grant proposals uh estimating the consequence of a grant proposal and seeing which ones to fund there are some other organizations in the space and I'm apologize if I don't know about yours and not mentioning them but one that I do know know about is called buttery and they're also focusing on this idea of uh funding projects terms of estimating the outcomes for the projects and I'm uh advising lots of these efforts and in particular I am chief scientist here at uh fy. which was previously named quiver uh we have at least one customer which is nosis to do governance for them and we'll do it for ourselves and we are interested in the question of how to do governance for organizations without bothering the bosses who don't want to be replaced and one promising application there is the relationship between investors and managers often they don't they're in dispute about when to raise more money and maybe a govern a f talking mechanism can sit sit as an intermediary there to help them make that decision all right now but I warned you about social Dark Matter key thing is that this I've been describing simple elegant idea there are a bunch of issues that we worked through in a talk yesterday I went through a number of the technical issues and problems but just in general all real world things are not just simple elegant ideas they have a lot of messy details and so we just need real world experiments to work out those details because in some sense we don't fully understand human behavior there there's a lot of social dark matter there's a lot of ways in which we make simple theories that should work and they don't so we just need to try things uh let me elaborate a bit imagine we put an autistic person in the sea Suite of a typical Corporation they are a person who knows the business very well they are uh very knowledgeable and uh informed about key decisions but they have no social Savvy whenever a topic comes up they just blurt out whatever they think the truth is without having any idea whose agendas might be squashed by that such a person will just not last in the sea Suite they might become a trusted adviser of somebody else nearby but they won't be sitting there but that's in essence what prediction markets are they are very smart and knowledgeable things that speak the truth without knowing who's bothered by that and that highlights the fact that modern organizations a lot of what they do is politics and massaging of messages and so we need to search in the space of how to uh place these things in organizations and in 20 years ago there was a wave of prediction Market applications and hundreds of firms tried them but even though they typically had more accurate estimates uh satisfied users the experiments didn't continue largely because of objections from managements who really got pissed when the market disagreed with them and the market was right and so we need to deal with that so A more concrete example even think of a project manager who has a deadline they want to know what the chance of making the deadline but they also want to have a good excuse if they fail how will they do that well their favorite excuse if they fail is to say well we were going along just fine until the last minute when something came out of left field some weird thing it's so rare and it'll never happen again so let's just forget about it and the problem is the prediction Market gets in the way of that because it tells you a long mind before you were going to fail you're going to fail and you can't say we didn't know that until the last minute so you can see uh we need to overcome this sort of a problem in organizations in order to be able to field them and that's why we need to be doing these experiments which I'm excited to be doing and that's my talk