Shimony Metagov 20230823
Metagovernance Seminar Archive | 2023-08-23 | Unknown
Speaker 1: Great. So Thank you, Orry. What, like, one definition I like for what, mechanism design is is it's the design of games that exhibit equilibrium behavior. And this definition comes out of the formal mechanism design space, which is kind of an evolution of game theory. So very much from an economics perspective, but also considering how political and governance decisions play into the...
Top Keywords
- mechanisms 0.020
- design 0.009
- mechanism 0.009
- crypto 0.006
- design space 0.005
- library 0.005
- different 0.005
- space 0.005
- game 0.004
- governance 0.004
- value 0.004
- token 0.004
Transcript
Speaker 1
0:00 – 0:00
Great. So Thank you, Orry. What, like, one definition I like for what, mechanism design is is it's the design of games that exhibit equilibrium behavior. And this definition comes out of the formal mechanism design space, which is kind of an evolution of game theory. So very much from an economics perspective, but also considering how political and governance decisions play into the game theory of of decision making and coordinating independent agents or agents that have the ability to freely enter, exit the system, collaborate, collude, etcetera. Now what got me into the the the crypto space is kind of this excitement that we have this new terrain upon which to design such mechanisms. So no longer is mechanism design the domain of just governments or massive corporations who have large platforms, but basically, all of us can notice a problem and potentially design a solution and deploy that and then and then get people to use it. And so that that's one difference is the kind of, accessibility that blockchain creates around designing mechanisms that can actually affect real people. The other kind of key difference is the voluntary nature of blockchain based games. So any sort of governance economic coordination system that you design through smart contracts doesn't have an ability to force players to to play the game. And so it's basically a total total open field of the the good mechanisms can proliferate ideally, and the bad ones can die away. And we can have this evolutionary process of discovering the best sorts of mechanisms for improving coordination. Now I wish that was actually the case, but as we've seen in the ecosystem, it's kind of, you know, two steps forward, 10 steps back. Because since it's such an open playing field, there isn't always the best incentives, to ironically move forward the space of building incentive aligned systems. So although we have this, like, powerful new platform or set of platforms for designing better coordination systems, we're kinda stuck in this back and forth where, you know, very few people understand all the different disciplines involved or are brought together with people who unders understand, respective disciplines that might be important, to understanding how humans behave and how economic incentives works and political processes. Also, people like to basically do what everyone else has done. This is the point down below here. So it's kind of easy if you're really, like, have a real, problem that you're trying to solve. It's safest to just do what's popular. And
Speaker 2
0:15 – 0:15
kind
Speaker 1
0:30 – 0:30
of what I've seen over the years is even though someone knows it's a suboptimal solution, it doesn't really make sense for any one market actor unless they're really massive to change or to take an take a risk or experiment with a mechanism that might achieve a better outcome. As well, the other kinda big issue I've been realizing is that a lot of these exciting experiments that have gone on over the years in the crypto ecosystem are sort of lost to history or they're, you know, in people's memories or only people who are involved know about the learnings that came out of that, what made the difference that made something fail or succeed. And so we don't have a robust understanding of the design space and also what what's worked in the design space, what hasn't worked. And, yeah, just the other point I want to make is that we have some groups that are more industry focused or they're participating in markets. And, of course, they are more constrained by by their bottom line, by their perhaps token price or business model. And so, of course, they they can't afford to be long term focused and see the bigger picture. Whereas, those in academia who might have more rigorous approach and a more long term focus, sometimes there's a gap between those practitioners and those that are building products that, you know, they want millions of people to use. Right. So this is the kind of reason that we got the mechanism institute together this year. The idea is to really advance these new building blocks that we just barely understand as an ecosystem and really take them further. So one is understanding what's already happened. Like I said, what works, what doesn't. But then it's understanding the underlying principles of why do certain mechanisms work and why do certain mechanisms fail and what variations thereof can lead to better outcomes. And through that process, we can actually design new mechanisms, and we can make new matches between mechanisms, which are solutions and different kinds of problems. And we think that the best way to do that is by working with a really cross disciplinary group that is both in academia and also in industry, has experience, you know, studying these kinds of things in, yeah, in a rigorous way and also in using them and in seeing the kind of messy ways that things actually run-in production in uncontrolled environments. Right. So the first project we took on was to do this design space mapping exercise. And so this is still ongoing. There's still definitely lots of mechanisms we've missed and lots and lots more use cases that we haven't properly documented. And so the first step was kind of how do you, break down the different types of mechanisms? And so this is the current set of six that we have, categorized, but there's lots of different ways you could slice it. And that in itself is a whole area that we're we're going further on with the research. But, yeah, I'll I'll quickly just run through a couple examples so you just get a a sense of the breadth