From Whiteboard to Mainnet Podcast | Episode 4: DAO Governance

All right, welcome to From Whiteboard to mainet. This is a podcast brought to you by Columbia University Center for Digital Finance and Technologies and also the Ethereum Foundation's academic secretariat. Uh today we're going to be talking about decentralized autonomous organizations or Dows for short. Um and so I'm your co-host Fahad alongside Theo. Um and we have uh four awesome guests here today. uh we have three academics and uh one one person who's in industry now. Um so we're going to talk both about research and about the practical implications of it. So let me briefly uh introduce our guests at the top here. Uh so we have uh Junguk Khan from uh Soul National University. Uh Jungli also from Soul National University. Towi from the University of Florida. Um and then we have uh Michael Zaram who's the founder and CEO at block science um and research director at Megatav. Uh so uh what we want to do here uh really is we want to start by talking about some work that uh Jungub Junguk and Tao have uh on Dows and specifically looking at sort of looking at it from an economic perspective. uh and then we're going to transition into having a broader conversation about the implications uh and how to sort of properly design voting mechanisms in the context of Dows. Uh so let me just dive right in. Um and I think you know the first uh thing to really think about before we get into sort of uh thinking about any research is um to understand kind of uh at a high level how Dows differ from traditional firms because uh Tao Jong and Junguk are all professors of finance. So they've studied uh at length and written on at length um what we call corporate finance which is the analysis of traditional firms. Um and so when we start to think about Dows, we want to understand first and foremost kind of how they're different uh from traditional firms. Um and that also then informs exactly how uh researchers start uh start to do formal analysis about them. Uh so first of all, yeah, what are the primary differences uh between uh traditional firms and Dows and also how does that enter the framework of analysis? Um and so um you know uh Junguk, Jong or Tao, any of you do you want to take that one? — Yes. Um maybe I can explain u briefly about that. So if you think about corporations in ter sense, they have leadership such as CEOs and board members. They make decisions for the uh shareholders. uh but in that case uh inevitably there is a conflict of interest between uh the leadership of the company and then the owner of the firm shareholders. So we call it the principal agent problem. Um the leadership is making decisions on behalf of shareholders but then they might have their own uh own agenda. They have uh they may not have align incentives. So this is a critical problem and traditional organizations on the other hand DAO uh decentralized autonomous organization um they don't have central leadership at least in the pure form there's uh various degrees of DAO there's hybrid ones which actually has some human leadership but in pure form DA it doesn't have any central leadership it is um organization um based on um blockchain uh smart contracts on blockchains. So um the all the decisions are made by voting um according to the tokens. So um they it doesn't really require any human intervention. Therefore we don't have a principal agent problem. So that is the kind of key difference uh between DAO and the uh traditional organizations. Of course there are many other differences. For example, uh in case of DAO, uh the users uh and then uh the the voters they are actually the same uh body. In case of corporations, consumers uh who are using the services and goods and the shareholders uh these are different bodies. But in case of DAO, it could be actually the same body. Um but um uh as I mentioned the critical difference between DAO and corporation is the absence of principalization problem. — Yes. So I think um you know basically um there were there were also many other differences some of them are captured by our model as well. For instance, you know, if we talk about

traditional corporations, uh we often talk about cash flow rights, right, control, um you know, institutional investors, uh stuff like that. But here in terms of the Dow setting when we talk about governance tokens uh typically we uh value is heavily tied to platform usage or network effects uh rather than traditional cash flow claims which is you know a very um thin element in our model. Um yeah so and also I would say uh in terms of the legal uh background although that's not really captured in our model because our model is not focusing on that element but I think that's also very important difference is that corporations typically would rely on courts on fiduciary duties you know listing rules etc but here in the DA space and this at this point we have very thin um legal infrastructure in place and oftent times people are just you know relying on the smart contract which is very different from the traditional legal uh um contract. — Michael, do you want to add a little bit of context? — Yeah, I would say one thing I I've learned from working with and in a bunch of Dows is that um it's worth noting that the principal agent problems in practice don't totally go away. They show up elsewhere. Um the natural uh way this happens is that because you have a primarily a smart contract as a locus of coordination where there's some voting that uh body that ends up making voting decisions isn't really operational. It can't really do anything. And so it ends up having to whether it's field proposals or procure services or decide which members of the community are going to be empowered, compensated or allowed to do certain things. Ultimately, you end up having the organization itself as principal and some agent of that organization acting in fulfillment of some commitment that they made through some voting based interaction. And so you still get a principal agent relationship and you still get principal agent problems. Um it's worth noting that the main challenge in Dows as I've seen it is the lack of clear separation between operations and governance which is normally codified via the um shareholder basically owner shareholder kind of senior executive management