A distributed OS for the world computer - Adrian Brink

Transcript

welcome great yes hello thank you very much for coming everyone um today oh I have a clicker as well right yes so today I'll very quickly go through AA a which like you need to it moves the notion of what we think of as blockchains really like it is not just like yet another L1 it is more of a distributed operating system that canun run on a global L1 that can run on many different l1s and that connects all these systems together so rather than having to say that well I'm currently on Salan or I'm currently on ethereum or currently on arbitrum you can rather say well I would like to do an action and in code you define whether you want to

execute this in something that has for example pbft consensus or something that has Nakamoto consensus right you can Define all these things in code AS intense and then sort of the operating system abstraction handles that complexity for you so that rather than you having to figure out how to use all these different systems you can use one operating system uh this also leads to the next point which is it allows you as a developer to sort of and I'll talk about this in a sec a little bit more but like think of this very fundamentally as in like currently we're living in this pre- Windows moment where everyone keeps building programs for specific CPUs and it's like you have to pick whether you like the IBM CPU or the Intel CPU and if you want to switch to a different CPU you have to check whether

different programs are available like as a developer like everything changes all the time and you do not get the standardization effect uh that you get with a noral as a distributed operating system so yes with that so currently we live in this very messy world where everything is all over the place there's zero coherence to this uh as a user I mean if you're active here you probably have around 20 different wallets because for some reason this is a standard we evolved into at the moment that all do various things in VAR slightly different ways um and you do not and as a developer this just gets worse I mean as a user this is already bad but as a developer you go well I have to pick maybe I'm building for the evm but there's now 15 slightly

different versions of the evm and like we're living in this ecosystem where it is very difficult to reason about the actual distributed systems properties that are at play here right like it is relatively simple to reason about properties that play in a single environment for example like ethereum main chain that's fairly standard that's like fairly well understood how to do this but it's very difficult to go oh I'm my users live on 15 different systems and they have various State and various components of those systems and now how do I represent this to the end user in a coherent way I mean like one answer to this is and I hope mats isn't in the room anymore I'm like Dune is very cool right but it's like the answer that we're coming up with at the moment

is like well let's just put a bunch of centralized servers everywhere and then this problem magically goes away well if the answer to the problem was more centralized servers we didn't need any of these systems like we knew how to build fast postgress databases 20 years ago this is like if the answer to a problem becomes well let's just bu buil an indexing service somewhere and someone just runs it they we've solved very little at this point we've just like built the shittiest kind of database Solutions known to man in the last 20 years um so this is the current state of the world um and with Ana and intense this fun model fundamentally changes because now all of a sudden you can think of this entire ecosystem as rather one unified

operating system it's a unified but distributed operating system where different parts of the operating system provide different guarantees for example some places will optimize for storage some people some places will optimize for execution um but fundamentally you as a developer and this user can treat this as unified thing where you get a uh State like a single operating system API that you can build against um and by the way like for a lot of users maybe the right trade-off model is actually that they want a bunch of centralized Services um because like this is the trust model that they care about and are almost fully compatible with that so like if you it it just makes it visible to the end user and the developer on what model you're actually in um yes this is very important um because

everyone is building intermediaries um and a no a distributed operating system it's a developer interface it's not an intermediary there's never a specific anoma instance you must send your thing to right like there's no specific Global anoma L1 that you must send your intent to uh where you want to send your intent depends on like the user's local trust preferences essentially or like if they want to use AA as an operating system on ethereum they just have to send it to notes that interact with the ethereum network if they want to use it on the global L1 they interact with the global L1 if they want to and I'm a big fan of like Sovereign things so if they want to run this like for their local Village they only need to interact with people

locally over for example some Mesh networking that they're running locally uh this is all compatible with this so there's no intermediary because especially with the chain obstruction people I mean a lot of people are saying like well we get rid of all this complexity by just introducing yet another middleman and it's another middleman even if that middleman is decentralized it doesn't really change the fact that now we've just added another hop into the trust model whereas here it's like you pick which servers you want to talk to which nodes you want to talk talk to and you send your intents there um yeah it provides a level of abstraction that is compatible with existing chains and VMS and this generally means that anything that's expressive enough to at least verify a uh recursive Zer knowledge proof will work with anoma um IE you can um deploy

