Information is the New Money - Phil Daian

foreign [Applause] [Music] so today I'm going to be talking to you all about hypermavization or information is the new money or why everything is med and it always has been or why geography and privacy these are the two memes I want you to walk away with run the world or why your project is now an Med project which if you've been developing in the space you may have noticed uh in the last year or so so having heard my talk title Shin on uh on Twitter got really excited and said man will we see a complete theory of why information is the new money tomorrow in Phil Diane's talk and maybe that could have been the talk but today it's not sadly I didn't have time to quite polish all the ends on a complete Theory so instead I'm going to try to zoom out like 10 levels from our previous talk which contained a lot of amazing data and talk about some high level Concepts around Mev information Theory privacy decentralization and all of these things that we're trying to build together you should see this talk as like some informed trolling which I think should be super fun for you um so take off your skeptic a hat and put on your fun hat and let's get started um okay so what are we going to try to do in this talk we're going to try to marry these two different areas of study these two different areas of research and modalities we've seen in Academia and when I was putting together these slides I realized I don't really Vibe with the institution of traditional marriage so I at least need to use like uh you know a same-sex marriage at the very least to represent this uh but even so you know not fully on board with this meme but anyway on the left here we have um kind of information based fields of studies or or technical Fields based of study where you reason about entropy bits of information um kind of computational algorithms on top of this Etc things like information Theory cryptography and privacy and on the right we have economics which traditionally has been also a modeling based field but in a kind of very different direction attempting to model human interactions attempting to organize human systems uh much more political kind of bent and here we have like the study of economics cryptocurrency and also Mev options so we're going to try to marry these two disparate Fields today with a single unifying set of abstractions that'll make both able to talk to each other better and hopefully we'll have good communication and a lifelong relationship so why do we think this is possible well here I stole the Wikipedia descriptions of both information Theory and money which are two things that I want to draw some very high level connections today and apply to Mev so information Theory according to Wikipedia is the scientific study of the quantification storage and communication of information at the intersection of probability Theory statistics computer science statistical mechanics information engineering and electrical engineering like we just said all very kind of technical or logical Pursuits whereas money is this kind of nebulous economic concept uh it's any item or verifiable record notice the ore here so that's a very interesting or that is generally accepted as paying in for goods and services or repayment of debts and generally people kind of flag these three properties as being interesting medium of exchange unit of account and store of value but you could argue that all of these three things medium of exchange unit of account and story value as well as payment are really just quantification right quantification is unit of account storage um uh store of value and communication AKA medium of exchange of information which here is called payment so you really see some like direct correlations between reasoning about information and reasoning about money that become particularly Salient once you apply them to blockchains so to unify these two things it would be nice to have a clean mathematical transform where reasoning about money can be translated into information theoretic models Frameworks and results and vice versa I'm not at all claiming this is original many people have been trying to do that for a long time and I'm going to talk about a little bit the history of these mathematical transforms and how we can apply these to Mev and maybe get somewhere um so in this talk I'm going to argue and this is really the outline for the rest of the talk that the key to unifying these disparate Concepts lies in both Mev and specifically in privacy Primitives useful in Meb options um so in how specifically we Define the Mev option that we're running on chain and in kind of the real world that is going to allow us to use both information Theory as well as money to create the best systems for users um so let's talk about the history of these attempts and what's happened in the past so one kind of uh strong existing attempt at this so um exchange attempts um sorry existing attempts have kind of all been under the umbrella of this thing that we call mechanism design um which is kind of a 1960s field that was started by this person called Leonid herzwick when he was kind of thinking about policy how should Central planners reach decisions on policy when the quality of the decision kind of relies on these information silos right and this kind of aggregation of information into a single output is what we call mechanism design it studies the economic as well as the the flows of incentives and information to make these kind of decisions um for everybody so some fun conclusions that underpin this whole field of mechanism design Etc um so first one is this thing called the gibbert Satterwhite I'm probably butchering this because you know it's been a while since I've taken a class on this theorem from 1973 and 1975 and this is two kind of philosophers as well as an uh sorry one philosopher and one Economist that were attempting to study the same problem of allocation and voting and making decisions based on voting systems and they basically in their own model of voting systems noticed that there was a tension between three properties where kind of all three properties seem undesirable to a voting system so the first property is if the rule is dictatorial so basically if we're all deferring to one distinguished voter if there's one voter whose input completely determines the outcome of the vote that seems super bad right there's this critical point you can imagine like the one swing voter in the U.S election that ultimately chooses the outcome we're giving this node a lot of power and that's surely seems undesirable