Episode 310 - Algorithmic Game Theory & PoS Tokenomics with Noam Nisan

Transcript

[Music] welcome to zero knowledge I'm your host Anna Rose in this podcast we will be exploring the latest in zero knowledge research and the decentralized web as well as new paradigms that promise to change the way we interact and transact online this week trun and I chat with Noam Nissan professor at the Hebrew University of Jerusalem and principal researcher at starkware Industries we cover his extensive research in academic background starting with his work on complexity Theory the move to

algorithmic Game Theory as well as his current work on Game Theory and blockchains and tokenomics at starware we cover his recent blog post comparing the measurable toonomic outcomes of different live POS systems and share how he aims to better communicate best practices for those designing these systems from there we we explore open problems about how to analyze these models once you add l2s cross-chain transactions me and shared sequencers to the mix now it's not that we solve any of the problems in this interview but it was really fun to explore this with Noam now before we kick off I just want to remind you that the next ZK Summit zk1 is happening in Athens on April 10th the application to attend is now open space is limited applications don't guarantee

a spot but you should still get yours in soon if you want to have a chance to get a ticket I've added the link in the show notes now T will share a little bit about this week's sponsor Alo is a new layer one blockchain that achieves the programmability of ethereum the privacy of zcash and the scalability of a rollup driven by a mission for a truly secure internet ELO has interwoven zero knowledge proofs into every facet of their stack resulting in a vertically integrated layer one blockchain that's unparalleled in its approach Alo is ZK by Design dive into their programming language Leo and see what permissionless development looks like offering boundless opportunities for developers and innovators to build ZK apps as Alo is gearing up for their mainnet launch

in q1 this is an invitation to be part of a transformational ZK Journey dive deeper and discover more about Alo al.org and now here's our [Music] episode today we're here with Noam Nissan professor at the Hebrew University of Jerusalem and principal researcher at starkware Industries welcome to the show Noam hi thank you and I'm happy to be here we also have tun co-hosting this one hey tun hey excit to be here so I'm pretty excited to have you on Noam so you came recommended by one of your colleagues and it was he was highlighting a recent blog post about the tokenomics and POS

systems but it was interesting in preparation for this interview and reading a little bit about the work you were doing I realized that I actually know I know you because you wrote this book called The Elements of computing systems oh building a modern computer from first principles now this was recommended to me like over five years ago when I started to learn about blockchains and then when I started to learn about zero knowledge like the circuits and all of this terminology um I actually went through your book and so it's so cool to have you on the show now wow thank you I'm glad to hear that you heard about my book yeah that was a fun book to write and hopefully also to read you also have like a video series around it a like going through all the the

pieces yeah that is true we have a course at corera that actually actually two courses a set of two courses that goes over the book and that book is really based on projects that you build yourself according to pretty detailed instructions and uh we go over these set of projects and of all the accompanying materials that you need to actually do them and a corsera course and these are courses are pretty uh popular at corsera apparently before you wrote that had there existed any sort of course that would do kind of that level of breadth when it comes to Computing I think that's what I loved about it was that like it really starts from like the very Basics and then slowly works you up yeah had that existed before was this sort of

like a new idea so I think our gimmick if you wish was to actually put everything together so you would have books for specific computer science courses one book for compilers another book for doing Hardware and so on and we thought that if you simplified everything to the utmost you could put everything into a single semester's course and we thought that would be really an important teaching Aid because for once students would get like an overall picture of the whole you know of the whole picture from the ground to the top and that would be immensely useful totally because usually students you know go through a sequence of courses each one sort of disconnected from the rest and they somehow lose the unity if

you wish of computer science so we thought we could make by making everything totally simple just push everything into one course amazing definitely recommend it to listeners if they haven't checked it out also if you are already a computer scientist and know you're way around it but you have friends who don't fully understand it definitely share that book with them as well tun when I told you about this interview though you know noam's work from kind of a different perspective yeah I mean I think a lot of the algorithmic game theory learned over the years came from a couple of your textbooks I mean I guess the obviously the big book in the sense of you know is sort of the nice intro for someone who maybe has a math background but doesn't really know anything about game the I

guess there's two books there's like your book and um yval Perez's book on Game Theory on Game Theory yeah those were like kind of the two things that made me kind of switch more into this field maybe like 10 years ago or something like that see cool what's the name of the book book I think it's just called algorithmic Game Theory right yeah so there was this new research field just emerging we're talking about like the you know 2005 or something like that and it was called algorithmic Game Theory some people call it economics and computation which is you know very similar and you know a lot of people a lot of researchers in the world were starting to work on that area but it was just an emerging field so a few of us

you know came together and wrote this like what we helped would be like the hand book of the field basically you know editing lots of different chapters by all the you know leading if you wish researchers in the field trying to put together what was known in 2005 about this field and we just called it algorithmic Game Theory so I think we've just touched on like two works that you've created but I think it would be really good for our audience and for us to hear a little bit about like maybe the phases of research that you've gone through just to give us a sense for like what kinds of yeah what kinds of problems have you been working on and hopefully that can lead us a little bit to what we want to talk about

today okay we can start as early as you want you can go to the 80s if you want you can go read more recently well up to you up to you where you want to start so so I'm afraid my research does go back to the 80s so I did my PhD in the 80s in Berkeley and my my first area of research was computational complexity I was working mostly about pseudo random generator interactive proofs a computational complexity algebraic complexity things like deep theoretical work and I you know maybe about 10 years I worked in that field and then in the mid90s you know the internet just popped into our lives so I'm old enough to remember that but I was already a professor the middle