of this design space. So on one side, we have mechanisms that let you transfer value between parties. Things like airdrops or earning rewards through a a block reward system with a fixed amount or an amount that varies according to an algorithm that multiple parties are competing to win. We have bounties, things like harbor taxes. Yeah. And we can go back later if anyone wants to dive deeper on one of these, but I just wanna give you kind of the broad swath for now. Right. Then we have fundraising mechanisms. So there's been a lot of interesting experimentation here even though the space has kind of gotten way more conservative largely due to regulatory concerns in the way that new systems bootstrap themselves into existence. And I think there's a lot more that hasn't been invented on this on this item. A lot more we can learn from, you know, other other spaces, you know, for the pre crypto universe. But, you know, we we've also come up with some novel ideas in the crypto world. For example, you have continuous auctions out of the Nouns project, which auctions off an NFT every day. And this is kind of some blend between a revenue a mechanism for, selling a product for revenue and also continuously fundraising. And then those NFTs that get sold also are the governance token of the DAO. And this kind of slow, you know, one new member every day is an interesting kind of membership, value transfer, and fundraising mechanism. And we all are familiar with kind of ICO and the crowd sale era. But then we have, you know, bonding curve tokens, which have been explored in different niche use cases, got popularized, you know, more recently through AMMs, through Uniswap. But then there's this concept of a bonding curve token with token whose whose supply is actually algorithmically determining its price. You know, Vitalik has written about, you know, Daiko mechanism. And a lot of these, we've seen, you know, maybe a few implementations, but, yeah, there's just a lot more, I think, that in controlled settings we could we could learn about. Yeah. This is what we're calling signaling. So, we could also call an identity perhaps. It has a lot of overlap. And so these are mechanisms that aren't necessarily economic in nature, and they're not necessarily governance. Right? So they're they're about, signaling some, information in a validatable way. And this is unopinionated as to, you know, how good these different mechanisms are or what they're trying to achieve. But this this is a few of the examples. And governance, I think, is the most familiar to us. There's a really wide design space of really interesting governance mechanisms inside and outside of crypto, you know, very few of which have actually been, used successfully in practice inside of smart contracts. And so this can include different ways to calculate people's influence over a system, different ways to then, trigger thresholds and types of actions that happen as a result of governance. Another an interesting overlap, I think, is we used to have a category called budgeting, but we actually took it out because, really, you know, we see I see budgeting as the overlap of a governance and a value transfer primitive. So this is where we we're just trying to build more of a pattern language where these higher order patterns can emerge from the recombination of lower level, more fundamental patterns. Yeah. So here we have kind of market design primitives. So different ways to design, spaces where buyers and sellers can interact. And this also includes, like, the development of of components of primitives that enable different kinds of market mechanisms to emerge. And here we have, like, a, yeah, a sample of value capture mechanisms. So if a system is is accruing new value, creating new value in some way, how is that value returned back to to those who have been contributing to it? Yeah. One, I could point out that's an interesting one, maybe lesser known as contract secured revenue, where, contracts receive rewards from a protocol, from an l one or an l two, let's say, in proportion to the usage that that contract is getting. So you can start to see how there's overlap in in value capture, value transfer, and higher order emergencies like, let's say, public goods funding that could come from these lower level patterns. Yeah. So moving on from the library, this is kind of the the main piece of open resource that we have today, but we have a lot more plans for where this is going. So on top of the library, we already started building an LLM assistant. So basically a mechanism design bot where we feed it an embedding of the library and the case studies that are documented within the library as well as a lot of engine like, prompt engineering around that with principles of how you can design good mechanisms that can make the bot sort of figure out, like, a a good v one to get someone going in the right direction if they wanna use the best mechanisms for the the problem at hand. Yeah. Like, I was saying earlier, we're also looking at this as a problem of formalizing a design space that's quite new and novel. So, yeah, we're still on the early kinda days of using smart contracts to solve coordination problems. And so we think that other than kind of coming up with a few categories and a pattern library, there's there's basically more intensive materials that we can create around guiding the process of designing mechanism solutions as well as model ways to model the impacts of mechanisms I can show. And I'll I'll show a bunch of examples of this in a minute. Yeah. And then the the other main focus is, yeah, actual applied research. And so the idea here is to take mechanisms that have promising consequences and to actually put them to the test. So to take, let's say, Harberger tax, which you could use to allocate intellectual property rights in a way that might be more less monopolistic. And we can actually test these in controlled settings and run them with different kinds of communities and have insights on what's the best parameters to use, what what features should be turned on or turned off for these different kinds of use cases. Another kind of way to surface new approaches and new mechanisms and also new combinations of mechanisms is through design competition. So working with people who are already