specifically to decentralization Etc so that's bad let's uh let's say that's bad the second tension is between the rule limiting the possible outcomes only two Alternatives so in their model you can get past number one and three if you only consider two Alternatives obviously seems bad right we live in a very complex world and only kind of the listening information or choices between binary Alternatives is super limiting and the last one which kind of underpins and is the fundamental result of mechanism design is if you have if you kind of want something that's not dictatorial with more than two Alternatives you're always going to be subject to Tactical voting so in many cases a voter will not be incentivized to vote their sincere ballot they'll be incentivized to strategically play based on their model of what other people are playing AKA play the metagame As we call it crypto and for voting systems this is super undesirable because you collapse into this world where everyone is reasoning about the metagame people are never truthful and it's very hard to think about what actual outcome the vote is eliciting if everyone's playing strategically rather than giving input that's correlated to themselves um so a fun result that actually underpins the fundamental mechanism design studies of like the 70s and 80s is that if you just add one property here you add utility functions to these individual agents and you allow the mechanism that you're designing to do transfers um over these kind of uh this kind of shared set of integers um you can do things like model human rationality you can build much more complex mechanisms using things like auctions and assumptions about how people will behave against these transfers that allow you to bypass many of the results of classic social Choice Theory so basically what this says is that by adding transfers by adding money we can do much more than we can with pure voting we can do much more than we can just with information elicitation if we don't have money and that kind of kicked off the whole field of mechanism design so money is a very powerful primitive that being said around the 80s and 90s people actually started questioning this result and thinking can we actually take half a step back here and rather kind of going then going pedal to the metal on money and adding rationality to our life maybe we can look at other models that allow these nuanced uh kind of transfers of information that let us achieve the same results as money does and bypass all these classical voting and possibilities um but do so in a different model so this is one paper that kind of looks at this I think the kind of paper itself is mostly irrelevant but I thought the intro was like a very good phrasing of the problem specifically this paper looks at continuous mechanisms which is one modification you can do to the model rather than looking at discrete utilities kind of allow continuous functions Etc same for transfers um again that's kind of a sidebar to what we're talking about but the core of what they were saying in this intro is like let's look at mechanisms without money and tweak them and see if we can still do things without at least classically what's considered money this transfer of like a shared unit of account uh in this exchange right um so what did they say they basically said in many contexts these monetary transfers simply aren't available and we still want to solve mechanism design problems so specifically they were looking at things like um delegation within a single firm where you can't like transfer really money within a single firm or incentives aren't as clean or more political political science so direct politics incentives were often money is viewed as corrosive or regulation where like money transfers may be prohibited in certain contexts among others and they actually found that basically um if you tweak the the model slightly where you basically have principles um commit to the principle of a mechanism commit to a rule and then the agent report uh private information that executes on this rule commitment you can basically achieve a lot of the same properties as you can with money so they're saying even if you don't have money as long as there's a commitment to a rule that determines the outcome and then there's information fed in that is kind of acting on this commitment by the agents we can still get a lot of the same beautiful properties as if we have money uh so this is kind of what I'm going to be talking about for the rest of the talk how does this reasoning about commitment and information relate to money as we view it in cryptocurrency how does it relate to Unchained tokens as we call them money and how does that relate to kind of like the bull monetary case uh for cryptocurrency itself a fun another connection which is in the cs700 UPenn lecture dope class highly recommend you check it out if you have some free Cycles it talks about how you can actually use differential privacy in mechanism design so by tweaking the model to allow for differentially correct outputs and differentially private inputs into these commitments and into this information you can also achieve like a whole class of different mechanisms that aren't possible without differential privacy um so basically what I'm saying is information theory is this framework that's tying together um partially through mechanism design which is one application of information but also in all these connections with privacy and with economics and with money trying to tie together reasoning about these different uh contexts and these same outcomes we want to achieve um so how does this make sense for crypto well it actually gives us a roadmap for crypto like we need to marry information Theory with money through commitments and privacy it's a clear roadmap of what's happened in the past what's been successful in like reasoning and we need to Port this over in a more clear formal way into the systems we're building in cryptocurrency um so this is a little bit of the trolling I'm doing here I have a lot more slides that I'm not going to get through so I'm going to go through it slow because I think this is actually one of the fun ones um this is presented without proof so shout out to roon it was probably uh triggered by this without proof part and also you know maybe you can help help us prove all this um there's a few kind of intuitive results that seem obvious once you start thinking this way one is that if you have trusted private smart contracts so let's assume we have this magic privacy box that we all trust that gives you strictly more