Liquid democracy in the wild with Andrew Hall | a16z crypto research talks

Transcript

well I'm excited to give the last and final talk of governance week I'm in the advantageous position of introducing myself so I will reserve the first 20 minutes to extol you with my background now I'm Andy Hall I'm uh I'm a professor in the at the Stanford GSB uh I've been uh studying decentral decentral governance for a while now I'm uh very privileged to get to work with the crypto research group here I'm going to present to you some joint work uh with my co-author and my pre-doctoral research fellow show Miyazaki who's at the GSP with me as the organizer of governance week I feel I made a very foolish decision to schedule myself last and Peter in particular is a tough act to follow and I'm not nearly as audite

nor as my accent as cool as Peter but I'm going to do my best to entertain you in this last and uh and final presentation um very big picture the motivation behind this project and a lot of other stuff we're doing to study governance in web 3 is this observation we have a lot of reasons both practical and philosophical that we think Democratic governance is a good way to set things up both in the physical world and in the digital world and we've been using uh tools to do that that were kind of come developed a long time ago when physical constraints were different than they are today and so many many people over many years have continuously asked this question there's

got to be better ways to do this now that we have access to so many new tools and so that has been a long thinking idea is that modern technology should let us improve the way that we run elections that we use voting rules uh and so forth and yet it's actually been pretty elusive to figure out what the right ideas to pursue in this space are and today I want to zoom in on one that I think is especially interesting I'm going to tell you a little bit about why I think it's especially interesting that is sometimes called liquid democracy which is just a really cool awesome name so obviously we should be into it um the funny thing about liquid democracy I'm going to tell you a little bit about how it works I'll very briefly touch on the history of it which is fascinating I

wish we could talk about more uh and then I'll and then I'll talk about why we're studying it but so this is a popular idea it's especially popular in Tech spaces though it actually has a Origins that go beyond uh Tech and what is the idea of liquid democracy how is it different from representative democracy there's at least I think four key attributes the first is it's open and what I mean by that is that anyone who's a voter is like a latent representative as well as a voter I can give my vote to anyone else I want I can say Peter I want you to be the the representative uh it's not I'm not constrained to a list of candidates that some some process has

selected for me so that's pretty interesting uh it's continuous I can change who I give my vote to whenever I want so instead of having elections at some regular period anytime I please I can switch my vote to someone else one idea that sometimes comes up in it but isn't always present is sort of similar to this idea I can change it whenever I want in some ways this is almost mathematically equivalent I can give my vote to different people for different types of decisions so instead of having one representative I'm asking to be a a jack of all trades I can give my vote to Peter when the decision is about uh rank Choice voting and whether we're going to implement it I can give it to Tim when it's about uh designing the economics of

the blockchain and the final one and this is actually important to the word liquid but often actually not present in most interesting voting systems is the person I give my vote to can then further give it to someone else and that's the liquid in some conceptualizations of this comes from the idea that that then allows votes to flow to the people most expert in making the decision so we put these things together and we have this kind of fascinating really radically different looking voting system than a normal set of Elections for Representatives right so you can imagine it's this very open smooth system where lots of people are giving their votes over to other people

with the hopes that they're then aggregating into the best set of people whoever they may be uh holding the most voting power that's kind of the idea well let me say one last thing on this um the idea for this actually predates modern technology the the it's debatable who really came up with it first but in some sense Lewis Carroll author of Allison Wonderland is credited with the first concrete fleshed out proposal for something that looks like liquid democracy surprisingly he did not come up with as cool a name for it as liquid democracy despite his creativity but the idea really goes back a long way and then Gordon tulk very famous uh sort of um public Choice academic in the mid