experimenting in the space and coming up with, like, formal design challenges around solving a particular set of problems and mechanisms. And then working with crypto protocols and then peep groups outside of crypto to actually do focused research in solving some sort of incentive or co governance problem, etcetera. And we're looking to start with the crypto space. Right? Because there is a lot of just mechanisms design space inside of crypto. Like, crypto protocols have lots of issues with giving grants or governing themselves or managing identity. Right? So there's definitely plenty to do within the crypto space. But the the bigger idea is also to take the mechanisms that the crypto space has been spending so much time and energy developing and starting to make those bridges outside of crypto to see where can we run pilots, which partners can we bring together to actually test these out, in more real world settings in a more consistent way. So this can be areas related to human flourishing, you know, like, energy and food security, wealth distribution, human rights, as well as existential risk kind of topics like nuclear proliferation or AI alignment. Right. So I can now walk through a couple more specifics, examples from the kinda outline I just gave. So, yeah, one one we started already is formalizing the way that the economic or mechanism design process happens. So this is inspired by the mechan the mechanics dynamics, as aesthetics framework from the game design space. So this is a highly cited paper, for developing basically video games and different kinds of board games and and things like this. And so it the idea is that we can adapt this into cryptoeconomic games. And there's slight tweaks we need to make, but overall, we can we can recover a lot of insights from this. For example, it talks about how the game designer really just has control over the low level mechanics of a game. But the at runtime, there's kind of these second order emergent dynamics that emerge that are not directly in the control of the designer. And then the byproduct of that is the actual experience that the player has, right? So the designer and the player stand on like opposite ends of this pipeline. And we can now update this model to think more about crypto. And what we're designing are not so much games that people play to have fun or to learn something. It's games that where the players are actors that have self interest, and they're trying to maybe they have a coordination failure, and there's some higher order impact that participating in this game could could have for them. So at the narrow level, there is here. Here's a canvas version of this. But at the narrow level, there's the actual impact that the the game dynamics have on the different players, based on their motivations and what they're trying to get out of playing. And then there's also externalities. Maybe there's a negative externality of a certain game, or or the game actually improves, some like, has a positive byproduct. So this is kind of an adaptation of the, business model canvas, or the platform ecosystem design canvas that integrates some of these insights. So how you could take these different mechanisms and kind of chart out the dynamics that you anticipate and ultimate impact and externalities that would happen. And you can also then apply this into, like, a methodology where you you do iteratively design you design games, you test them, you have an experimental setup, and then you can validate your hypothesis. Yeah. So this is an example of a model that we recently built with Gitcoin about looking this is just a tiny screenshot from a really long page that we're gonna make public in a few weeks. But, basically, it's, like, looking at different mechanisms and how you can quantitatively model their impact on on different aspects of the system. So, yeah, here you have, like, economic, mechanisms that you can choose between, and then the model updates. And you have, yeah, various, like, levers you can pull, to change your assumptions. Right. So this is the bot that I mentioned. So this is just a preview. So let's say you wanna solve a particular coordination problem, and then the bot can figure out for you what mechanisms might be the best. And then you can kind of query it more and find out case studies, what's what's worked, what hasn't, with this sort of combination of mechanisms. Here's the actual current prototype, that we have. So, it still needs a lot of work, but, yeah, this is an example asking it to build, like, an educational platform with peer to peer mentorship. So it looks in the library or it gets the from the embedding, the top results that are the most semantically relevant to the query, to the prompt. And it might suggest this combination of those mechanisms. And so we have, like, a template that we feed the bot with informed from the MDI framework I showed earlier. So the bot is using both the the framework and then the library to come up with an intelligent solution. Right. So this is just a I just pulled this together really quickly. Just a couple of the mechanisms that I'm most excited about that we wanna dive deeper into. So one is algorithmically constrained governance mechanisms. So the idea that, like, the the rights holders in a system shouldn't necessarily have full power over the budget and and protocol upgrades, but you could you could build more of these algorithmically constrained and also algorithmically determined, outputs from inputs such as token weighted curation, fundraising mechanisms that have investor protections baked in. Most of us are probably familiar with RAGEQUIT. There's also DAICOS and, yeah, just applying and redesigning some of those to to work for a broader set of use cases. Yeah. There's a few here related to value capture and identity. We we built a, like, a probabilistic, civil detection game also, in partnership with Gitcoin, to look at how to reduce the incidence of civil for passport. And the harbor tax Harberger tax example that I gave earlier as well as the AI example I gave earlier. And I can stop there, and we can start to turn it into a discussion. So, yeah, I can also pull up more tabs, show you show you more of what we've been working on, but I was just mainly curious to hear, yeah, how how everyone reacts. Thanks.