kind of staff you know thin bottleneck there and so the Dows are releasing that bottleneck a little bit but that also um creates a cognitive strain because now you have the attention cost on governance decisions over a much larger set of things often a much more granular set of day-to-day operational activities and then you see failure modes around the failure to engage the lack of expertise or misalignment amongst the voting population if they are engaged and so I think it's worth noting that Dows as such are fulfilling the same function as firms as kind of sites of collective action or locai coordination they have a different architecture and that new architecture is more like a joint constraint satisfaction problem. We've got a lot of room to learn how to use computers to do that effectively. But I like the distinction that was made in terms of the architecture compared against that kind of traditional bottlenecking at senior management. — Thank you Zang. Yeah, that's important to add. Um yeah, I guess we can agree that are diverse uh mechanism that is used widely. uh but coming back to maybe the basis of the study I believe there's a certain type of DA that was uh um the basis I'm curious to learn more from the office about your primary insights so you've been building a theoretical model of D governance featuring token trading under token based voting um and then you investigate conflicts of interest so could you perhaps share what is the most uh notable insights you've um you've found out over the course of your investigation. — Um yes um maybe I can share some of those um insights about our model. Um this is actually related to the earlier question about what is DAO? Um and then I mentioned that DAO is organization without central leadership. uh therefore it is uh um uh absent of principal Asian conflict but as Michael mentioned um there does exist some could exist um uh principal agent problem because there's a human interface but uh theoretically um if there is no human interface if it's really pure form um you know we may not have any uh human intervention therefore uh no principal agent problem but in reality of course there could be uh some human part and then

that could easily create some problems. But let's just go back to um just um a theoretical idea that Dowo is purely smart based smart contract based organization. Um so in that case um now the decisions are all uh made by voting uh and then uh whoever uh kind of captures the voting uh could actually uh sway the outcome of all this decision making. So that's where uh this um ownership concentration of token uh comes into play. um often there are large token holders what which we call uh whales. So if there are whales uh who have substantially large amount of tokens the whale can capture the decision making and if the whale doesn't have a line incentive with other users the whale may um pursue their own private benefit at the cost of other users. So that is the uh the critical problem of DAO. Even though DAO may have less of uh the uh principal agent problem from the uh traditional corporations but DAO now has um other issues with this um conflict of interest between large holders and then small holders. Of course traditional corporations may have that but then the DAO against DAO this is actually critical problem. So uh what we study in our theoretical model is um uh how this kind of situations uh could be avoided uh given the trading mechanism in the market. So uh often uh we think about quadratic voting mechanism uh which can mitigate this problem. So quality voting mechanism uh is saying if someone wants to vote more it becomes more and more expensive as the uh the amount of voting increases. So it could increase quadratically. So um often this can resolve the issue of um large uh holders uh sway weighing all the outcomes. However the issue is um uh the civil attack. Uh so um one can make many addresses and then uh make votings uh dispersed across all these addresses and then in that case um the the voting cost doesn't increase quadratically because all the small uh addresses will have um rather low uh voting cost. So how can we actually uh go around this problem? Uh what we show in our model is actually the trading mechanism um could be the solution for this problem. Uh namely the token ili liquidity. So when you think about liquidity uh or liquidity um it is liquidity is usually the good thing. Uh the definition of liquidity when you think about it is easiness of buying and selling certain securities in this case token. So if buying and selling tokens are easy if any that should be good. However, uh in terms of governance actually um uh this could be an issue because if there are whales who don't have enough tokens to change the outcome but then close to it then whales can easily buy more tokens and then change the outcome and then they can um they can pass a proposal that may be harmful to other users but then may benefit uh themselves. So, so uh this could be an issue. Uh but then as I said the quadratic voting um may not work here in case they try to attack but then in case um this is done by let's say tokens um and which are traded on the market then the liquidity becomes like speed bump. uh when the um the whale wants to take uh buy more tokens as they buy more and more actually becomes very expensive it doesn't matter whether they have many addresses or not sure amount of tokens they are buying in the market will push up the price and then um making uh the voting very expensive for them effectively this becomes quitive voting. what we show theoretically in the model is that uh the token ill liquidity is effectively becoming quadruping and that which may resolve all this some of this governance issue. So that's one of our uh theor theoretical results and then we actually have other uh results about long-term commitment and so on but then maybe uh I I'll just you know uh start here and then listen to others. Yeah. So, so I want to unpack that a little bit and then maybe you can bring a few more people in. So, um you sort of talked about two things there. One was