20th century was one of the first people to to to take this idea and to say now that we have modern technology his ideas were it was mostly premised on television uh and um uh phone calls observed that we're getting closer to being able to implement something like this and so tullk proposed a verion where uh you still elect representatives but you're the all the deliberations of Congress are broadcast on television and at any moment you want you as a voter could override the vote of your representative you could reallocate your vote uh continuously as things are being voted on um then in modern day it's really come alive with digital

technology and the idea that now we can really Implement quite directly this idea that we could all transfer our votes to whoever we want um as we're voting and so it's really it's actually quite old idea but it's gotten a lot of energy as technology has progressed broadly and so like so why did we pick liquid democracy I i' I hope I've made the case that it's very cool and the question that uh this this age-old question that we should figure out how to improve democracy in a very broad sense uh is reason enough to study something like this but it actually turns out to be a very direct practical importance today in web 3 in other parts of tech and in finance so I tell a little bit about that to explain why I think this is such a great topic for

more research so first of all in web 3 the majority of the most important and important maybe not the right word but like largest Dows they generally use something either the governor Bravo contract which we discussed earlier this week or something similar to it and that smart contract implements something that looks a lot like liquid democracy in particular it implements uh the ability for you to delegate your vote to anyone else they don't have to be an official candidate or anything like that you can delegate to them I'll just note and this will come up again later there's no redelegation so I'm I'm a little bit abusing the term liquid democracy by calling what's going on in web 3 exactly liquid democracy but it's very close and

I think a lot of the most to me at least the most interesting parts of the system are the ability to delegate to anyone else the the red delegations in practice I think it's quite unlikely we would ever see lots of that but maybe I'm wrong so it's already today stay active in important Dows and web 3 so we should want to know how it works how we could make it work better from its use in web 3 we can learn about its applicability to lots of other domains as well and so I just want to note two areas which that I think are ripe with opportunity here one is the rapidly accelerating world of what I would call AI governance where all the major AI companies including meta where I've done work on this topic

are developing different systems for the governance of these of these new technologies and in particular they're very concerned about setting guard rails around AI just to pick one example I'm not at all picking on this company because every company has faced this problem but when Google released Gemini there is this huge backlash uh based on sort of guard rails that had been put around how Gemini worked that seemed to reflect some people's values in societies and not others people got very upset it had to be pulled back there is out of many experiences that all AI companies have had like that there is this constant thread of rhetoric of we need to find a democratic way to set the guard rails around AI today most of that work is

being done via sortition so so anthropic open AI meta have all used some kind of a Citizens assemblies approach kind of like what exactly what Bailey was talking about yesterday to do that in the future as these things mature I think it's very likely that they will implement systems of voting for those decisions and I have already heard people working in that space asking me what is liquid democracy what is delegation we've heard that's what we should be using so I think that that conversation is already out there that'll be a very important application and of course web 3 AI uh will be a big part of that conversation as well the final category which might be the most surprising to you unless you're in the finance space is very surprising to me

is that the large the very very large asset manager companies are right now today implementing forms of delegated voting systems so proxy voting has always been around to some extent in corporate shareholder voting when I first heard about liquid democracy I was like great these Geniuses have reinvented proxy voting uh but it turns out you know proxy voting is quite crude and very very limited in corporate shareholder voting there isn't very much of it in practice and today Black Rock Vanguard and State Street which together are by far the three largest holders of voting power uh in in in publicly traded companies in the US because of the ETFs um that they hold

that they sell to customers they are all in the process of developing what they call voting Choice programs that pass through the voting power of their ETFs to their customers and the idea is we as these large uh asset holders we shouldn't have an undue amount of power over the votes that are about companies that you our customers are really the underlying owners of we would like you to vote uh instead of us it's uncomfortable for us that we're the ones voting your shares and so they are building these systems in which they're attempting to encourage their customers to vote more of their shares instead of Black Rock Vanguard State Street and the

manner in which they're doing this I will tell you more about it later in the talk is like a highly constrained version of liquid democracy and it seems extremely likely to me that in the future as they try to expand it and open it up they are going to it is going to merge with what is going on in web 3 with liquid democracy so for all three of these reasons I think there's you know timely practical reasons to want to study this not only the super cool big picture question of like can we redesign democracy around principles like this so that's the motivation there is already lots of research on liquid democracy I'm not at all trying to claim we are doing the first research on liquid democracy but I want to explain why the large literature that there is doesn't answer

most or perhaps all of the Practical considerations we need to know to know how to implement it well and this will touch a little bit on a theme from Peter's talk so thought Peter did a great job and one of the the spirits of this visit this week I think is bringing together people studying governance across fields and one thing we've observed I think in the history of all Academia is that when you can compare insights and ways of thinking across field you can discover new ways of conceptualizing things that help you make rapid progress on practical questions Peter I think did a great job of showing why the social Choice approach to the study of voting systems has been extremely useful and from a computer science perspective my impression is it turns is a very fruitful way to plug algorithm you know