of the '90s and this new invention the internet that was seemed to actually be an important uh change came into the world m and you know very quickly everyone in the world knew this was a very significant thing except for I suppose that researchers at universities who had created it in a way yeah so as users we immediately knew this was important but very few people changed their research at that point oh wow and I thought that I have to change my research I mean this is a huge invention things are going to change and there's going to be amazingly interesting research problems in this new area so it took me a while until I figured out that what I find most

interesting about this new world of the internet is the interaction between you know different people with different incentives with different interests different companies with different goals and how you can organize these kind of things and that required taking Game Theory and economics and combining them with your usual computer science and so since about 200000 or something like that that's the area that I'm working with a trying to combine game theoretic you know understanding and economic considerations into Computer Science World with the internet I mean there had been Game Theory studies before but do you feel like with the internet just sort of like put some of these marketplaces on steroids like all of a sudden a lot of friction was removed

like it just changed absolutely so I think that I think that the scale of things you can do on the internet really changes Game Theory so if you're an economist you're really interested let's say in how you do create auction for a single item because that's what economists really are worried about now you go to the internet and instead of an auction for a single item you have hundreds or thousands or billions of items whether it's like ads on Google or whether it's you know slots on the blockchain right for transactions on the blockchain so suddenly you're in this new world you're still looking at the same kind of auctions for items that economists are looking at and game theories that we're looking at but now you're doing it at huge scales and huge

complex ities and that changes the whole story cool okay so this is the switch that you made when you had been doing complexity Theory though was that computer science or was that more like math yeah so I I was in a computer science department but it's definitely true that my work was completely theoretical mathematical I probably for these 10 years I did not program so I love to program before my PhD I love to program now I started programming again when I started moving to this new field but for this decade I did not program at all I never needed it I never needed a computer I just you know proved things mathematically but then you made the shift over and you say you did that

around 2000 maybe you can tell us from that point what were the sort of General topics that you started to focus in on so I think a a large part I worked on a field called algorithmic mechanism design that basically takes the economic field that existed which is called mechanism design which is sort of a generalization of auctions how do you create auctions how do you create voting systems and so on and then said okay what happens when Z moves to computers when that moves to the internet now Things become complicated instead of one item auctioning you have lots of items with interactions between them and instead of actually talking about human institutions that people have used for

years now you can invent your own you create completely new mechanisms anything that you can imagine as possible for computer you don't need to worry about people and their limitations so basically what I was trying to do is take everything that existed in this area of economics and try to see okay what happens when we put it on steroids in computers had you done work in economics before as well like it sounds like it also brings kind of that department more into the space so the truth is no I have not done any economics before that I just figured out that I need game theories that's the kind of analysis you need in order to actually you know work in this new field where you have computers interacting over the internet and these computers

are owned by different entities so they must be trying to achieve different goals so I needed to study that basically and of course as usual as professors you know that we can't study anything by ourselves we take a very smart graduate students and these graduate students learn the stuff and teach us so really is that how it works that's how it always works yeah it's very difficult to learn by yourself once you're a certain age so so had also an added benefit that I'm a member of What's called the rationality center at Z Hebrew University this is really an interdisciplinary Center that focuses on Game Theory and its applications to various Fields so we have people there from economics from Psychology from biology H from law many different areas

and all of them trying to actually look at the common core of what do you do with Game Theory how can you analyze game theoretic and decision issues in different fields so you know that Center that I'm a member of was immensely helpful because I had all these people there that knew Game Theory and could teach me Game Theory and were happy to collaborate and we're happy to see how game theory is applied to new areas when you say that the era where you made the transition was around 2000 I feel like the some of the older papers of yours that I read actually have a little more of this like complexity Theory bent versus economics bent where they're like about the bidding languages right where it's like okay I have a combinatorial

auction I want to Define what a notion of a succinct bidding language is versus a non- succinct bidding language and you know the language part looks like CS but then the efficiency part sort of has economics but it started more CS so I think it's kind of interesting to look through your history of writing where it like started as this like you look at it and it looks like hey I'm defining kind of NP language or or some language with some predicate I make some notion of size of the language I tie the size to efficiency how would you describe that kind of transition so I I think you're completely right in the sense but it's not only my own transition but I think the whole field has been gone in this direction so it started with a bunch of computer scientists figuring out that we

need to use economics and Game Theory to study what happens in computerized systems and at the beginning you could see where we came from you know we think like computer scientists we talk like computer scientists and we try to formalize sing a computer scientists as the area progressed we try we're now I think more balanced so we know more we you know we understand more economics we talk more like economists we use more Game Theory so as the field progressed we're now more to the direction of Economics so probably I'm still not like an economist in any sense of the word but uh definitely more than I was 20 years ago and I guess in that transition when did you sort of feel like blockchains were an

interesting thing to study so I have to admit it's pretty late I still remember we I was at this conference of economics and computation and a bunch of researchers were discussing this new invention the Bitcoin thing and I still remember the number it was you know Bitcoin was $3 then and we were all reached the conclusion that this is way too expensive in no way is it going to you know pay for itself and I just left it for another few years and H so so then I didn't think about it for a few more years and then the Ico wave came and then I thought H this starts sounding interesting but you know the icos maybe are a bit iffy and I think just lately

maybe two or three years ago when I started hearing the word web 3 and it really clicked on me that now this is really an interesting uh possibilities are opening up here I want to understand what sort of held you away like so cuz that Ico boom that would have been like 2017 2016 2017 but like did you feel at the time that it just like wasn't serious enough or you did you still feel like it could be a phase that was just going to pass by let me put it this way so I think Z I thought like many economists professional economists think today so if you talk to a professor of economics about the blockchains the usual response one of the responses that I got from a very famous economist is oh