quadratic voting and one was uh illquidity. And you mentioned that the liquidity could look like quadratic voting. And so, maybe but I want to put quadratic voting to the side for a moment and talk a little bit about the illquidity. Uh because uh I guess one way to think about what you're saying is that if I'm already a large holder, whale, I could buy a little bit more in order to become pivotable in a particular vote. And if it's if the market's very liquid, then that may be very easy for me to do. And that's going to cause the problems that you're describing. So I'm kind of wondering though then uh and this may be even a question for Michael because this maybe is a bit more in terms of like what we actually see in practice is um like do we see for instance people uh just buying directly ahead of uh votes um uh because in some sense you know the the holding that anybody has at any point in time at some point had to be purchased right um and so really it seems to me what matters is kind of like whether they're doing a significant purchase at a particular point in time, right? So, for instance, like if I'm always 40%, and then I just want to become I don't know, let's say it's a majority vote thing and I need 50% or whatever, right? Um and then so I just tend to buy whatever 11% or whatever um when I right before a vote or something like that, then then it's more then it's not about like amassing the full 51% position. It's about amassing just this incremental position. So, do we see this behavior in practice where large players will suddenly sort of top up their positions just for the sake of a vote? Um, I I'll just leave that for the entire room. I don't know if anybody wants to pick up on that. — There are some um cases of where that has happened. I wouldn't necessarily say it's a widespread practice, though appropriate for an empirical study. I've had some personal experiences where I've been involved in votes where that very clearly happened. One that's quite old now that comes to mind was an old Aragon vote which really looked like it was going in a particular direction and the whales dropped in at the very end and swung the vote um many years ago now. So hard to recall the specifics. It's an emotional memory funny enough. Um but I think the point here is probably that um if a vote is important enough, if it has enough bearing on the outcome for a particular whale or class of whales, you do see some of that behavior. I think it emerged probably most strongly in the um vote escrow kind of emissions routing process. There was another example that comes to mind where someone kind of totally captured a stream in the balancer protocol at one point. I think there was a basically a dead pool and someone managed to capture all that liquidity and then route a ton of rewards to it. And again, this is very much governance attack territory, at least in my opinion. Though, because it complied with all of the code in the smart contracts, there were sharp divisions about whether it was considered legitimate behavior at all. I think this kind of I'm going to call it gaming the metrics is always going to be prevalent. And you see this around other mechanisms like uh explicit vote auctions as well. So I think one property of Dows empirically is that there's a financialization of governance authority. It manifests in a variety of ways. But just because something's financialized doesn't mean those methods are always exercised because there's cognitive costs, attention costs. You actually have to really care for like minor day-to-day votes. the likelihood of this kind of financial activity uh in order to swing outcomes by whales is pretty minor. In fact, half the time they don't vote at all. If they are voting, then they have to have a reason to really care about the outcome to put extra again financial and operational labor into making trades and positioning themselves to improve the outcome. So I would say like broadly speaking um the binding constraint here is actually how much people care about the outcome. And so the thing that Dows often do is um have many votes. And so it's interesting to look at in which circumstances what kinds of activities major changes almost like constitutional level parameter changes are probably the only thing that are going to merit that kind of behavior. — Yes. Um yes f you know I um I basically want to uh second Michael's um points that oftent times those uh you know governance attacks don't happen but I think when they happen the damage could be very large for instance I guess you know some of the people in the room probably are aware that in 2022 a whale used the flash loans to buy tokens that worth about you know $1 billion to gain about twothirds of the voting stake in beantock and then uh you know this whale proposed a uh malicious proposal to drain about $180 million worth of cryptos from the asset pool of the platform and I don't think this is

an isolated case there were uh there were a couple other cases even before this one so although those events are very infrequent but when they happen the damage could be very large — so I Think there's a really important distinction that we should make too about the properties related to kind of resistance to attacks versus properties associated with governance as steer like governance mechanisms as literal steering. So you have constellation of good faith stakeholders with different interests who are kind of advancing positioning using the mechanisms as they were intended to advocate for support. They're you know whipping votes so to speak. they're communicating with each other to try to get people to support things or they're spending money in order to increase their influence in order to drive an outcome which is showing skin in the game etc etc. So like broadly normatively dis um separating the behaviors that are you know again attacked in the sense that someone wants to exploit the system pull money out in a way that is not pursuant to the purpose of the DAO versus like participation in the DAO and the size of social and economic mechanisms as they exist to advance the interest of the particular stakeholders at hand. And I think the reason for separating those is that from a mechanism design standpoint, I would author very different requirements and I would test against both sets. So the tests for the requirements associated with steering mechanisms, they look kind of like signal processing or market design problems. You're structuring like, okay, well, you don't know what