algorithmic analysis into the study of voting turns out to miss some really important insights that require the use of Game Theory and strategy possibly in a way similar Tim I don't know to the growth of the the sort of EC type community in in other parts of computer science that there's this fruit to be born by the combination of strategy and Game Theory with cs techniques I think Peter did a really good job of showing why that's definitely true for the study of voting systems and is true for liquid democracy so one thing we see going on with liquid democracy Now is a move from a computer science literature that has been almost entirely social Choice driven to a strategic literature like this mors at all paper I'm citing here that asks when voters are strategic will

you still get good properties from the system of Delegation what we want to contribute I am not smart enough to do that kind of work I am an humble empiricist and it turns out that in the same way that bringing Game Theory into the study of liquid democracy is is very very fruitful bringing data into it is very useful too because it turns out that the Practical considerations of implementing liquid democracy have a lot to do with how actual people will behave with within this extremely weird alien system of voting kind of like Peter alluded to with rank Choice voting and so we have a very rich theoretical literature on liquid democracy there's already others in economics in the social sciences starting to apply Game Theory tools to a particular set of

questions about whether liquid democracy successfully gets people to allocate their votes to the right you know to the people who have more information in theory but we have almost no empirical work as asking in practice does any of that have anything to do with how people really behave inside these systems there is one fantastic paper it is as far as I know the only empirical systematic empirical study of liquid democracy uh it's this cling itall paper and it is about the pirate party in Germany which implemented a form of liquid democracy through a bespoke online platform it's very cool uh to make internal decisions and so that and that paper is really focused

build from this theoretical is very focused on this question of is there overd delegation Within These systems which would be a sign that we're perhaps not aggregating information effectively we're overly centralizing because we're all giving our votes to to Eliza or Robert or something and then we only have one voter and that would be bad and that is an important practical consideration but there's only so much progress you can make studying you know one question in one empirical setting there is no existing work that looks across a set of cases of using liquid democracy and there's none that have ever looked at it in a in a place where the decisions are high stakes with no no offense attended to the pirate party in Germany but the internal decisions of that party have not generally been super

high stakes uh in the way we would think about a political system making decisions in general um so what we want to do in this paper is really get to those practical questions about liquid democracy and I want to lay out for practical questions about how liquid democracy Works in practice um that have not yet been answered so the very first is the most basic Peter alluded to this participation problem if you make it possible to delegate to someone else does anyone pay attention notice that that's a possibility and do it that's a very basic question not surprisingly to be tractable and useful theoretical models tend to make optimistic assumptions about that just to figure

out like if this thing works you know if we make optimistic assumptions can we establish the possibility that liquid democracy is going to aggregate information in in favorable ways and so forth there's in practice though there's a very basic question of does this work at all do does anyone push the buttons to delegate their vote so that's the very first thing we're going to show you evidence on uh we're going to show you kind of like who's actually use you know who's doing this how many people are doing it and that I would just say by the way it's just a theme of web 3 and Technology governance in general is we can create these extremely baroke really cool engineered systems for governance they often crash against the rocks of the reality that participation is so low I'm going to show you some in this case

I'm actually quite optimistic that that's not the case Okay then if you have people participating in delegating which ones are the ones that differentially select into doing this not everyone is going to pay attention and delegate their vote who are the ones that do who are the ones that don't if they decide to delegate how do they make the decision of who to delegate to you know are they choosing carefully are they choosing wisely or not and then finally and maybe most importantly if you think about these practical considerations if you think about uh a web 3 AI platform setting guard rails through this system or you think about Black Rock one day implementing it for shareholder voting how do you actually build a system around these voting rules that gets people to pay attention lets

them figure out who their options are creates the conditions for campaigns where you can understand the debates and decide like yes that's who I want to delegate my vote uh to that's all external to theoretical models of liquid democracy but the first thing a practitioner would want to know if they're actually building a liquid democracy system so I'm going to talk quite a bit about that fourth question our progress on this question is extremely simple to explain we just went and got data for the first time from a set of high stake settings where liquid democracy or something like it is being used in practice our contribution is that simple turned out to be way harder as a humble political scientist pulling data from the blockchain was not as