Speaker 4
0:45 – 0:45
Yes. Thank you for that for the presentation. Really exciting to see both the library all put together in some of the public resources as well as the specific research and experimentation y'all are focused on. So I'm gonna jump to some of the questions that I see that came up in chat, and we'll call up folks to to get up a chance to give voice to what they were thinking. For those who dropped comments, please just virtually raise your hand so I know that you do actually wanna come up and and comment beyond what you dropped in chat. Otherwise, I think Val's was the first, question that came up. So Val, are you able to or do you wanna hop on voice and, speak to some of your question there?
Speaker 5
1:00 – 1:00
Yeah. Sure. Yeah. Thanks so much for this presentation. It was awesome, and I learned a lot. And I'm curious to dive deeper. I was just curious, like, the the team's kind of criteria for inclusion of a mechanism into the database. Like, does did it have to have, you know, a preestablished does it have to be kind of preestablished or pre named? Did it have to have, like, a certain number of users that had implemented it, tried it out, and, like, perhaps some evidence and, like, you know, results of, like, kind of how it worked. If you could just talk a little bit more about that, that'd be really interesting.
Speaker 1
1:15 – 1:15
Definitely. So right now is what you see is the first version of the library, and so it's very much trying to nail that balance between two highly abstracted and two low level. So some criteria I mean, were about, like yeah. Like you said, if it has a name, like, if it is a thing, right, that can be discreetly described. Because a lot of these, there's a blurry line between if it's a piece of another mechanism, if it's a mechanism itself. Yeah. So so the short answer is that it's just a first pass, and it I guess I can talk a little bit more about the process. It's basically going through resources that that I've collected over the years by by working on actual mechanism design projects. Yeah. Going through blog posts, and there's also smart, yeah, libraries like OpenZeppelin and, I'm forgetting the other one, cookbook.dev where you actually have implementations. And then also things like ETH research, forum posts. Yeah. So so a lot of them are not even aware of an implementation. So they're just theoretical. Some, for example, like prediction markets have been talked about for years. There's plenty of discussion about them. There's some early implementations. But but for example, for something like Futarchy, which is decision making by prediction market, I'm not aware of any actual implementations, but that's definitely a mechanism, that's been theorized about. So it it's a really combination right now. It's I think it leans towards, yes, adding it. We wanna just kinda capture as much as possible for the time being. But the next step is to, one, refine the, the categorization, like I was saying, but then also to do actual expansion of the of the discussion section and the description for each mechanism and the case studies. So, yeah, that's, you know, where we're looking for research collaborators to basically for each mechanism to really break down, you know, where it works well, where it doesn't work well, what problems it's been applied to successfully, what parameters are important to consider when using it, and as well as, like, the the links to the implementations, like, actually describing how each example case study uses that
Speaker 4
1:30 – 1:30
mechanism. K. Thank you. Next up, we're gonna jump to, Amir who has his hand up, then we'll jump to Abe's question, in chat, and then I have one. And if anyone else wants to come in with more questions, either please just raise your hand or drop them in chat. And I see Seth came in, and I'll keep calling on people from there. But, yeah, Amir, please feel free to jump in.
Speaker 3
1:45 – 1:45
Hello, everyone. So there's a lot to unpack here from gamification to there's a lot to unpack. I just want to ask one question. I'm working on online voting, and I'm trying to create a a business model. So from the speaker, it he reminded me of the idea that people, they can propose ideas to vote and whoever gets more approvals or more votes gets a fee because we're talking about programmable money. I I'm just wondering if the speaker has ever thought about anything like this marketplace for ideas about online voting in specific. Because you guys should do it governance.