people's opinions are. You shouldn't tell them what they should be. But if the mechanisms are exercised to express preferences, then the mechanism aggregates those preferences into a decision according to a process that's transparent, legitimate, codified outcome is deemed the decision of the group. That's effectively DAO as bureaucracy. However, the smart contract as implementation of the bureaucratic process. The other side of this though is that when you move to these formal kind of codebased systems is that you create room for gaming those mechanisms and exploiting them against that interest. And so you can use different techniques to discourage, prevent or otherwise um kind of reject things that are determined broadly to be illegitimate. And I I'll stop for the moment but I have some um other work actually academic paper related to a case study where we explored how like one dowo differentiated those two mechanisms. — So uh if I can just follow up the Michael on what you were saying because so okay let's say we're talking about the first piece which is steering. I think you were describing it as the steering case. Um it seems to me like uh one thing that academics particularly in finance are very familiar with going back to the corporate finance literature and so on is that uh in some sense the this the set of participants who happen to be shareholders to a firm or token holders to a DAO is not the whole universe of players and there are other players there's other welfare to be concerned considered and so the classical example here is something like um the person in charge of the firm has some private advantage from implementing a policy that is advantageous for that particular person but is actually socially not the optimal thing. Um and so then there's sort of like broader welfare implications that are not so good about that. Right? So for instance like if you take the space of voters as fixed and you say like okay what's the best outcome for all of them? That's a slightly different question than saying, okay, what is the best social outcome, which really and I think this gets a bit more into some of the work in the paper here, right? They're thinking very hard about things like, oh, people can actually buy tokens in this setting. And so in some sense, the set of players is not like some exogenous fixed set. We have to kind of think about the fact that there's this uh other margin. Um uh and so I just wanted you to reflect on because when you were talking about the steering mechanism, it sounded very much like if we're thinking about microeconomics, a game with a fixed set of players who have preferences. And so if you could just maybe illuminate that a bit because I think I misunderstood — necessarily mean a fixed set of players, but to be clear like you at any given time when there is a vote, there is a collection of votes and they represent the interest. So again, um you know, I'm a control systems engineer. My PhD is in distributed optimization. I'm comfortable with open systems and evolving sets of um members and games that are dynamic games, games that are in an open world setting. I'm actually not a very big fan of economic game theory. To me, it forked off of the kind of aerospace and defense variance of game theory in like the 50s and 60s and went very academic, whereas the, you know, the other stuff being arrow and defense went very dynamic. And to your point involves open environments and

interacting with comp complex change. Um, sorry, tangent. Um, point being though, I'm not assuming that this is a fixed set or like a canonical microeconomic game structure. I'm actually saying like there's information in the world about preferences and that those that information is incident on this system through these voting mechanisms and the to the point of splitting the problem apart. It's the same mechanism but you need both like effectiveness properties which are related to whether it channels the information incident on it into decisions that are you know considered effective and legitimate versus the more defensive side which is suppose an adversarial actor tries to hijack those same mechanisms to say drain a treasury. It's not a different mechanism it's different requirement. So you have to author formal requirements and write property type proofs about those things or at least demonstrate through simulation scenario tests both of the form you know use as intended essentially no adversarial actor just can it process the information incident on it into decision and does that make sense and when an actor who is adversarial shows up and tries to hijack that how resistant is it to that hijacking and that can include things like the cost of um achieving a certain amount of power. So 51% attack or um essentially um there's a variety of different techniques you can use, but ultimately you're writing properties that say, hey, how hard, how expensive, how much time, how much money does it take to uh induce uh an adversarial outcome. And that is something that will vary by the mechanism, but also by the liquidity, by the shape of the markets, by the distribution of the holdership. And so you don't necessarily get like an a priori result. You might get a functional result that tells you quote how hard it is given the current um you know token distribution and the current market prices and estimates of the price impact of trades. — Yeah. Let me add a little bit about the ongoing discussion as uh just for clarification in our paper we just try to highlight potential conflict of interest between two you know types of token holders right the big whale and then small users but you know we just want us start discussion start discussing about you know how their conflict of interest affects the you know Dow's resilience but in fact our discussion is more toward you know what's the optimal form of ownership distribution, right? So maybe just one big whale versus small, you know, users, you know, in that case, maybe the dictatorship or autocracy, you know, that's definitely the bad, you know, situation. But also at the same time, in a complete democracy like a small users, then they don't participate. Maybe they are too small. So they don't have any strong leadership to lead the DAO. So that case may not be ideal either. So I think that eventually you know