straightforward as I've been led to believe but we did figure it out uh uh and we just have this data that no one has had before and that's that's the uh in a nutshell what we've done there's just this incredible laboratory of these Dows uh that are making very high stakes decisions using this technology so if for anyone in the whole world across any of these settings interested in it this is where I think they should look first uh so what we're going to do is we're going to zoom in we're going to make a couple of decisions to kind of very much zoom in we're going to look only at ethereum based Dows that was just where we had the most knowledge and we're most comfortable and we're going to look only at onchain voting which is going to massively restrict our scope we could

def easily brought in this to include all the voting that goes on on Snapshot uh much of which also has a liquid democracy component but we really wanted to zoom in on the high stakes the self-executing irreversible decisions that kind of like are the life blood of Dow governance and just as a reminder I think this is a kind of repetition for visitors but just to make sure you're clear on the setting right these are settings where there are governance tokens they can be sold and traded and it is one token one vote and so when I'm talking about delegating I am talking about you taking your wallet which has these governance tokens in it and delegating them to someone else technically within the governor Bravo system by the way you have to delegate

all the tokens in your wallet of course you could always split your wallets and only delegate from one but that's just something me to note and just like in the theoretical laying out of liquid democracy I told you before it is indeed open any address can delegate to any other address and it is so is open in that sense and as I mentioned the important thing we think really matters is the stakes of these votes can be quite High I will note one more time just in I always say this because there's some real liquid democracy purists out there not sure there's any in the room today but this is not there's no redelegation allowed the governor Bravo smart contract does not allow for redelegation can I UND delegate at any moment yes yes people have debated one of an there's so many

interesting design choices to make here and one is you could try to lock delegations at some point there have been some high-profile moments in which people have undelegated at Key at Key instances I assume that the answer is no to both but there's like two Notions of redelegation there's the person I've delegated to Red delegating or like me delegating twice in the case that somebody abstains oh deleg twice so you mean I delegate to you you abstain and then I move my delegation no I say like I delegate to Andy but uh if Andy doesn't vote then delegate to someone else if Andy doesn't show up I see yeah yeah so that's not implemented in the in

Governor Bravo at least I assume neither is implemented that's right that's right I will say and I'm going to come back to this the participation rates of delegates are lower than I would expect in our data and so I think this issue of exstension is is quite important uh the the participation of delegates tends to be higher than the average token holder but it's nowhere near 100% um and I'll show you data on that what I really want to emphasize so that's like I think the first contribution of a project that we're very excited about is like we just got this data and we're going to tell you facts about how liquid democracy actually works in practice and suggest to you that it looks a lot different from the theoretical debates over it but the second thing which I think at least from a social science perspective and

maybe also from a practical perspective might be the more important part of this is these Dows that we're going to study have intervened over time and they've each built not each of them a set of them have built these online platforms to do the work I was talking about of creating this social and tooling environment around liquid democracy to make it actually work and so they have built out I'm going to show you what one of them looks like these websites that facilitate getting you to delegate and I just want to pause make and make it clear to you why that's so important these theoretical models which again I find very valuable I'm not I'm not trying to pick on them don't get to any of these practical things so they

just say like if this all works if everyone knows everyone else and you want to delegate here's how it would aggregate which is something we need to know to understand if this idea makes any sense once we've done that though we then have to step back and say how would this actually work for a real governance system and the first thing you'll be confronted with is it's really not that useful as a governance system if I have the option to delegate to a bunch of addresses out there and I don't know anything about them I don't know who they are they haven't told me anything about their positions they're just addresses in floating in space that's not a governance system that's just silly and so of course this was bootstrapped through people participating in the forums and and posting on X and so forth but most of

the big Dows over time we're like we need to build some kind of Hub that will get everyone to coordinate get people to sign up to be delegates get them to say a little bit about why they want to be a delegate what they would do if they win your vote your tokens and then make it pointand click accessible for the voter to delegate to the person they like and there's every Dow has done this in a different way but I just want to point out if we step back about these practical opportunities to build out systems of of of Delegation and liquid democracy in other places it is of incredibly high value I think that we have a set of dows where each of them built their own version of this Hub or or they I mean they've shared some