you're working on blockchains and stuff let me sell you some tulip bulbs right so so one way to look at it it's all just a bubble it has no intrinsic value and it will burst and you know maybe people will get rich in the middle by you know being in this like Ponzi scheme kind of thing and that was sort of the general way I was looking at it I suppose at the time maybe not so strongly as some people have strong opinions today but I didn't see you know what it could really contribute as a significant you know human endeavor it's funny because like the Tulip example like I've lately and I feel like this might get some people mad by me saying it but I felt like the nft thing felt

very like a tulip market but tulips remember still to this day have some value they're not completely worthless well the existence of the joint stock company is due to the Tulip bubble right so like I I think the idea of selling shares in a company to earn some future profits you can thank the toolet for you know yeah between that and the British East India Company is sort of like some combos but one thing I will say is like I like you kind of had this similarly like nebulous view of a lot of things in crypto until sort of maybe 2015 or so when people started at least proving some properties of these networks like I feel like the idea that like hey we ran a bunch of servers

and somehow they're reaching consensus but we don't explain why you know like cap theorem doesn't hold or what or like what the probabilistic definition of these properties is so I'm I'm more Curious like from your perspective you know what kinds of results were the ones that changed your mind more maybe more like academic results about things within the space that change your mind versus sort of Market driven things so I have to admit that for me the real pull was not any academic results but really just uh some kind of realization of the human usefulness of the web 3 in the sense that I finally got into some kind of thinking that this could be

as important let's say as a limited liability Corporation invention of you know a few hundred years ago where it changed it seems like completely crazy how can you take Al loone and immediately from the beginning say you may not return it even though you yourself have money but only the corporation doesn't have money to return the loan right so that sounds like a crazy invention and yet we know from history that that was an amazingly useful human invention and at a certain point I figured out that you know this web three you know agenda if you wish of running things in a trustless way without trusting any single entity you know I finally thought that this is really useful for many things and it's useful from a human perspective so once

I had this motivation say oh this is something that may be really useful it's not such a way to make money or or you know or play with the whatever then you start looking at okay so what are the challenges here and of course there are immense number of challenges lots of difficulties because it's very difficult to actually do very trustful system without trusting any one particular person and then you have lots of really interesting uh you know research problems so we've talked about everything up until that kind of LED you to being interested in the blockchain side of things but what were you able to kind of bring from your previous work into this blockchain world what were the problems that you you noticed right off the bat that you felt you could jump

into I see so I suppose it's the first problem was the transaction fee mechanism design because that's very close to an auction but it's not quite an auction in the sense that you also need to worry about the incentives of the you know validators or sequencers or miners or whatever so it's like a sort of like a two-sided thing and it's a repeated thing so you're really into uh you know basically my field of research which is algorithmic mechanism design so that was the first thing that you know seemed oh this is exactly up my alley these kind of questions now I think it took some time so a lot of the early work in Academia on blockchains was actually figuring out their safety

whether they work well or not right all the Byzantine agreement all the analysis of a longest chain all these kind of questions were really saying okay does this system even work and in what sense does it work and how can you attack it and how you can protect against these attacks and so on once at a certain point I sort of accepted okay so you know lots of research it sort of works now let's see what it means economically speaking and that's where I thought that I could you know put in my expertise one kind of thing about transaction fee mechanism design is at least recently I would say you know a lot of the results tend to be kind of either negative results or like impossibility style results or like

sometimes sad sometimes insightful in the counter examples but like kind of not you know you're kind of like okay well maybe it doesn't kind of fully drive I'm kind of curious how you think about the role of impossibility results versus sort of constructive results or approximation algorithm style results when you look at this space because you know I I think one of the Beauties certainly of complexity theory is these impossibility or like hey like this is sharp P complete you can't do it type of results but there's also a sense in which for some reason people have always found these like weird probalistic loopholes around some of that that you don't expect sometimes around these kind of impossibility results if you like

slightly change the framing and then slightly change kind of what your state space is what kind of lens do you take towards transaction fee mechanism design and how do you view impossibility results that's a really great question so I think that I think in general in computer science so in the ' 70s we had this like NP hardness results NP completeness and oh you can't do this you can't do that you can't do that you can't do that and for a while that's what complexity was you can't do stuff but I think very early afterwards we figured out no this doesn't mean you can't do it it means you have to be more delicate you have to find a more sophisticated model you have to desire less maybe you can't be perfect but you can approximate maybe you can only work on 99.9% of the instances and not at

100% and so on so I think in general that's my that's my point of view I love impossibility results because they really focus you on what you can and what you can't do but I think that the point to Fig a point is you should take an impossibility result as something guiding you towards what you can do rather that in saying that you cannot do something so if you can't do something that's you know 100% incentive compatible as you said as you cannot do mechanism design 100% incentive compatible that's fine you don't need to be 100% incentive compatible you can say maybe it works almost perfectly in large markets maybe it works almost perfectly except for these very h strange situations which we can account for and so on so impossibility results just tell

you what you should be doing instead yeah I mean I think the interesting thing about blockchains to me and sort of at least personally why most of the work I've ever done has been in the spaces they somehow are good at giving you some almost like intuitive way to think about impossibility results you know if I look at like prediction markets from 20 years ago and 10 years ago right like all the results were can't do anything right like oh like it's sharp complete to Source liquidity sharp complete to figure out how to solve some of these problems but then you know somehow in practice people who accidentally just wrote a couple Reddit posts figured out something that looked like had 90% of the flavor of what the original

prediction markets were trying to do uh and then you know when you formalize it you find oh it's actually like hey you took the wrong frame of reference or representation of the problem you like instead of trying to map it directly to like these variables you know you kind of have to like if you look at it in a slightly different environment slightly different frame of reference it suddenly is a lot easier so I Ian with that kind of in mind I guess what what are the things you're most interested in research-wise now and sort of where in your journey are you have you like reflected from the impossibility results to something new or you know so so I suppose I'm working on a bunch of things mostly in starkware but also thinking more generally so there are the