there could be some optimal allocation of the power balance you know kind of power structure among you know several whales so that maybe some you know pluralism type you know situation three or four whales they actually check and balance between each other so that you know one whale cannot control the entire D system but this actually you know a couple of numbers of whales they are they have a strong participation incentives the voting mechanism and then you know discussions. So that may be you know in reality could be a better form of the DAO power balance or power structure but our paper for clarification it's not touching on this you know more advanced issue but we just want to highlight you know when a single whale actually can easily accumulate the voting power and they steer the D you know in a way to personally benefit not you know benefiting for the other users then what type of agency problem you know that could arise and how to mitigate such problem through the you know kind of mechanism design tide in the contracting you know solutions. So definitely I think that we need to actually make more deeper discussion about what could be the better power balance within the DO system you know but that kind of research is actually ongoing and then we may actually have to answer those questions. I think the one thing I'm trying to do though is highlight that you actually can't carve out um one problem when you're doing mechanism design in a particularly in a Dow setting where those mechanisms are effectively serving multiple functions or have multiple purposes and potential threats at the same time. So like when we talk about the category of let's call it requirement which is like resistant to exploit by a single actor with a whale position that is a requirement and it's one I totally respect and studying it deeply and understanding like how bad it can get and or what mechanisms or

approaches to mechanism design mitigated is super valuable but it's actually can't be decoupled from the other questions that I've been surfacing. So, as the industry voice, you know, Kio and I talk a little bit, I'm trying to highlight how this relates to practical design decisions. And so, the way that I'm framing it at least is that um when you look at a particular mechanism that's going to fill this slot, it needs to be resistant to several different kinds of attacks. It needs to be useful for at least one and often multiple kind of purposes. And again I'm highlighting the kind of big governance change type purposes which are like hey can we make major parameter changes that are viewed as the kind of the fundamentals to the Dow and then they're also used often for more micro more operational decisions which creates attention issues. So you have to worry about requirements around attention and interest relevance. Um the effectiveness of the changes visa v the DAO's uh you know ability to perform its purpose pay for things whatever it's doing varies by DAO and you'll also have to worry about the various dimensions of exploit attack conflict of interest etc. But since those are all about the same mechanism, you have to be able to say, "All right, here's a category of mechanisms. Here's the circumstances it's being deployed into. " And from each of these angles do the assessment. So it's a split from like a depth first assessment of a particular kind of um attack or I'll call it safety or resilience property that we want versus the breadth first which is for any given mechanism it has to satisfy that and and other things. If you go too deep on one thing, you end up with a mechanism that is like maybe really good against that, but it's um effectively overoptimized and it can't perform or isn't um resilient relative to the other requirements. And we talked briefly in the preamble about you know quadratic voting and I like I point out often that it's not a great use case for pseudonymous environments. It's hyper um specialized to reduce the effects of um having high amounts of money, but it sacrifices its effectiveness in a pseudonymous environment. And in fact in it kind of exacerbates the effect of social media influencers because social media influencers influence a large number of individuals and even if they are discreet individuals, it magnifies the effects and you see that in some of the Bitcoin stuff. And so the point I'm getting at here is that like any particular property assessment has to be placed alongside the other let's say three to five core properties and assessed in parallel. And so the mechanism design problem is making the best across all of those rather than necessarily being amazing at any one and like trade-offs always end up showing up. And so again that's not a particular criticism. I appreciated the paper. I just I was trying to kind of provide the how do you use this information in practice and the answer is you place it alongside of your other requirements. You do some trade-off studies and you do the best you can given the situation you're in. — Let me just maybe add some context before we transition on. Uh we might be getting a bit too much in the weeds of methodology here actually. Uh so okay, one thing to note though is that when you're thinking of an optimal mechanism, you do need kind of a wellposed objective. So I think economists uh tend to think about particular objectives like optimal welfare for instance and by the to be clear something like optimal welfare can capture multiple properties like for instance security is not irrelevant to optimal welfare considerations um uh and so Michael when you're talking about different properties I think a lot of times one of the difficulties like economists have when they talk across engineers is that sometimes maybe not okay I shouldn't put all engineers in one category but for instance I know in computer science there's there many times particular properties that you want your let's say stepping outside the DAO context let's say you want your consensus protocol to satisfy I wanted to satisfy safety livveness I may want to satisfy a bunch of other things right um but of course uh coming up with an optimal mechanism requires kind of a single metric and you could have a metric that for instance weights um you a very simple one would be like I uh I I have 10 properties and I just count the number of properties my mechanism satisfies and that's essentially how I value the mechanism. Now in that case I may have multiple mechanism of being optimal or something like that. Um but the point is that the idea of having different properties that you're interested in to satisfy um is not exclusive to the idea of having a single objective function because you can have an objective function that essentially collects these. they might have to wait it or do something like that. But uh and I think — it's a pretty basic thing, right? It comes to your point, it comes up in engineering, traditional engineering all the time, right? You have a bunch of