technology but built a version of this sort of Hub or delegate program as we're going to call it at different times and that's going to allow us to kind of exploit I would call it in the social sciences we' call this a natural experiment where we're going to attempt to get some actual causal evidence here's how voting within a liquid democracy system changes when you go from it being this state of nature to some extent to this coordinated activity that has a website and and all of these other bells and whistles and so we're very very excited that the way web 3 Works has facilitated for us this natural experiment as each Dow has tinkered with its the creation of its own version of

liquid democracy at different times I mean I think do we take these two things together and we think wow we have this very like this is a very novel opportunity to understand what's going on with liquid democracy in the wild as we call it okay in case I run out of time let me tell you what we find right now so that everyone knows first I'm just going to tell you what are the overall rates at which token holders are delegating their tokens those rates to some people I think they look low we're talking like 20% of tokens in circulation or something it's actually pretty comparable to corporate shareholder voting depending on your goals for voting that you could call that a higher low number it's higher than I expected my experience has been you know getting participation in these systems is very hard 20% was higher uh

than I expected what I think is particularly interesting when we ask this question of who's doing the delegating to the extent we can tell and the Civil thing here is a huge impediment to making strong statements it looks Bottoms Up that is it is smaller token holders delegating to experts more than we're observing large token holders delegating to someone else that was also quite surprising to me could be large token holders chopping up their wallets um but I'm not not sure they have other strong incentives to do that I don't think they care I don't think they were anticipating me drawing this histogram so we just but we just don't know uh we can't tell you a lot that's like yes the voters are super

ultra informed and they're choosing the delegates that match their reasoned views and so forth we just don't know but I'll tell you a basic fact that I do think is encouraging which is delegates who have an established track record of voting more often as I said the delegate delegates participation is itself a challenge those who have a track record of participating more often have received more delegations from more addresses so it suggests at least this basic thing that somehow the system is helping voters to figure out like there are delegates who I can delegate to and they'll kind of do more and there's some who do less and the ones who do more are getting more delegations which seems good and then last and this is I think the most interesting part when the Dows

build these programs these systems around liquid democracy to help people we see first of all they're quite effective at getting rates of Delegation to go up so you get a bunch of people delegating who weren't before and that actually then induces higher rates of overall turnout in terms of voting and I'll explain to you when we get there why it's actually not obvious that that would be the case and so it seems pretty encouraging that you build these programs and you're getting more people delegating and more people voting and I think if we look again back to these all these different other possible applications of delegate delegation voting that lesson is going to I think be quite important that you can build these programs that probably will learn

ways to build them better and better over time and they're doing their job they're getting people uh using these systems so very quickly we're just going to pull all the wallet level data on the onchain votes in all ethereum based Dows the data I'm going to tell you about today goes roughly 2021 to 2023 a challenge with studies in the space is that new data is coming in every day which is great for the world annoying for researchers trying to write papers uh Dan Bonet has a vision where all papers should just be dashboards that update live every day which I love maybe chat TBT could rewrite the findings as the finding St um uh but we're going to take this snapshot and time and tell you about this time period we are going to

then bring in data on tokens and circulation that's going to be our denominator you could use other denominators we think that's the right one to use to analyze governance participation so it's kind of like here's the set of tokens that could have been voted could have been delegated here's how many were delegated were voted and then because of these restrictions we put on because it's only the onchain voting because it's only ethereum based Dows and for a couple other reasons we're going to end up with 18 Dows of course we would love to have more but that's a decent amount covers SE roughly 1,700 proposals that were voted on from roughly 250,000 voters so it's a pretty good chunk of information that we're working from let me just zoom through a couple of the basic facts cuz I want to make sure to get to the the um

effect of these websites before I finish so this is sort of just to give you a sense of the proper nouns uh not pun intended for nouns Dow fans uh the names of the Dows in our in our sample so this is just a descriptive plot on the x-axis here we have of all the tokens in circulation What proportion have been delegated to someone other than yourself note there's this little thing where Lots you know you have to delegate to yourself to vote in Governor Bravo so ignore that we're not counting that you have to delegate to someone else uh uh and on the vertical axis is the proportion of tokens and circulation voted and this is on average across all proposals and you can see I mean there is a bit of a you can see there's a positive correlation it's noisy and you