questions of you know mechanism design of transaction fee mechanism design nothing is perfect is possible and then but you have lots of things around it you know we have all the meev thing which of course if you you know can destroy anything but maybe you can do a little bit and maybe you can talk about how much m is possible how much is it not possible how do you protect against it and so on maybe you can't be completely incentive compatible but you know practically maybe you can so for example one of my you know favorite points you know usually mechanism designers do not like the first price auction at least because it's not incentive compatible it's what we as we say in the blockchain world it has bad ux you know people need to figure out how to manipulate their bid and so on

but the truth is it's not so bad in the blockchain world because all your wallets are sitting on the meno they know exactly the situation of the market it's lots of different players lots of different transaction you can get a very very very good estimate of what bid you should put in even if the mechanism is first price like in Bitcoin right so once you look at the the context some things which are considered bad in classical economics Theory maybe are just perfect for us and also in classical economics theories they like you know first price auctions are used a lot because we know exactly how to analyze them how you should bid and so on and that can also be taken to the blockchain world so that's like a very simple example and I think there are

lots lot is going on around it and trying to figure out complex mechanisms for transaction fees which for example in the L2 space you really need because you have many different resources so the multi-dimensionality is sort of jumping at you and you need either to accept it or to approximate it away but you need to figure out how are you going to deal with it I think the interesting thing about blockchains is that the traditional sort of constraints you have in algorithmic Game Theory might change right like maybe individual rationality doesn't need to always hold maybe incentive compatibility doesn't need to always hold maybe you care about you know Cil resistance and credibility much more than you care about bid shading in

a first price auction right like what you're talking about in the multi-dimensional resource case though you have like you know a sort of different set of constraints right like there's sort of some types of resources that you need you know very fast response on like execution some type of resources you might be willing to take a lower sort of response rate on like storage or proving that L2 did something wrong to the L1 to withdraw how do you think about the types of constraints you want to construct right like I you know I think when you read EC papers or you read algorithm Game Theory papers people always jump to like the the five things everyone knows right it's like some type of individual rationality some type of

IC constraint some type of like for all strategy IES that aren't like this this is dominant or expected dominant whatever right what do you think the the constraints game theoretically for the multi-dimensional market will look like because I suspect they actually are quite different than kind of your classical ones so first of all I do think in terms of incentive compatibility but I definitely do not think of it in a theoretical way so I know I'm not going to get incentive compatibility in any theoretical way I just want to be close enough practically to incentive compatibility and there are many ways to get that and that's what I want to to to to do so that's my first point of thinking then when I'm looking about my goal I know my goal is efficiency what is sometimes

called social welfare and an algorithmic mechanism design trying to make sure that we're not losing too much resources and now again the focus is than too much so I don't want to be 100% efficient that's going to be impossible and not really necessary I'm happy of being 80% efficient in some s cases even 50% efficient is completely good enough because you get the security around it you get simplicity so for example it's very important for everything to be as simple as possible why because you're going to get complications whether you like them or not so whenever you can avoid a complication it's going to pay for itself so even if you lose a little bit in incentive compatibility and on efficiency and so on if you Simplicity will do that for you that's really

important and it's going to pay off so basically that's what I try I try to get basic efficiency basic incentive compatibility none of them perfect and try to do everything as simple as possible so for example if you tell me about talking about multi-dimensional mechanism design my first attempt will be to be single dimensional approximate everything by a single resource if you can get away with it that's perfect that's great and sometimes you know your bottlenecks are always going to be in one resource and the other you can just you know sort of with a little bit of loss of efficiency take care of sometimes you have two different bottlenecks and you have to deal with them separately and then you need to do a really multi-dimensional thing annoy your users because they are going to have to pay

separately for two different things they're going to have to keep it in their mind but if you need that that's what you do yeah so one kind of thing at least you know I've been thinking a lot about in this regards is you know if I look at the three pillars of algorithmic mechanism design you sort of have mechanism design contract theory and then sort of signaling mechanisms right like something where I provide signal and they kind of You could argue that the contract theory is sort of a mix of the two right like there's some signal provided by the principal but there's kind of this question in blockchains because you have this public Revelation everywhere and you have a bulletin board everyone agrees on you could do better signaling mechanisms like you could have the user signal some type of intent and

have the mechanism input some of that signal versus just purely historical algorithmic pricing how do you kind of view given that you've written papers sort of at least in I know in two of those you have the Bobby off paper on contracts 2012 and then you have a ton of auction papers how do you view the three different prongs as important in decentralized sort of mechanisms so I think to begin with you know before we can talk about how do you put them into the blockchain world we're only starting so as opposed to auction Theory and mechanism design which we've really you know adopted into the computer science world with much work and lots of papers and I think pretty sophisticated understanding the other two pillars that you were talking about

contract theory and information design we're only now starting to do that and I think there's really interesting work to getting you know more algorithmic contract design and combinatorial contract design and similarly combinatorial information design that's just starting now and I believe that these two Fields will grow in the future first you know among computer scientists trying to do Game Theory and so on and once we understand that a little bit better I'm sure that we'll also find it way into blockchains but at this point I think I don't know how to apply any one of these newer field at least for US computer scientist into the blockchain world in a useful way I think it's bound to happen but maybe not in the next two

years I'm going to pull us a little bit out of this rabbit hole and bring us a kind of a little higher level basically the work that was first suggested to me when having you on the show which is this blog post called simple tokenomics for a proof of stake utility token I kind of want to talk a little bit about that work and then I have some questions about like how these things apply to l2s but what was the goal of writing that post this is on the starkware blog it's quite well presented for someone who's not super familiar with the material like there's a lot of defining of proof of stake and all of these things in the beginning which I always like it when posts do that when they kind of like bring the user in they don't throw too much lingo right at the start instead