properties. You have things that are hard requirements that are essentially booleans like did you or did you not satisfy this? These are constraint set. Then you have objectives. But if you have a multiriteria objective problem, then you have to have an aggregator function. You can do like minmax raw style. You can do sums, you can do a your choice of aggregator. Then given that you can still apply weights. So I like things like geometric programs where they're multiplicative. Um like that's because you can do like softmax and stuff with that and they're still um uh like quasy convex. You can still solve them um optim like with computers. But the gist of it is that you ultimately need to construct something that produces a paro surface where although you have weights or parameters a way of selecting the relative importance of your many objectives once you actually start doing that you can reveal paro surfaces. You get a sense of what the optimal surface is and then your trade study is making decisions about like where on that paro surface you want to be. The thing that drives me crazy is when people pick things that aren't on the paro surface and I feel like there's a lot of that. — Okay. Well, I I think we're conceptually agreeing here. One of the values, by the way, of having a practitioner in on the conversation is precisely it can help the academics understand what is more practically important and what should be considered that maybe is not. But I think that's a longer conversation that we can have uh afterwards offline. Um uh Theo I think uh you wanted to bring in a related uh topic here right? Yeah, thanks actually yo um as you earlier mentioned um the whales I think it's worthwhile discussing delegation as well so many in practice use delegation mechanisms which I suppose they facto create other sort of sorts of whales who have voting power like other whales but actually do not have the control including the financial of trading these tokens. So to what degree do you think your um uh results from the theoretical model are transferable to Dowist practice that use delegation models? — So um you know the uh yeah so basically delegation does reduce the ability to trade a you know around the governance votes right. So basically um you know it is harder for the delegates to buy votes etc. But still I think some of the economic logic does apply. You know, some of the economical logic from our model still applies because uh you know, even without uh transferable voting power, uh you know, those dedicates can create a persistent concentration. And you know if the delegates have different expectations or different preferences about certain governance outcomes versus other users, they actually could still uh you know steer uh you know voting outcomes uh in a way that's beneficial to the delegates versus other users. So, you know, you would still have this misaligned incentives between those dedicated so-called dedicated whales um and users. yeah I mean you know um so um however I think uh there were other dimensions that are not available in our current model is that for denigates you could you know make things more transparent because you know a lot of those denicates do care about their reputation right so basically you could make the model involving delegation more transparent than what we have in our current setting um etc. So yeah — interesting is anyone want to add to that? Yeah, also like the delegation is actually one way we fix the participation problem, right? Because in our model, you know, there is no, you know, such thing like a directly modeling, you know, some kind of lack of attention or you know, lack of participation type issue we actually encounter in the real world Dows. So the delegation on behalf of the others actually the delegating party you know can actively participating in the voting processes. So that you know they can overcome this you know the real world problem. So but at the same time you know given the participation you know one delegate you know delegating party you know can have conflict of interest from the others maybe you know against some you know the vast majority of the users then the same problem like we you know describe in this paper like a big stake you know holder versus you know

many diffuse you know small holders you know or some other delegates you know who try to represent them actually there could be conflict of interest. So how to actually intervene and then try to actually restore the resilience of this Dow system you know against such you know power imbalanced game you know so that actually mimics and our model in that part actually mimics the real world situation I think but uh in re in reality you know there is actually more complex layer of problems including the participation issues and also the conflict of interest issue we describe in detail well in our paper. So I think that we need to actually you know learn step by step you know those you know vast majority variety of issues you know as Michael said there are so many different issues in the DO space so the one mechanism design cannot you know fix them all one size doesn't fit all you know so we may need to prioritize list all kinds of problems agency issues in that specific space and then try to order them in terms of the pro you know significance of each problem and then try to think about you know what's the multi-dimensional mechanism designed to fix those problem to a certain extent. We our paper is just a starting you know point of the one of the biggest you know agency issue maybe conflict of interest issue you know among token holders even in the absence of central leadership you know who