Speaker 1
2:00 – 2:00
Yeah. I I can give you some examples that I know of. Like, basically, I don't know if this is exactly what you're talking about, but, like, joke DAO. Basically, there's incentives to come up with an idea, whether it's, like, in the in the prototypical joke DAO, it's it's coming up with a joke, and you basically get paid in tokens if your joke wins the joke competition. And you can set ahead of time first place, second place, third place prizes. I know that four years ago, DAO stack implemented the same thing. They called it the auction scheme or the the competition scheme. So it was a plug in in DAO stack, which let you predefine the the length of the competition, the the first, second, third place prize. And more generally, I think maybe govern incentivized governance is kind of interesting to think about there. So, yeah, incentives for for voting, maybe just incentives for voting on the winning idea. Of course, that can have perverse side effects, but, yeah, curious if that is what you're thinking about.
Speaker 3
2:15 – 2:15
Yeah. The best percentage is what I was thinking, and I I'm just I was wonder worried about the the perverse the the bad side effects it could have.
Speaker 1
2:30 – 2:30
Yeah. Yeah. I mean, if you're talking about incentives for voting, there's very clear perverse side effects if that's not done properly. But, yeah, I I I definitely like the idea of competitions where there's multiple winning outcomes. Another another idea that I've talked to someone about lately, but I haven't seen it implemented is kind of more well defined markets around proposals. So today, you know, you there's maybe a request for proposals, like an RFP, or someone makes up a single proposal to get paid to to build something. Yeah, and basically designing markets around predicting and and options, for providers for different services.
Speaker 4
2:45 – 2:45
Great. Thank you. Next up, we had Abe with a question in chat. Are you able to to come in and voice your question, or do you want me to read it out for you? Give a moment. Sorry. Yeah. So I will read it out. So can I just ask the main motivation of the institute again? It looks like, mainly educational at the current stage. Thanks.
Speaker 1
3:00 – 3:00
Yeah. Definitely. So, definitely with the current resources that we have out there, which is just the library, it's purely educational. But the larger goal is to actually push forth new mechanisms. So I would say research is what I would tag that under. So, designing mechanisms ourselves, working people that have working with, new mechanisms or theorized mechanisms that haven't been successfully used in practice, and then actually running experiments to get them out into pilots. So, yeah, to me, that dovetails nicely with education. Like, we want people to know what's been tried so far, what's worked, what hasn't, what what are the kinda contours of the current design space. But, also, it's very early. It's it's like doing web design in the nineties. Right? Like, there's definitely a lot to learn from what people have done, but we can probably design a lot of best practices, new principles, tools that aid in the process of doing that kind of design. So we're I would say over time, gonna be progressively more focused on that side of things.
Speaker 4
3:15 – 3:15
Great. Daniel, do you wanna jump in next with, your question?
Speaker 6
3:30 – 3:30
Yeah. I would love to. Thank you, Ori. Super interesting presentation. My question, which I think is maybe a bit related to Vals, in your opinion, you know, do you think that for most projects, they'd be better served taking from the library of, I guess, you know I've heard I heard the term once ludic elements, which I really got a kick out of as, like, the as, like, the unit of a game, a ludic element. Do you think that for most people, you know, the bot you described is kind of like a, you know, let's take off the shelf and do a mix and match of our existing library versus, like, we really need to develop something kinda greenfield or we need to do something that that's really new. What do you can you comment on? Do you think, like, the elements that we have are sufficient? You just need to use them the right way, or do you think that we really need to develop brand new things that are radically different or some combination of the two?
Speaker 1
3:45 – 3:45
Definitely some combination, but I would lean towards developing something new. Right? So I guess, let me break that down. So, for a lot of for the long tail of use cases, a combination of some existing mechanisms probably is gonna get them 80% of the way there. But let's say to get all the way there or for way more complex or large scale than ever before tried use cases, they probably need to, like, start from a, you know, a blank canvas. I always think it's useful to to look at the library or to, like, talk to the bot and, like, you know, understand what's been tried, what's worked, what hasn't. Because then you know, okay. I can draw inside of the lines, but I know where the line is. So I know when I can, you know, go outside of that. So I always think it's good to know this the current state of the art even though it's, you know, gonna be, hopefully, like, 500 times, more, like like, ahead of where it's at now in a few years.
Speaker 4
4:00 – 4:00
Great. Next up, we have Seth, then Francisco, then Louis, then Steve.