can just see that different Dows are very different from one another some Dows have quite a lot of Delegation uh some Dows have very little some Dows have higher turnout though never you know extremely high turnout uh and some have very very low turnout but this is just to give you a sense that this is the universe if you look at the names here like we have all you know a lot of the Dows you might think of when you think of Dow governance or in our sample if you plotted that for like concentration of token Holdings in non-liquid uh systems would it look sort of the same tell me what you mean by that like concentration concentration and participation so like levels of Delegation could also correspond to something like a a level of concentration in the eventual Vote or

level of like you know how many tokens were in effect being like actively voted or something like this is is there in in a world with like a plurality voting scheme where you can't delegate is there some analog of like this chart we could imagine thinking about to figure out whether it looks the same I'm not sure I'm falling when you say concentration are you trying do you want to like measure the the heaping in voting or you're just saying like oh sorry well sorry you're right this is like levels of this is total level of Delegation not like the resultant concentration so that's not the analog I'm trying to figure out I I don't have a prior on what this picture should look like so I'm trying to figure out what like what it should look like what it would look like in some other voting

scheme that didn't let me show you some other pictures and let's revisit the real this is just a figure to put some proper nouns in space but yeah I'm going to show you some evidence on concentration in a moment and then maybe that's the right um right thing to think about for the analog so here what I'm showing you is the average as a proportion of all tokens in circulation across all these Dows the average rate of Delegation so how many tokens divided by total in circulation have been delegated and how many have been voted that's the purple line that's lower and and the two points I just want to make here are the rate of Delegation you could call it high or low it's around 20% of all tokens in circulation um and

the rate of turnout is actually lower than that and this is just one thing that you can sort of infer from this plot is that not all delegates are voting all the time right if all delegates were voting over the time then the turnout rate would have to be at least as high as the delegation rate and that's not the case um there's some time Trends over here which we could explore but I'm not going to talk about that right now this scot is getting towards the concentration question a big motivating question in the study of liquid democracy to date has been okay we want to aggregate information there's generally not a notion of ideological competition in liquid democracy which I think is a missing part that we need to study more there's just this question of

like how do we aggregate to make the best decision in the sort of like Condor um uh tradition and de and lumpiness of delegations become a concern there if a few delegates get tons of voting power delegated to them because they're the most popular the smartest or whatever even if they have the best signal of the correct Choice you've lost so many other signals by G having everyone give their vote to this one person that you might uh it might be bad for the system what I'm showing you here is two over overlapping histograms the underlying dark purple one is the underlying share in terms of tokens so it's um again as a percent of circulating tokens what

percent does a given address have modulo cble issues here as usual uh and in light blue is what that looks like after delegation so when you think about delegates now how many tokens do they hold for voting purposes they don't actually hold the tokens um what you can see is there is this much longer tail right where these are delegate there's way more delegates with higher shares than there are underlying token holders so indeed there is heaping going on here but it actually looks quite a lot smoother than what was doc has been documented in the in the pirate party now this histogram is truncated and there are a few cases I'm not showing

you of delegates who have obtained quite large shares of the vote and that's something we would definitely want to keep Ane on uh in the future got it yeah you're right that that's the analog and and I guess you know the the ownership structure is somewhat indd to the choice of voting scheme but that's pretty hardening right like it would be hard to imagine that a system that doesn't allow delegation would would have a figure that looks like dramatically better than that somehow yeah yeah yeah yeah yeah that's interesting okay this is the other basic fact I wanted to tell you uh which is that token holders on average tend to delegate to delegates who have a who are voting more often and so this is just each point here is a bin average of

many underlying points and this is just a correlation it's just showing on the x-axis for each delegate here is how you know as a as a fraction of all tokens in circulation how many tokens they got delegated to them from other people and on the vertical axis their rate of voting and what I want to emphasize here is like delegates rates of voting are not that high uh and there could be many reasons for that um but it is the case that the ones higher up here who vote in general vote more are getting more delegations so something is causing these two things to correlate it is not necessarily a highly informed process in which I as a token holder I'm like I want to delegate to the person with the highest turnout rate and I go check