expressive power than public smart contracts from a mechanism design framework from a modeling framework of like how powerful is our model which is what we've been talking about with these various Frameworks you can prove this really easily so look at just instantiating like let's say a vcg sealed bid second price auction in crypto if you don't have uh trusted privacy it's impossible to do this securely with two untrusted parties let's say Ethan Solana and uh Avalanche want to run a sealed bid auction together with like the output of the validator set being the bits you could see how you could build this with private smart contracts but it's very difficult to imagine building it securely without unless you have centralization which is which is also private smart contracts but in a much more adult way what does this mean this means that once you have privacy you can build money more auctions so you can make auctions more secure and more powerful with privacy as a tool not only that but there are many auction structures for example the one I just mentioned with vcg not all auction structures but many that require this privacy interface to be able to be trusted to be executed on a chain um so private smart contracts are strictly more powerful otherwise known as informationally better money they can give us more options to approximate more properties of more powerful money than we can without privacy assuming the Privacy comes with no downside so this is the big asterisk here if the Privacy centralizes our whole stack and self-defeats its own goals this doesn't work if privacy can't be built and comes with like a huge cost this is much less clear so we're assuming here we have this magical privacy primitive but assuming we have this the world becomes amazing and the last thing is that if we build systems that are interoperable and have Global reach we have more efficiency right this is the basic premise of decentralization the pi is bigger when more people participate even if a system becomes less efficient the economic value of participation shines through and gets born out so thus the best efficiency feedback loop we can build is this uh loop of interoperability and trust and Global reach uh proof sketch this is Trivial so if people don't participate in mechanisms they don't trust obviously a global mechanism is more valuable than it will um so we're not in such bad shape here because crypto actually already has transfers and transfers in crypto are fully trustfuls whereas privacy and crypto is not right now so we're in kind of this gray Zone where there's a few classes of mechanisms we can build anything that requires public transfers that can be satisfied by the blockchain but there are many things that we can't do the ones that require privacy the auctions that are really more efficient if you have privacy and this is what's causing externalities in the economics of our systems today because many of our mechanisms require trust so we're seeing this privacy because this privacy is economically useful it lets us build more mechanisms but because the Privacy is not trustless we're resulting in trust and we're resulting in externalities um so how can we solve privacy what is the abstraction boundary here what does privacy actually mean and how the hell do we solve it that's kind of the focus of the next um part of my thought so first giving you the bull case let's say we do solve privacy and we come up with this magic black box that I just showed and we have Universal trusted privacy amazing um what will we gain if this happens well first of all powerful Network effects for this better class of mechanisms I just talked about this is going to embarrass the crap out of tradvice so tradfi doesn't have Universal trusted privacy that's why you see regulation that's why you see broker-dealer conflicts that's why you see these auctions that nobody knows how they execute because you don't have this concept of universal trusted privacy so we should make this look embarrassing that's my opinion as a space this is our mission the other thing is we should embarrass classical mechanism design because a lot of the results even regarding transfers assume you have a trusted Auctioneer to do these transfers and if you have privacy and blockchain with some asterisks that we can get into in a separate talk um you can kind of go beyond what's possible there that would create a clear adoption case for crypto right because many markets would just start working better than they do otherwise better for users with better commitments better mechanisms and more efficient apps and you know that's how crypto eats the World by building privacy that is actually better for users and Universal um so where are we today we have this kind of Mev supply chain we're going to talk about various components here including Mev share and Mev boost both of these components rely on privacy so the reason we can separate these roles and gain the efficiency of separation of Duties and keep the decentralization of the validator set of the Builder set of the Searcher set Etc is because we're able to have these abstraction boundaries uh map share and map boost being two examples of money reasoning abstraction boundaries we have today but the Privacy here is not perfect the Privacy here is flashbots just saying trust us and that makes it super sub-optimal and you can see that if you look even at the Mev Matt Mev uh uh not MVP boost any of you share flow here which kind of shows you the transactions that a user makes which represent commitments and the selective data sharing which allows them to Define privacy this is how we build a more efficient Mev mechanism that we believe is best to redistribute uh value back to users and this is the what we believe is the economically best way however it relies on these two very centralized Concepts to achieve this economic output we want so from the information theoretic point of view it's super subpar and from the kind of decentralized system reasoning it's extremely centralizing which makes it very sub-optimal um okay so here I talk a little bit about how to use various techniques to go in depth and get out of this including committees sgx and tex crypto economics MPC fhe and the centralized systems that we're seeing today our belief is that you layer all these things together in a maximally decentralized way to truly achieve decentralization so I'm going to skip this part of the slide which kind of talks about how to use Tes how to use economics how to combine them and get to this world where basically we're going Beyond sgx and we're achieving