Speaker 7
4:15 – 4:15
Honey. So, I mean, my question speaks for itself, but, I'm always interested in the complementarities between on chain and off chain. So I'm wondering, I didn't see any, like, you know, Web two tools, in the toolkit. But I wouldn't be surprised if it's in there somewhere, either as a tool or maybe in the philosophy. Are you just sort of agnostic to it? Is it there somewhere? Is the idea that you get that stuff for free in whatever kind of platform people are using? No. Yeah. Is it in the philosophy? Is it in the values? Are you encouraging it? Are you suggesting it? Are you suggesting it? Are you ignoring it? Do you hope to just, like, make it go away once there's a complete enough accounting of of mechanisms?
Speaker 1
4:30 – 4:30
Yeah. I I would say that we're I would say on chain maximalist is pretty strong term. But what we're trying to map out and build a science around is self executing agreements. So the different variations and flavours of self executing agreements, because I think that's the area that's very under explored and that has so much potential. Now, of course, there's immense value to, you know, the numerous, off chain, web two, whatever, TRAD mechanisms. But there's also plenty of research and plenty of, experimentation around those. Right? So in terms of filling a niche of of what we think is missing, I think it's really rigorously understanding the design space for self executing mechanisms. Now maybe there's a blind spot there because there's some interesting bridges or, like, hybrid mechanisms. But, yeah, for now, it's starting from the extreme of only looking at fully on chain mechanisms. And then, yeah, over time, I think, you know, we'll we'll see how those compose with off chain access. And even in the methodology that I I mentioned, it that considers what are the off chain dynamics. Right? Like, what are the institutional and institutional arrangements already in this ecosystem that this mechanism is gonna serve as an intervention inside of? So definitely considering the the off chain dynamics, but studying more formally what's going on within the smart contracts.
Speaker 7
4:45 – 4:45
Just to just follow-up real briefly. Could you give an example of one of the hybrids that kinda stands out in your mind?
Speaker 1
5:00 – 5:00
Yeah. In terms of current industry practice, I would say the snapshot safe is a pretty classic hybrid where, the execution the the the signaling is done through signatures that are, you know, private public key pairs. And that that that results in a signal that then the safe signers have a social obligation to execute on chain. And so, technically, it's not autonomous execution of some set of token holders or whatever. But, functionally, it gets it gets most of the way there in a lot of cases.
Speaker 7
5:15 – 5:15
Thanks a lot.
Speaker 4
5:30 – 5:30
Yeah. Thank you. So next up, we have Francisco, if you're able to hop on and ask the question.
Speaker 8
5:45 – 5:45
Sure. Thank you. Hey, Ari. I was thinking about the synergy of all these mechanisms. Because when I was looking at your presentation, and now I'm looking at the web page, I see the different mechanisms. And I see a use use case where you try to understand the different mechanisms that protocols have. But for me, a very interesting thing would be to understand how they all behave together. Like, what is this organizational sorry. The mechanism soup looks like. Because sometimes you change one thing, but you change the whole formula.
Speaker 2
6:00 – 6:00
So do
Speaker 8
6:15 – 6:15
you think there's a way to understand how these different mechanisms behave together? Is there a way to simulate that with AI or something like that?
Speaker 1
6:30 – 6:30
Right. I I think you're making a good point, which is that these are complex systems, so we can't know ahead of time when if you remember, like, the MDI, if we're just on the mechanics, we can't know ahead of time the second order dynamics that result from that and the third order impact. So I would say we can and we we we can look at, like, case studies and understand the the actual mechanisms themselves to get a good sense for, you know, what could happen in isolation given a certain mechanism. But, yeah, definitely, when you start combining them, they're complex systems. And as we see in, like, the world of DeFi, even in that kind of, high risk, they they haven't figured out how tweaking some sort of, like, you know, the way that the Vyper is implemented could, you know, introduce a reentry error into a smart contract. So I think we're in similar territory where, yeah, we we don't know. And that and that's why I think the methodology needs to be very experiment focused and, you know, running simulations, like, with AI agents. That's one part of it. And then also taking it to, controlled settings, test nets, capped, mechanisms where may maybe the mechanism has a smart contract enforced cap on how big it can get because we're not we don't have high certainty in the safety of it. You know, maybe if Lunaterra had that, you know, the ecosystem would look a lot different today. So taking unproven mechanisms and baking in programmatic constraints on them and also setting up controlled experimental setting setups where we can learn some of what you're asking about.