there's many other possible explanations for this but that's sort of a basic pattern that we observe have you looked at this like like you could look within delegate over time as I get delegated more or do I does my voting rate go up the sectional correlation yeah and we haven't looked there's also things we could probably do with like after the website is rolled out and stuff like that no I haven't looked at that yeah do you also look at variation in these websites with how the delegates are displayed because I know development with like Agora on optimism and Unis swop is going from kind of displaying based on token power or to weighted random this such a good point

let me get I'm going to hold that for a moment and come back to it we haven't looked at it but I agree it's like exactly the kind of thing that needs to be studied next let me Zoom forward in the interest of time I'm not even going to explain this ugly looking graph I'm just going to say this this is where we're showing you the best evidence we can modulo the Civil problem that there's something bottoms up going on here this really surprised me this is basically showing it's medium to small wallets that are delegating at the highest rates the larger wallets are actually generally preferring not to delegate yeah H is this sort of wallets just or is this also smart contracts that you're looking at this is just wallets can you explain you're probably

getting something important that I need to learn about like what would be the what would you want to see about a smart contract oh no I think it's I guess it's better if it's just wallets but like if like Unis swap for instance a Unis swap pool was in there that held quite a bit of a voting token and that could be like a an address with a larger share of tokens which wouldn't delegate cuz it can't okay that's a really good point we should I'm pretty sure that's not in here but that's a great point we should definitely make sure it's not um okay let me let me Zoom through here okay so this is the part I wanted to make sure to get to as I said in the real world building a voting system delegation making it work is going to require doing

a bunch of stuff that the theories understandably don't countenance and what we're going to do is take advantage of the fact that different Dows rolled out what we're calling these programs at different times they also did it in different ways I'm not today going to exploit any of that information to learn more but that's definitely where we're heading is like can we learn from the fact that Unis swaps version was rolled out is different from optimism and so forth um but typically the way these programs look is there's a website that shows you all these delegates you could delegate to it has a convenient point-and-click button so if your wallet is linked you can just pointand click delegate uh you can learn about the

delegates sometimes there's a quote unquote white list where some committee or someone blesses some people as being like verified delegates and so forth you can see how there's a lot of like options for how you could do it this is kind of a prototypical example I'm pulling here this is unis Swap this one is built by Agora which is like a service provider for voting in general and and Agora is one of the more popular interfaces for doing this and the way it works here I just want to give you a sense and I want to suggest to you kind of like why this is interesting but also why there's definitely scope to do more in the future it's basically just a set of tiles uh showing you all the different addresses you could choose to delegate to and what Eliza was getting

at earlier which I think is super important is there's many options and so a challenge here this is like old school uh you know platform design user issue here we're going to have to figure out some way to like display this information to you and this has been a challenge for these programs so far so it's very very valuable that they're putting all these options in one place and making it easy for you to delegate but when there are many potential delegates how you order them potentially matters quite a lot and so you can see here and this has changed over time they've been kind of tweaking this over time which is I think makes sense you can fil filter it so here it's showing all stakeholders whove you know on all issues and right now it's weighted

random which I think means they take how many tokens the delegate has and then they sample how they order them so that your probability of being ordered high is higher if you're if you as the delegate have more tokens others have done this I think Optimus at one point was doing just pure random um Unis swap used to just sort by size and you can see how the way you do if there's a lot of slack in the system and users are going to default to top ranked choices if we've learned anything from web 2 the suggestion would be there's going to be massive effects of these defaults you can see how you might want to think really carefully about the different things that would incentivize we don't have a lot to say about that today I

just want to emphasize that that seems like a very important research point for the future if you sort it by size you're kind of like you're encouraging a Snowball Effect where the large delegates keep getting larger which could be very good if you think it's kind of a reputation system where getting large is a sign that you're a good serious delegate and you want to keep that going but you can also see how that would be a deterrence to new people wanting to become delegates and and would definitely potentially exacerbate this risk of overd Delegation over heaping in a small number of delegates and so forth so it's a It's Tricky how to think about how to do this um I'll just note again there's a very

optimistic thing going on in the background of these where the hope is that you click on these things and every delegate has a statement and has posted a video and you know all about the reality is you click through and many not just on un on all of them and many of the delegates haven't even issued a uh a position on anything or they have a one- sentence position that's like I'm going to vote to do whatever is like best for the Dow or something which hopefully is not an informative uh platform um uh in equilibrium um but that's kind of where we're at and so I think it's it's a fascinating space where we're just starting to build these things out we want to know what has their effect been so far and then we want to know