the operating you can deploy anoma to those systems you can deploy anoma um because anoma in the end is just a large uh recursive rollup or like recursive circuit essentially where you can as long as you can verify verify that circuit you can run this wherever you can verify verify the circuit um to deploy to things like Bitcoin this may get harder I don't know the Bitcoin people are doing crazy things though with ZK at the moment so we'll see where we get there um yes and fundamentally it enables a new paradigm of intent Centric applications and this is really like I mean I've focused a lot here on the distributed operating system part but one of the other core Innovations of anoma it's It's fundamentally intent Centric it is not transaction Centric it is and this really just Maps better into what users actually do because when you

like when Bitcoin came around we lived in this world where like I go to a store and I'll give someone some Bitcoin and they give me some milk that's a fully balanced transaction where I just do a payment right but nowadays like 90% of all the use cases are I have some a for B who else has something to settle with right like I would like some eth for Bitcoin who has some Bitcoin for eth but those things aren't transactions they're fundamentally intense they are partial State transitions right where um I cannot directly settle my intent um and so we came up with a this is one of the other things that so like once you have generalized intents you also need a way to do generalized counterpart Discovery and for example unop is like the poor man's version of counterparty Discovery where we said well counterparty

Discovery is actually very difficult you know what we'll just dump all the state in consensus and we have like the single Global lock called a block time to do all Global counterparty Discovery uh I mean we're seeing right now this absolutely doesn't scale because it also makes no sense like if we in this room and we want to do some Trading we should do this locally we can like do our local counterpart Discovery if there intersections between our intents we can settle them locally if we need to have access to more liquidity we can sort of reach out right but the more the moment you broaden that sort of counterparty Discovery Space that moment you take trade-offs around privacy around security around um actual latency and like I like to joke about this but um no system can be faster than AA um because

AA fundamentally like is deployed to the scale that you care about it it's scale free in that sense like Solana can never be faster than this because no amount of global consensus can be faster than us running some local consensus right and like if the only thing that we need is some local consensus to do some local settlement that's all we need and this will always be faster than going I mean it's at least going to be what like 800 milliseconds faster depending on speed of light conditions around like how fast we can get to Global consensus so even the fastest thing is always going to be fundamentally Limited by um the speed of light constraints um this also comes to the other thing I'm I I think blockchains weren't like this industry wasn't there to solve the problem of

like how do we do really good globally decentralized things like the idea was like we want to be res resilient and resistant against capture and like the answer is not like well we have this big thing that's very hard to capture the answer to True resilience is we have many local things that are very hard to in uniform uh comp compromise right and this comes from like my thesis is fundamentally that sort of on the day on the eve of World War I Bitcoin and ethereum just halt because most fiber goes away um and Global consensus is going to be shut at that point and so like you want to have systems that um are resilient in the sense that they are capable of adapting to the actual underlying topology that's currently available um yes why does this matter

for ethereum though and why does this matter generally across the space well because it allows you to velop ER to fundamentally write one application that where the user then makes a choice where does this application run what specific execution context does this application run but so you as a developer you don't have to make the choice oh I'm going to Target all the users in arbitrum or I'm going to Target all the users on optimism you just make the choice uh of like you build an application against the operating system and then the users make the choice because some users may want to be on ethereum some want to be on arbitrum someone salana someone may want to be on multiple things at the same time right and so now as a developer you don't have to make that distinction anymore like you just buil

an application and users to pick where the runtime is actually happening um it also means that you can treat all underlying chains as one unified system or one unified chain to you you just thinking about like there is one operating system um and of course different components have different guarantees but fundamentally you're not treating them as like heterogeneous components but you're rather treating as like this is one homogeneous development environment that you're working against um you also get unified intent liquidity and this is I think actually incredibly important because like I as a user don't want to go manually hunt for where I should be trading e for BDC I just want

to say like I would like to do some e for BDC intent and honestly give me anything out of like these six acceptable options maybe like ethereum and salana um or like and near and wherever I get the best price please execute me there if it's like in under 5 Seconds of execution right this is sort of the idea here that you can describe your trust preferences or like your constraints for what you want to do and then the system sort of exposes this to the developers so that solvers can then come and pick this up and go like okay like I have these constraints and I will decide to settle that guy on salana right now because this is going to be the fastest um and I think this is like the really big unlock because