Speaker 4
6:45 – 6:45
It'll be really interesting to see where where those sandboxes form and what people will see as the safe areas to experiment. But to to keep the the list of questions going because we're we're got a lot coming up. So, Luis, if you were able to jump in. Okay. Sorry. Let me find your question in chat. Okay. So sorry. I missed that comment. So could you please say more about the applications to AI, specifically alignment and or x risk as you as you have in mind?
Speaker 1
7:00 – 7:00
Yeah. Yeah. So in the AI front, there's a lot of different sides to this that I think mechanisms have a have a role to play in. One is in the process of training models and the problem of, like, data ownership, attribution. And so, of course, lots of people have suggested ways that, yeah, crypto could be a part of, like, a provenance chain and also, like, a value share, mechanism where models, give back to the the training providers of training data or the or the owners of the IP behind training data. Then there's, like, model usage. So yeah, like, for example, if a model is running in a centralized server that OpenAI runs, probably there's no, you know, interesting mechanisms or problems that that the solution space can help with. But if you have open source models and, yeah, essentially validating that certain text or other sorts of outputs came from a certain model, yeah, basically having a way to prove that. Now that might be more on the domain of cryptographic techniques, which I think is distinct from what I'm talking about, which is these mechanisms coordination mechanisms. But if you're using cryptographic techniques to do those checks, you might also need things like time stamps and ability to challenge and to reward someone who challenges and flag kind of like a a false connection. Yeah. And so that's where mechanisms could come in. Yeah. There's definitely a few others that I'm missing on the AI side. I mean, AI is complicated. Oh, yeah. Another one is just governance. So people talk a lot in, I guess, the the regulatory space about what to do about AI. Right? Like, runaway AI or concentration of power amongst the the the highly capitalized companies that can afford computing data. How do we decentralize the gains from AI? Right? How do we rein in the kind of power of the the new, like, AI gods? So, of course, the regulators are are setting up laws to do that. Right? Like, you can, like like, speed limits or or moratoriums or, yeah, breaking up kind of monopolistic behavior. But I think mechanisms I see as an alternative or even a compliment to regulatory approaches to solve problems like that. So if if, you know, let's say, a few industry big big industry, players or even a lot of open source communities, can decide on some sort of governance and, like, some sort of set of boundaries that you can get penalized for going outside of or if there's proof that you go outside of, you could perhaps design some sort of, like, decentralized AI governance or AI safety framework that you have an incentivized network running around, kind of looking and detecting and enforcing.
Speaker 4
7:15 – 7:15
Great. Thank you there. Steve, you are up next, so please, hop on and ask your question.
Speaker 9
7:30 – 7:30
Yeah. My question's pretty simple. Can I talk to this design large language model that you got going on? Because he's trying to take my jib. I mean, like yeah. I mean, this is, like, exactly what I do. Like, I designed an education platform with peer to peer mentorship, for example. Like, I thought that's all I do is design new mechanisms, pretty much. It's like, my ADHD is so much like a limited token window. It's, like, uncanny. So, yeah, I mean, I I I just wanna talk to it and and you know?
Speaker 1
7:45 – 7:45
Yeah. I mean, I I can, yeah, one on one show you it, give you access. Right now, it's not ready for for prime time. I would be embarrassed if we, you know, put it out there. There's all these, like, keywords that you need to, like, put in the right way for it to spit out the nice output. But, yeah, I mean, I designed it for that reason too because I spent, like, endless time talking to a particular client about, okay. Well, yeah, there's this mechanism that could work for that. No. You don't wanna try that. You know, that ended up not working out. And I was like, okay. One, we need a library for that, but, you know, why not go go the bot direction too? So, yeah, we we need more people that actually done this stuff so that we can make it more intelligent. And then, obviously, you're not just gonna say, okay. Implement the code. Okay. Deploy it. Okay. You know, get millions of people to use it. Yeah. For for a while, we're still gonna need humans in the loop, and I think the bot can only be so creative. It's it's more gonna come up with recombinations. I don't think GPT four is, you know, quite yet, like, able to really get creative and come up with new, yeah, cutting it, like, use cases and combinations that weren't already thought of, but it could be wrong there. Maybe we need to just get better prompt engineering. So, yeah, let's
Speaker 3
8:00 – 8:00
Still have a chance. I got a
Speaker 1
8:15 – 8:15
chance. Good job for a few more years
Speaker 2
8:30 – 8:30
now.
Speaker 4
8:45 – 8:45
And I know we're we're getting close to the hour. And, Yanis, I see that you have two questions in, so please feel free to to prioritize whichever one you wanna ask more in case we only get to one of them before we hit the hour. And if you're not able to get off mute, then I will just read them in order.