obviously what could we do to improve them uh in the future so I'm just going to show you quickly since I'm almost out of time these two basic questions which are have these increased rates of Delegation to other people and have they then increased turnout and I just want to highlight like those are worth asking right delegation's only going to increase if it's actually right that these frictions in the state of nature before these websites are big and we don't know like maybe these token holders are so into Tech that they don't mind that in the past they had to like I don't know delegate through the terminal or whatever you Engineers do maybe they don't need the pointand click system maybe they don't need the information they're already super aware so the first question is just what is the margin

what's the size of this margin of like making it easier doesn't matter but then the more subtle question is this one about turnout because if you make it easier to delegate and you get more people to delegate there's at least two things that could happen one thing that could happen is I care so little about all this I'm some token holder out over here in space I'm not paying attention to any of this I'm not voting and I don't care if you build a website to delegate I'm not delegating either and the people availing themselves of this new website are like the enthusiasts who were already voting and now they're like oh this is is cool I can delegate now so the net effect on turn will be nothing CU you're taking a person who's already voting and getting them to delegate on the other hand if the story is more like

in the history of democracy that there's a bunch of people who won't go to the trouble of regularly voting on every decision but will go to the trouble to delegate their vote once to someone else then this could have big effects on turnout and so we need to know which of those worlds we're kind of in and the data can help us answer that so I don't have a lot of time I usually like to give a whole uh overview to the to the Cs people of how we try to extract causality from observational data it's very cool one of our proudest achievements in the social sciences no offense to the theorists in the room uh uh but basically we're going to take advantage of this fact that different Dows implemented this at different times right and so I'll just just to wrap up just as I'm ending here

like the point is if we just compared Dows who have these programs to ones who don't and we said hey look the turnout rates are higher and the ones that built these websites that's the kind of thing someone might do and that would not be good evidence that these programs are actually the thing that caus delegation rates and turnout to go up because guess which Dows have the money and the time and the interest to build these it's the ones that are already more important more resourced already have more people involved already have higher rates of turnout so we need to do something to get from that correlation to that causation we're going to use what's called a difference and differences design where we compare changes for the Dows that did used to not have these websites and now have them to changes from a control group of dows that didn't

Implement them over this time period and just in a nutshell without going through this so I don't keep you too long we're going to find very large effects the the creation of these websites increases rates of delegating by about let's say let's call it four or five percentage points and so if your mean of Delegation is around 20% proportionately this is a very you know quite large increase and maybe surprisingly uh it causes quite a large increase in turnout as well so the creation of these programs apparently is drawing in token holders who weren't all already voting but who in fact are attracted to this new system choose to delegate when they

wouldn't have voted otherwise and so it's kind of I think it's a sort of an interesting discovery that indeed making uh a system of indirect democracy easier to use is going to in is going to increase the amount of tokens flowing through the system to voting okay so let me conclude uh with some practical implications uh of these findings for web 3 and and Beyond so we know that a major obstacle in all voting systems and liquid democracy right is the willingness of Voters to participate we have some signals here that participation in the Signum is somewhat broad but it's lower than we probably would like it to be we have some evidence too that the delegation with the sys a little bit Bottoms Up which

again I think is what we would like to see but we might want it to be more and we see that voters are at least somehow finding their way on average to the delegates who participate at higher rates so these all seem like building blocks of a system we could make work potentially there's tons of other questions we need to ask um and then finally we see that okay if you build these programs you can get quite a lot more people uh to participate but and I think this is a big fact we take away from this in addition to figuring how to improve these programs down the line and we have some thoughts about how to do that we should be getting these delegates voting more it's weird that we're only getting the delegates in these systems to vote 30 40 50% of the

time if we're going to delegate to you ideally you're voting all the time we would expect a legislator to vote essentially all the time and so maybe that's a problem that we're asking too much of these delegates maybe they need to get paid more there's a lot of questions about how to do this but getting delegate participation up seems like the lwh hanging fruit now that we've built uh these systems and so I will leave it there uh and happy to talk more afterwards thank you