you defined it but yeah what was the goal of that work and then I want to ask you more questions about it the goal was that you know I was thinking about issues relating to our own token our own mechanism design our own minting and our own you know staking rewards what we will have when once we are completely decentralized this is for starkware this is for starkware yes or Stark net maybe well this is for Stark net when I was working for starkware cool right so you know and these are sort of you know slippery issues and many people are looking it from many different angles and I was thinking Okay C can we have a very simple kind of explanation of what's going on there yeah and I really

wanted to get into something very simple and I say it like an economics 101 sense that sort of gives answers that make sense so of course you know it's going to be very simple and it's not going to capture everything but you know I always want to start with a very simple point of view and from there we can you know make things more complicated and more exact so I you know I was thinking in our context and then I tried to see okay how General is this can we generalize stuff and I thought that I reach two I think important principles that I think could be like the basis of uh you know thinking about economics of a certain types of systems specifically of systems that have utility token of proof of stake systems with utility tokens and I

wanted to write it down in a very generic way not related to starkware itself or to arto stuff but in a very general way it could try to appeal to people and say you know you should try thinking about stuff this way and you know that's how I like to think about it so that's what I was trying to do in that you also compared a lot of kind of top blockchains that are alive that you know there is data on now about how it performs what's the staking ratio how many people are actually staking you have kind of a nice chart which shows the staking ratio and adjusted rewards I never seen that compared actually so this is really kind of nice to see when you're laying all of this out are you trying to provide just Frameworks

for people to think about their own tokenomics when they're designing them or were there any like best practices that you actually learned through this work yeah so I I definitely have an agenda here of best practices okay and these are two very simple rules if you wish rules of sum which I think you should should be the beginning of thought about how to do these kind of tokenomics and the first rule was like a microeconomic kind of analysis saying how should you price your transactions and the point here that I was trying to make is that they're correct pricing if you wish now of course remember this is extremely simplified only as a starting point right but the correct pricing should be

What's called the marginal costs so that especially means you know how much more would it cost you to run this transaction relative to everything else without this transaction in particular the most important point about this kind of pricing about this you know the way you price this thing is that you don't want the transactions themselves to pay for all the you know startup costs if you wish for all the fixed costs as economists call it especially you don't want the these transactions to pay for the capital cost of staking because that's an enormous amount and if you let the transactions pay for them you're basically driving up the price of transactions you're artificially reducing the number of transactions that

are going to you your system and that's going to be bad so you don't want transactions to pay for that so that's the first point that I wanted to make and that sort of tells you how you should price you should price according to marginal cost that needs to take into account the congestion which is sometimes the most important thing but not capital costs and that leads us to the second principle so who will pay for these fixed costs I mean someone needs to pay for that if I'm not letting the users pay for them who is going to pay for them and the point that that I was trying to make is that it's fine to Mint this these new costs that's actually one of the strengths that we have in the web 3 world where we have our own tokens where we have

cryptocurrencies this is something you can do with them you should pay Capital cost you should pay fixed costs from new minting because if you're looking who is really paying for that the people are really paying the current owners of the tokens why should they pay because basically if they get more usage of the system you know their platform grows so they will eventually actually hold shares if you wish of a more significant ecosystem so for them it's not really paying taxes but more like investing in the community so that's the second point that I was trying to make which is like a macroeconomic them because from there you can figure out how much minting do you want to do how much you know our

Staker is going to get and so on from that work though and well as I was looking at this Stark net is going to be an L2 how do any of these things actually translate because l2s originally like rollups they they were just meant to use the consensus of the main chain the da that was like the settlement layer and they were just going to have and they kind of currently do have this like these sort of like centralized sequencers that just did that but there was no additional proof of stake Network now that's changed in a lot of the networks they actually are introducing some sort of L who proof of stake I mean it's not I know it's not actually Pro it's like a decentralized sequencer or something some sort of multi-agent

decentralized group that's going to make decisions how do these and like there's tokens and there's tokenomics on those l2s as well how do these principles translate it becomes very complicated I feel like I like you're like we're starting simple but I feel like it's going to get weird so okay so so so that's actually very very interesting question so I think all l2s really started if you wish like normal corporations but just working in the blockchain space so for example you know starkware itself its first product was called Stark X was just a centralized thing you give us your transactions we you trust us and we'll put it you know an ethereum and L1

for you right so this is just you know we're sell software as a service we're doing something for you completely classical kind of company only we happen to actually do something that only make sense in the blockchain world but we're not ourselves in like a web three company if you wish we're only working in that space in a standard business model that was our original product but when you start looking at it so people who are you know using cryptocurrencies using blockchains and so on the reason they're doing it because they want what you get is a promise of not trusting anyone anyone specific but getting this you know consensus this generalist trust without a trusted party and if you're now an old-fashioned kind of company

giving this service then why would they trust you even right I mean so so it's sort of you know having this intermediary in the middle sort of takes away the whole point of doing this kind of totally so it's not 100% because you do get eventually it's on L1 and so on but to a large extent that's a situation so that's why I think both my company you know starkware and many other companies in that space figured out there's nothing they can do they also have to be decentralized them eles yes to gain that kind of trust because otherwise you know their users are not going to want to use them but some other decentralized thing so that's why H you know l2s are becoming decentralized as well I think that's completely necessary

from that point of view now once you're you know in L2 then the way that I'm looking at it your to economics are going to be like just like your you know providing any other service so suppose we're providing a service I don't know of running you know popping balloons to the air or something like that but doing it for some reason in a decentralized way so we're providing a service there's going to be some utility that we provide otherwise there's no reason for us to exist and the token is going to be used to pay for that utilities and to run the system and so on and now you're just you know it doesn't really matter whether you're in L2 where the system you the service that you provide is to put something on L1 or the service that you provide some other thing but you're