Speaker 2
9:00 – 9:00
You you you want me to jump? Is somebody else on the stack?
Speaker 4
9:15 – 9:15
No. No. No. You're you're next up. So
Speaker 2
9:30 – 9:30
Okay. My my question one is is more general. I mean, I was very interested in the value creation and also value redistribution mechanism that you talked about. And I'm interested, like, how do you implement them and how transparent is the implementation to the participants? In which case, they can choose what kind of, like do do they want a winner take all type of game? Like, it is which is the the the corrupted version of the monopoly game, or it could be a a game with, like, multiple equilibria where people can still they cannot beat each other, and then they can still keeping their acting with each other. That's so so so so let's let's try this question, and then if we have time, I can I can give some more? Thank you.
Speaker 1
9:45 – 9:45
Yeah. Yeah. I think the winner take all dynamic is a failure mode of how these mechanisms can play together and then and entering a real world context. So I don't I think that certain mechanisms are probably more prone to, like, a winner take all dynamic, like, let's say, naive like, token holder voting. Right? But then you can have modifications of it. Like, let's say, a rep weighted a reputation weighted multiplier on token weighted voting combined with a one person, one vote, multiple or or a human proof of humanity, like, multiple on a on a vote, and then you maybe you you dampen the winner take all. That's more on the governance side. I guess you're also asking about value transfer and value capture. Yeah. My my opinion on that would be looking more at the spectrum of actually, let me pull something. I know we have no time here, but, yeah, I didn't get to this. But, yeah, I think there's, like, a a spectrum of the kind of ways that we can tokenize rights. And and a lot of people get stuck perhaps on the tradable side because that can scale faster. But then we can build in constraints around transferability and also around distribution of surplus value, like you're saying. So, you know, requiring certain badges to be held by a certain account or a certain amount of tokens for a certain amount of time or a certain number of interactions with other accounts in order to, you know, let them achieve or receive the reward. And I would also, like, just call that gamification. And so just just looking at ways that you could redecentralize if you've realized that a mechanism has a perverse kind of
Speaker 2
10:00 – 10:00
And and that will be decided top down, like like, from an, like, omniscient, like, dog like
Speaker 1
10:15 – 10:15
Yeah.
Speaker 2
10:30 – 10:30
Entity in the system, or it could be, like, from the from the grassroots, from the bottoms up, kind of, like, emergent dynamically from the participants?
Speaker 1
10:45 – 10:45
I think you're asking, like, kind of the the meta governance question. So who is designing system? Who is deciding who how the system updates? Yeah. That that I think the the methodology that we're working on, that that's one of the areas that we're looking in on is how do you do you know, I mentioned, like, participatory, action research. How do you co design a system with the stakeholders that will be affected by it? And and what kind of cases right. Like, from from the complete, like, DeFi bot universe all the way to the small community building a community currency. Right? In each of those cases, there's an approach to go in deucex machina. I'm the designer. I'm a genius. Here's the design. Have fun. And there's, you know, a collaborative way of doing it, of working with the, yeah, the people who understand the problem best and also, ideally, providing them the the tools, resources, the bots, the frameworks so that peep the the affected stakeholders are the ones designing those systems. And in terms of updating them, I guess you could just look to the crypto space to see. Yeah. In some cases, it's a really well informed populace that has, you know, an ability to materially self govern, and in others, it's like a technocracy or, you know, a plutocracy.
Speaker 4
11:00 – 11:00
Great. And that brings us right to the hour, and I I dropped the link in chat for those who wanna follow-up with any questions. I know I I have a couple questions, and so I'll be dropping those in Slack. There's a thread there, and Ori's already tagged there. So, hopefully, you'll have a chance to jump in and see as the availability. But, yeah, I just wanna thank everyone for joining. And as is our norm here, I'm just gonna ask that regardless if you're on video or not for folks to to come, off of mute and just give a quick round of applause as a thank you to Ori for a a great presentation and q and a today. So, on we'll count down. So three, two, one.
Speaker 3
11:15 – 11:15
Yeah.
Speaker 4
11:30 – 11:30
Great. Thank you again, Ori. That was a really wonderful presentation and discussion, and looking forward to where else things go. And, yeah, excited for our communities to intersect and collaborate.
Speaker 1
11:45 – 11:45
Most definitely. Thanks for the opportunity. See you, everyone.
Speaker 4
12:00 – 12:00
Have a good one. Stay