might behave on behalf of himself or herself you know there could be some different types of agency problems in this dollar space which is actually acknowcknowledged by Michael in the beginning right there are still agency problem in this business so I think the our paper is kind of starting point to describe such issue but there are a lot more you know going on you know how to make them you know having strong participation incentives at the same how to coordinate them in the desirable way so I think that those are the important issues we need to solve in academically and also practically I guess — so I think practically the way I would imagine using your work is to use it to develop both requirements descriptions of those conflict problems so that they could become either constraints or objectives per our disc design discussion earlier. Right? So you can assess whether a particular design is likely to produce or the circumstances under which it's likely to produce that class of problem and you could also use it more for tools empirical tools for trying to measure or identify signatures for or predicates for those problems. that moves it away from being a mechanism design focus because I think the mechanism design problem itself ends up being um well frankly I tell most people not to try to design new voting mechanisms. It's kind of like crypto don't roll your own the cryptography. use a tried and tested mechanism for which you know the properties and for which the properties are appropriate for the situation. That doesn't mean no one should be researching and developing new mechanisms. Please continue to do that. But that like in practice using a new um a new governance mechanism is rarely a good idea. And that um the the key is that the research that is into governance mechanisms is often most fruitful in helping us describe the desirable properties of a design and to assess the current state of a living system. And that as those um things mature, it does end up leading to situations where those me new mechanisms do get tried out. we see how they work in practice and then we get a sense of when and where and how to use them. And um you know I realize that may be a little bit um unexciting but like there are reasonably good libraries of mechanisms. I think uh Our Shimone put together a great like library of mechanisms. I'm sure Tio can share a link. And that when it comes to the academic research around algorithms and mechanisms, the first class learnings that make their way into industry for you know seasoned professionals are how to write down the requirements. So you detail certain kinds of attacks, failures, you write proofs and do analysis of those cases. And then I write down in my mechanism design work. All right, here's another thing I want to check for in the list of many things I want to check for. And if I'm doing an empirical study of a of an actual DAO in life, then I'm looking for um whether or not it appears to be an issue. Again, deriving metrics from that research as well as in a more advanced case looking for predicates. And I think predicates are important because you're looking for the prior states that represent the risk of or the potential for that failure mode because most of the time once there's a failure it's too late. And so understanding how to detect predicates and intervene in advance is a major element of kind of organizational or institutional resilience and something that I think still need to develop.

Yeah, I was actually regarding that I want to add just one thing because maybe Jangok you know can add more. We were surprised when you actually found in our paper such that the targeted illi liquidity like a simple locking period for the important whales you know in the Dow system they turned out to be quite powerful you know governance device — so maybe we don't have to think about the complex solutions maybe we can just simply say hey you have a large stakes then you should show the you know interest in this D system over the long run then just signal your you know good intention through the long lock in period and when I talk about this lock in periods as a solution even in reality in the venture capital space you know that's an outside doubt but you know when there is a significant information asymmetry problem and potential you know moral hazard and adverse action that simple rule and that simple solution actually restore the trust you know between you know many different parties with the potential conflict of interest and importantly in our paper we showed that such idea like a lock in period as a targeted ili liquidity for certain groups of whales that could be a powerful you know device to solve this conflict of interest problem. So maybe expanding such idea simple yet you know powerful economics you know and apply them to the real world da you know maybe could be a way you know that the academic findings and the practical solutions can meet together at certain point. So maybe Junguk you could add because you know we thought about this issue for a long time right. — Yeah. I was about to actually um make a comment about that and thanks for inviting me to — give additional comments. So um what we're doing is not so much about how it should be. It's more about how it is. So uh we are not really designing any mechanism there. What we are doing is um we're just trying to explain uh how it is now and then it may actually work as it is. So um uh at the moment if uh you know many Dows are adopting uh tokenbased system and then uh those tokens are traded on the market. So we're not saying they have to design this in their way. But then if those tokens are trade on the market uh it just happens that those ili liquidity is on their side especially when the uh governance is contested where uh whales are um trying to capture the um uh the outcome of the uh the protocol. So uh so in that case indeed we empirically find that when the governance contested um the token liquidity uh is actually boosting uh the the growth of the platform. Uh so we actually verify that uh theoretically and empirically and um as Jung just mentioned um this um uh lock in period uh of tokens um this is actually um doing this kind of same mechanism as the uh token illity in a way because those whales who are uh locked in uh they it is as if having