still doing it in a nice decentralized way so so from my point of view l2s are just one example of the basic toonomic kind of point of view of platforms providing a real service that's useful to someone and have a token to support that would you then just use the same kind of breakdown in best practices on the L2 I would say yeah I mean again as as a first starting point right of course so there are always going to be differences for example maybe you need a little bit less Security in terms of fraction just because you get get a large part of your security from the L1 so maybe you need Less on your own and so then you start talking right but generally speaking yes I so wonder what

happens once like as these networks have evolved we started to see I mean two years ago we were talking about bridging we were talking about like tokens going between these things synthetic tokens and the shared sequencers and the me that could be captured from these like I feel like the toonomic that simple tokenomics does it not get influenced by all of those other things around it eventually do the people who design these systems have to have the foresight to take into account all of these crazy like in between cases that are happening this is also to you tun by the way cuz I know you you play in this territory I mean that's a difficult question to answer because the answer is always you know both yes and no right so I guess

it's like you can't predict it but you do need to keep it in mind on some level yeah so I suppose once you're designing a any particular system you have to understand everything that's happening in your system and try to take into account everything now the question is how do you do that so the way that I can think of doing that you start with basic principles and then see okay what do we do here here and there and there how we if you wish plug in everything that we know into this framework that we started with which was simple so for example let's say if you're talking about you know me is a big issue in our space of course and my point view of me is still the very simplistic thing it's sort of implicit fees on the users the fees are

not set in the correct way they have no relation to cost to marginal cost they have no relation to anything except how much you can take advantage of this particular transaction so I consider them something that's bad if you wish okay so maybe you know there may be some positive aspects to it maybe you can't eliminate it all together all that you need to take into account but from my point of view if you start thinking about me and your platform the starting point should be we're against it now let's see how can we handle it and nevertheless maybe we can't completely eliminate it maybe there are some special circumstances why it's positive and then you start actually getting into the

details but having a you know a departure point is sort of useful I think I want Trin to answer to the me is bad because I feel like he's been spending time trying to convince me that Meb is necessary evil I think you can't yeah you can't you can't get rid of it I mean yeah I I feel like 2023 was just for me just writing a lot of papers showing lower bounds of how much you could remove and couldn't remove but I I just like generally think it's good to have this idea of lowering it as an objective but with the understanding that in a permissionless world you know I can take an application on its own I can give you some bounds on

like how much me exists someone deploys another application Again permissionless by adding the other application I have super linearly or potentially multiplicatively increase the amount of me between the two of them and now all of a sudden my initial analysis doesn't work anymore so I think in the permissionless world there's a sense in which you're never shrinking it to zero uh and in some cases if you can direct it towards positive things like if there's some types of constraints you can add you can at least have some guarantees on worst and average case uh Behavior but yeah I I just I do think like there was a long time of people being like our thing gets rid of all me and most of those were you know not not

correct I I believe I agree with what you said so H but still you you know when you push you can decide which way to push right which direction yeah con finding constraints that help you have a better grappling on it I think is the ideal end state but I think there there was a while especially I think 2021 2022 where people would either a say they've solved it and hey we just shoved everything into a ZK circuit or an NPC and now we have no MV and it's like that's not really true yeah but tun I want to kind of throw that question to you too of like once you introduce and and maybe I'm wrong in thinking this like bridging

shared sequencers the decentralized like the sort of proof of stake on proof of stake when it's like a POS base chain and like a POS L2 with like just these extra things does that actually impact how the tokens perform like the cost the fees or is that like do those l2s still like each individual roll up does it still get to exist in this sort of bubble like apart from some of the other stuff this is me actually trying to understand like do the fees do the tokenomics change because of those other things inevitably the answer is yes but the thing is those other things you view as like a particular application that

application generates some fees the application also has some value that enters or leaves yeah and the question is like do the fees compensate the average net flow of that application in some Manner and you're right that these transaction fee mechanisms all will interact with each other and you're never going to get a perfectly isol ated one like the analysis and in most formal results kind of is is unrealistic in the sense that like you know I have one roll up with a particular set of fees another roll up with another set of fees uh one of them has fraud proofs one of them has ZK proofs so one of them is continuously paying fees you know to have the proofs generated the other one is only having these like worst case there's fraud

that's when you have to pay the very very high fee those are two very different systems but yet they're interacting once they're bridged right trade twoes and the sequencers arees them think you're right I think it's it's very difficult to study formally and maybe it's better to yeah come up with some heris first before kind of at least that's I know it's a copout answer that's it's an answer they saying it's like oh it's too complicated to analyze but yeah I do really think it's like each little universe will choose their own fee structure and first you have to see how that I evolves before you can talk about how it interacts with other ones so let me say something so I think you can see that you're like a computer

scientist I don't know by training or with heart but because you know computer scientist not by not by training no I did I did pure math before but okay pure math okay so so with computer science when we study a model and we expect the model to really represent reality very well so if you know we compute something and the answer should be 5.3 and we get 5.2 we think some things has gone bad economists they're looking at it completely different they know they're working with very complex systems with lots of interactions and with people also thrown in which is a you know a whole different level of problems and if they expect something to be 5.3 and they get 5.2 they're ecstatic they say wow our model is so exact it's so perfect

it's so great yeah right so you know the question is what are your expectations once you manage expectations and you say we need to basically figure out how these things work not completely not then you can start studying these you know systems not in complete isolation but maybe just you know try to formalize like one or two parameters that they affect each other and so if your expectations are like an economist expectations which are low H then you can probably do things yes r s equals 0.1 is always a great low lower Bound for for people yes that's maybe extreme but something like that yes it's funny as we talk about this I