decentralized privacy I will leave you with this to say that this is the goal of our Suave project it's to achieve this defense in depth on this boundary between the ideal mechanism design for Mev extraction and allowing users to have these abstractions to control their privacy that can then be implemented into cryptography that can then be implemented into differential privacy into information theoretic reasoning Frameworks directly um so we see the Suave Primitives the very formal core of suave as being essentially the translation layer where expressing your transactions through these Primitives allows for a direct translation to and from mechanism design and reasoning about the class of mechanisms we can build the types of money we're building as a community and where the basis of this money comes from and how that relates to Mev and in the other direction uh translating into information and privacy and asking the question of how do we implement this and how do we protect this information web share and MAV boost is kind of the Solo Cup to our Holy Grail it's good enough for now it does the job but it's centralized so it tastes a little bit like plastic right so it allows us to start experimenting with these translations but it does need to be replaced eventually with the Holy Grail and that's what we want your help with that's what we want you to help us with at flashbots and that's what we're doing so this is real we're already experimenting with uh introducing more decentralized techniques and more decentralized guarantees including economics including sgx into our existing mark Market come join the conversation we believe this forms the bull case for all of crypto where Nev feeds into these Suave Primitives people start expressing and leveraging their privacy in negotiations this fuels developments of sgx Tes and encryption Technologies like fully homomorphic encryption eventually cryptography wins and beats uh tees because it's better and I was criticized for putting the AGI bullet point last because I was told AGI feeds Mev and not the other way around but there's your bull case for the economics of growth in the future join us at flashblots.net and I'm overtime so that is all the time I have today thank you very much awesome thank you very much uh so we have a couple minutes before vitala comes on uh one question I wanted to ask you earlier on is like what are kind of your initial intuitions on you like you were describing how it's strictly better if we assume that we have this magical privacy thing that comes with no costs and it works perfectly and everything and but particularly based on where we are and where development like looks like it's headed like what what do you think is the kind of practical state of that is it it's good enough but it has some cost associated with it um that's outweighed I think it's an extremely nuanced question and I would love to give a talk on this this is kind of the extreme version of like the reasoning uh but I think that what I would say today is like the pressure for privacy like the economic side is really powerful like people want to build types of systems that can only be built with privacy that's why math boost exists that's why MF share exists that's why custom dap layer centralized order flow auctions are being deployed um and and that's why people are talking about things like Builder features and using that for for PBS and bank sharding and things like that at its core is because like privacy and commitments together enable like new classes of mechanisms and people really want this um and so what we've seen today is the centralized stuff has kind of been good enough quote unquote for now like it hasn't collapsed it's somewhat economically Diversified but it is a huge centralizing pressure on our systems and fundamentally I don't think that's good enough um adding sgx I think is a better step in that it kind of shackles these actors a little bit more and restricts what they can do um and you know limits them but it's also not a silver bullet because it's still centralized it still relies on Intel Etc um adding committees or crypto economics similarly has like its own downsides so I think we need to like do a combination of these things that just maximally decentralizes every single Technique we can use and then put the max decentralization pressure on because the economics is not going away so if we don't build like the decentralized mechanism for that to express itself we're going to end up in track five so that was like a very compressed rant but that's kind of the state of things today in my opinion one of the other slides you were skimming through because we thought we're attacked on time uh it's talking about kind of the post relay world and how we have to remove all the centralized components part of that starts to become kind of out of the hands of uh people like yourselves where let's say that ethereum doesn't actually end up implementing for example in protocol PBS um like how bad would that be quote unquote in your mind slash were there just other ways that we would have to figure out how do we decentralize that role in a better way if it's not built into the protocol itself and we keep the protocol super simple yeah yeah so I mean I think in many Futures it's not critical that ethereum Foundation if it's hard pushes hard on this maybe the community is just aligned enough around making it like decentralized enough and assuming we can decentralize things like much more than they are today and predictably like it will play out okay for the L1 maybe it's okay for EF to take a step back and we're not like totally screwed um so maybe we can do a lot of this without the EF in terms of removing the relay building decentralized Alternatives building like you know sharing Primitives for users and things like that that being said I think if the EF in practice doesn't have that in its back pocket as like a nuclear option things get really bad because like the worst case would be some angle of centralization in trenches here and we're like in a textbook malloc and so that's why I think PBS is super useful because like many things the EF can do will like reset that molok and I do think it's their a little bit their obligation to stay on top of that just because otherwise their own objectives will be defeated I do agree with you it's beyond my pay grade in this case so like I can say what I think is the right thing to do but ultimately like it's up to the community and we'll try to maximally decentralize whatever is left basically awesome thank you very much uh we will be moving on now