sort of a greater ili liquidity only designed for them. So um uh what what's happening there is um they are um they are sort of their incentive is more aligned with the users because of their long-term commitment in that case and then this is what we find in the model and then indeed in our empirical result that we also find the result that uh imposing more long-term commitment uh in the form of lock in and other um other forms uh actually boost the platform growth. Um so yeah uh in a way u kind of this is um what we find both theoretically and empirically and we're explaining how it is rather than where you know where it should be like that. So maybe there should be that better mechanism but then at least that's what we find. Yeah. — So my question is whether or not you participated in any of that you studied empirically. So like just essentially autoethnographically like balancing out because there's in my perspective you've got like theoretical kind of mathematical level analysis. You have empirical at the level of data collection and then I always triangulate with um kind of ethnographic methods. I work with researchers at RMIT in Australia Ellie Renie Kelsey Nabin and Ellie's other students um and I found that without that triangulation it's very hard to like really make sense of one of these sociotechnical environments. And so um the question is really like did anyone participate in any of the Dows that you studied and how does your like experience in participation line up as the third leg of that stool with the uh mathematical

and empirical results that you found? — Actually Zag maybe let me give this a question back to you because this is a really good one and it lines well with something I meant to bring up. So the data collection for the study ended in 2022 right? and the Dow space is evolving fast. So maybe Zum could you uh briefly describe some major changes you've observing over the course of that period because of course um yeah space has changed a lot and auto delegation has evolved and the regulator landscapes as well. Yeah, I mean the thing that I've noticed most recently that is like changing things is that everyone wants to involve LLMs and I get actually there was a little bit of discussion of that in the paper I linked um which we can probably share in the notes around detention economies. this is related to this problem of who makes decisions, but I yeah, I mean, there's probably nothing more impactful in my lived experience since we're kind of bridging off of that than the sudden like desire for people to kind of assess and participate either individually through the use of just like talking to chat GPT or Claude or Gemini or whatever people's preferred models are to the development and deployment of systems that are kind of rag based and or have access to let's say forums andor you know pulling onchain state that's a little bit more advanced I don't see as much of that but like the LLM craze has like really captured the imagination of Dow participants and I see it particularly in the face of the binding attention constraint the cognitive constraint and then there's also the fact that people are increasingly building tools that actually can directly interact with smart contracts and for folks who are interested in principal agent problems I actually went on a different podcast in AI I one I don't know some months ago and talked a bit about principal agent problems in AI systems where you really have to think about whether the AI agent is in fact your agent or someone else's in terms of like whose interest it's advancing and although that wasn't a cryptofocused conversation it was informed by my experiences in Dows and with crypto um community members you know both using tools that had been provided for them in service of um their financial or governance interests as well as the observation that the exploitive actors are increasingly automated actors who are programmed to just hunt for opportunities to arbitrage. I think that's actually an interesting factor to your paper about like these kinds of misalignments and exploits because at present there's nothing really to stop or to constrain a AI based agent from kind of entering a DAO governance environment and exercising automated market makers and then voting power to say steal funds and you know that's a little bit hazy but I would say there was a recent study I saw apologies I don't have it prepared to quote about the increase in um the um this kind of exploit on an automated basis. Um I'll see if I can find it and share it in case you want to add it to the notes. It's actually interesting one of my student who just joined the you know Colombia IUR department actually he wrote paper with the Augustina Capony about you know what if the agentic AI participates in the Dow voting process you know how does the AI in a asentic AI you know how does it do compared to the human you know based voting and I think for sure there will be the growing participation by you know Asian authentic AIs in this voting process. So there could be another you know conflict of interest between human controller versus agentic AI or even across agentic AIs. So I I fully agree with you know Michael's words actually this you know Dow space is getting more complicated you know by merging it with the you know AI you know the based world so you know how to restore the resilience of this system you know what might be the you know everlasting governance mechanisms I think that will become more and more important I guess in this you know complex world yeah — as we are coming at time I think this is a really great way of of signaling to both builders and researchers that there's a lot more to build to experiment on around our governance and surprisingly little best practices perhaps but also to researchers to take a much closer look also interdisiplinary research including ethnographic studies as Daram said to make sense of this very exciting field of darkness that is still emerging. So I would like to thank all

these authors Yumi Yamapani and Michael Zorgam for joining. Um thank you Farad for hosting this session and um please share the link on YouTube with everybody who may be interested and thank you everybody for joining today. — Thank you. — Thanks for your work guys. Okay, the public stream has stop.