almost like I remember like a few years ago when there was only one roll up and it was a super simple rollup that didn't do really anything and I feel like you know then there was a few more and we could look at how those behaved but we're on the brink of and it sort of already started of just like a deluge of l2s and roll up and interconnected networks like the bridges have been there for a while but I don't think they've become seamless I think they're still a little bit clunky but I feel like that's going to change I think we're going to have much more fluidity between all of these different places is there anything that you're like worried about when that kind of scenario happens

like is there any sort of concern on the horizon of like what could hurt these systems so when you you say concern in the blockchain world I you know my main concerns are still like at the you know Society level we have this whole system we're trying to make sure that governments don't control everything that corporations don't control everything and we know that it's not always a good idea right so no government means you know terrorism crime and so on and all of us are sort of hoping that we as human you know the human race are smart enough to get the good without actually paying the price of the horrors that can happen so if I'm worried about something is that you know great human level now when

we're talking about the small things about what will happen with L2 I think it's a really interesting question what will happen so one can think of you know if you wish two extreme possibilities one of them is all these different l2s interoperating with each other with money moving back and forth and with everything seamless as possible that's one possibility the other possibility will be a consolidation right there will be like one or two or maybe three different l2s each with a significant purpose and all the others with just wither away and I don't know what would happen probably maybe something in the middle but I think it's really interesting to see or alternatively everyone starts moving over to salana which I kind of don't think it's

gonna happen that's if you believe in the ethereum world right absolutely the AL ones is another possibility yes absolutely I don't know why I've become like the biggest Salon shell in the last few weeks I like I have no connection to it by the way I just yeah I think one thing about the complicated system so now now I maybe I you know I agree maybe I'm I'm over relying on symbols versus heuristic models but I will say the following I kind of think you know the classical mechanism design world I feel like really changed when people were like okay what are we really designing here well we're pricing some notion of the externality caus to each user and then that's how we measure welfare

revenue or like that's how we should price entry fees and there's sort of some sense here in which we don't know the externality that one L2 causes on the other by changing its fee right like the existence of salana to ethereum what's the externality that causes to ethereum users indirectly because now maybe all the volume that was trading in eth usdc suddenly is trading in some meme coin to usdc and so all that liquidity leaves ethereum causes prices to be worse and you know somehow welfare decreases right I I I think the hard part about permissionless systems and maybe this is the opportunity of them is that you can never really figure out how to boun price the externalities completely and then like that's probably why you know it's like you get all these

impossibility results that are just like Grand proclamations but I think that there's also a sense in which it doesn't matter to real usage that they exist it's it's it's a little bit like neural Nets in that way right like all the negative theoretical results have literally negative bearing on the fact that the things are used every day right like no one cares I wonder like could you almost like find out if all of the usdt on Tron has any impact on usdt on ethereum and we know that's true right we have seen that happen wow okay because there's like a really clear singular case a a really simple example is if you look at the distribution of sizes of transactions so like take the

distribution of like the number of $1 transactions $2 like bucket it by some interval and compare the distributions pre Tron usdt being over a billion and post you see that the the small siiz transactions on ethereum all disappeared like the weight of the distribution went down significantly so if you condition on the small sizes you look at some Divergence measure some measure of distance between the two distributions you can see that it's statistically significantly changed so that's an example of an externality right like the introduction of a cheaper blockchain lower fees caused all the low value transactions to move and increased the high value transactions because the people who are high value transactions

well they were already large eth holders anyway so they're that's what I mean by like these things I I don't know how to measure this notion of externality like introduction of each new rollup or each new fee mechanism to the whole system it they have these kind of weird things that you can measure afterwards let me you know pitch again the Simplicity point of view so just taking it real complicated yeah so so here's like the simplest way I like to think about these things and I think they capture a lot of what you're saying you know but just from very simple principles right so all of these systems have very strong Network effects you know there's huge you know the externalities basically of having a bigger system that has a lot of value so

these are classic Network effects the more you have in one system the bigger the effect is so if you have you know two different blockchains each one separate that has much less value than having all of the you know all of the users all of the operators everyone from these two separate blockchains being on one blockchain okay that's just a classic Network externalities kind of thing so from this point of view any split if you wish in the blockchain world is bad for the world right because you want everyone to be working together now of course you know that can only go so far because once you have only one system there's no competition you can't trust so of course you're basically you know kicking the shoe under yourself

kicking the chair under yourself I suppose but but I think the the first principle should be is that you know separation is definitely bad you feel like a system that sort of sticks more together is going to be the stronger one but I mean we kind of yeah I think for the future there's going to be a lot lot of movement between l2s at least like there already is right to some extent you just the fact that l2s are above L1 the same L1 let's say that gives them some interoperability especially those that are evm compatible I feel like those ones especially you see a lot of flipping like you see apps

deployed on all of them people will go and play on all of them I don't know that there's like a moat necessarily I think I think you're right so if you're all of your evm compatible then you get the added benefit of you know the same developer base if you wish right so that's another Network effect that you're getting but even if you're not an evm based you get the network effect of you know clearing let's say using ethereum or the e or and clearing together and so on so there is some Network just done by being in the same L1 of course if you were just one L2 there will be even stronger Network effect because you could transfer you know between people on the same network the same blockchain much cheaper right so there will be even more of a network effect right so there are different

levels of network effects and of course even though more Network effect is always better each one of them pays a price for example if you're limiting yourself to be an evm as you know Stark word does not then you're losing something because you now have to be an evm right so you gain something that you get more developer based but you're losing something that maybe you have a suboptimal language for your purposes right so that's a kind of trade-off that you need to play with always yeah yeah so one interesting kind of thing that seems to be starting to happen and I I'm kind of curious how you think about this is that you know a lot of the rollups that have sort of networks of rollups so say like rollups that all use op stack

or optimism rollups or arbitrum stylist they they kind of share some execution semantics and also share sequencers it's almost like they're a federation they're kind of like the US right where like each state has its own taxes but they have to like tithe to the federal tax and you kind of have this like loose Federation of you know I can change my transaction fees up to some amount but the sequencer has some like minimum they charge there's sort of some notion of like we're sharing our fees in some way how much do you view these kind of like Gomer of rollups versus individual ones because I I kind of you know I know you said earlier like hey they're the ones that have the highest utility will kind

of have this network effect and you'll stick to kind of those few left but then you also see these kind of like federations forming and I kind of think this like federative thing changes the the way you think about transaction fees because there's now some component that's common to all of them plus some amount they're allowed to change and there's you know these kind of constraints how do you reason about a world where that happens um okay so I think that's exactly the tradeoff so you have on one hand the pull of one network is bigger better than separate networks on the other hand you need to specialization because different blockchains serve different purposes and this is somewhere in the middle and I think it seems to me that there are many places where this is useful and actually

we're seeing that it's happening so it probably is useful to someone so and I think there is an interesting uh area there that will develop the Federated idea tud I've never heard it put that way I like that I mean it's basically the European Union or us in some ways right there's like some economic Freedom via the sequencer but you're paying some minimum tax and between op stack and stylus I feel like you're already starting to see this happen but I'm sure every rollup will eventually have this kind of like you know once people have trust in their particular infrastructure uh as well as like the sequencing Behavior as well as like their proving Behavior whether it's ZK or fraud proof

then people you know who want to make something that's slightly different that needs a little bit of specialization will naturally kind of break off the question to me is is like what's what do you think the equilibrium is do you think it's like the big federations or the dominant ones or do you think it's like isolated you know groups like cuz like this is like betting in the 1600s on you know whether the European Union would exist or not yeah so I think you know these are fascinating questions so I believe there is a place there so you know the nice thing about this Federated kind of idea is you get to choose how much of the network def you get versus how much you know and if you wish freedom to choose your own optimizations which are maybe different from the other you know Sister chain right and it's

really interesting that you can play this game and choose the sweet spot for you and what the equilibrium will be I mean I don't know I don't think anybody knows and that's what's so interesting about this space it changes every year that's why I said it's it it does really feel like being the 16 or 1700s I'm not sure what the French Revolution was maybe maybe FTX crashing and so now we're just kind of like looking into like which types of societies Will Survive the longest that's true for from my point of view so I still remember like the internet boom right like the '90s and for me one of the nicest things about the blockchain world is that it somewhat reminds me the excitement and the like the the you know the world of building a new world is very much like

that so and I that's something that I haven't seen since the mid90s S such excitement and such a new I don't know human creation be huge human creation being developed so I think it's really interesting like in '95 we couldn't imagine I don't know social networks and not to mention the social effects of social networks on the other hand we imagine door Dash and Uber just a little too early right they needed a phone yeah cool I like that you say that it reminds you of that era um I have a question about ZK like zero knowledge I mean starkware has got ZK at its core Starks um do you work on that like is

that something you focus on or is that something that like a different part of the org works on yeah so the truth is that I don't really work so much on that first of all because there are so many strong so strong people in and that you know in starkware in that area that's not something that can help anyone I mean they're amazingly strong they're mathematician crazy mathematicians I mean I'm not just just talking about Ellie was like really a professor May his career on this but other people well so I'm not working that but also for me it's like sort of like a deu because this you know this really goes back to my original work like 30 something years ago on complexity Theory it really developed from interactive proof systems

which became suin proof systems and so on so for me it's something that I left I don't know 25 years ago and I'm looking at what happened it's like you know seeing your baby grow all grown up and come full circle to combine with what you do in an interesting way it's really an interesting way so so I really love looking at it and I love you know hearing all the sophisticated mathematics that these guys are doing but it's not something that I can actively contribute I suppose at this point well I had one last question about the I I will say I should have asked it before I just realized do you think like in the work that you're doing would you be designing like if Stark net decides

to also become a Federated system of sorts is that sort of more the realm that you'd be focused on like how to do that and what like token economics would look like in such a scenario yes can can you tell us if you are planning this Federate no I can't tell I can tell you mean what should be clear we're also talking about it right it's not a bad idea and there's lots of things you can do and you can't do and definitely you know we're thinking about it and I don't think I can say more first of all because you know it's something that we're definitely talking about but I can say that we're talking about it it's not and I I that's definitely something that I would be interested in pursuing you know as a personal kind of endeavor that

sounds very cool uh so you know thanks again for spending a lot of time with us today I had one final question given your kind of publication record in writing really influential books and as someone who you know grew a lot from reading your books are you planning on writing a book on anything in blockchains web 3 tokenomics any of the things we talked about you know what's what what sort of your view on that so I definitely don't have any specific plans of writing anything yet so usually I write a book when I have something inside me that I have to basically get off my chest tell the world and then I you know I don't need to worry about it anymore because I was able to tell so I suppose that if I stay enough in the

blockchain world and if I have something a way that I want people to look at things that will be a reason to write a book before that no it's lots of work well no I'm thank you so much for coming on the show and kind of like going through this journey with us also I love that idea that something you worked on in the past complexity Theory and interactive proofs comes back together with the work that you do yeah thanks so much for sharing that with us it was great being here and thank you cool thanks tun thank you excited thanks for having us and I want to say a big thank thank you to the podcast team Rachel Henrik yonas and Tanya and to our listeners thanks for [Music]

listening