fundamentally n of us are going to explain to any normal person outside of this room how any of this stuff currently Works uh like we're living in a ridiculous world where there's like 500 different systems that all work in very slightly ways and like my parents will never be able to interface with this level of complexity so we need something that is just going to be easier to understand and that provides some sort of unification um here also for the liquidity uh yeah you get unified defx and ux um and the other big thing is I was joking a little bit earlier around sort of the well if you run into a problem just put a server in front just like run a centralized infrastructure provider and like that solves some of your

problems with aoma you just focus on building your intents and like and like building your intents and applications and you don't have to worry about building additional infrastructure like you don't have to worry oh where am I running my rollup sequence as a centralized service so that because this is essentially what a rollup sequencer does it's counterpart Discovery right like where do you this how do I run this um you don't have to worry about oh do I need to run my own Sol algorithm because fundamentally this is also what a lot of these existing systems are doing you rather just describe your application let the market and the sort of rest of the ecosystem handle all the other complexity um yes so fundamentally a little bit of a recap so we're here moving from a

virtual machine to an intent machine um and this is I think like historically we've moved um sort of we moved move from this transaction Centric world with Bitcoin um to this virtual machine Centric world with ethereum and ethereum like things and in the future like with our Nora we moving away from this virtual machine Centric world to this intent machine Centric World um so fundamentally a VM I mean it's very simple I think everyone in this room should be broadly familiar with them it has an instruction set some program counters different layers of memory and importantly it runs program sequentially so an example here is like the son virtu machine the eum virtual machine you name it I mean at this point we have probably tens of different virtual machines that

are all um kind of similar and this also by the way includes all like ZK virtual ZK VMS they are fundamentally the same model except now we're running some proof over the actual execution Trace um and so what is an IM in comparison well it takes a batch of computes computes valid possible State transition and then selects one According to some criteria um and then updates the state transition uh update it's a system state with that state transition so you can imagine this if for example there is like six individual intents that um aren't individually setable right because they're partial State transitions but you take all six together you put them into a batch together um then they F form a fully balanced State transition

and then you can update the world State using this uh and so one ex one example here is and I think the only one so far alide we will see more and more of these things coming up I mean we saw this with the evm the evm was the first and then like a year in we started seeing like things that had slightly different trade-off criteria uh is the a normal resource machine uh the big thing to realize is because we all live in this like weird PVP battle where uh everyone's like ah my VM is better than your VM and I like well all your VMS are exactly the same they're just like different colors and I don't know why everyone keeps fighting over the different colored VMS um like the intent machine and theal resource machine like it needs to run on some virtual machine

um it is just it's a higher level of abstraction to make our lives easier it is not there to replace for example the evm um it does something fundamentally different to allow us to do new things um yeah so um the one thing I would like to cover here is a little bit that around limited data protection um and this is I think probably longterm the single biggest hurdle we will be facing as an industry um like I will never be able to convince my parents to use any of these systems I'm working on if their neighbors can see whether they went to the pharmacy or what they bought in the supermarket right like and I think this is just a very fundamental constraint because most people deeply care about their like local privacy

guarantees uh they don't want all their neighbors to see what they have in their bank account otherwise like the venmo thing where you could like had to pay someone to tweet about it uh would have actually worked in practice um and like everyone would be very okay with like open sourcing all their bank statements um one of the big things is when we say well we end up in a world of programmable data sovereignity um data moves to Edge devices so instead of currently we live in this world where like there's a large blob of data in the middle and then we all do some compute over this large blob in the middle this is a Super Batter and privacy perspective and b super inefficient from a compute perspective and so like we're moving to this world of like more and more data living in Edge devices and

intents are the right way to represent this like it allows you to represent Edge device computer essentially where you're like ah I have some local state and I'll produce some proof of this local state that is also referenced in a Merkle tree somewhere globally so that other people can agree on but we don't need to sync these Global State blobs anymore because a lot of the data just lives locally um so this is both usable useful for um scalability but it's also useful for actually data sovereignity because like data lives on edge devices and users make their own choices on under what conditions are they willing to reveal this and for example if you still live in the world of our intents are just limit or us I mean a you're wrong but B also um one of the interesting things here is

that you can never like be fully private if you want to trade you at least have to reveal to someone some intention about like I would like to do like Bitcoin for eth if you're willing to go if the only thing you're doing is like I am just I have this intent I'm telling you nothing about it you will have a very hard time getting this settled this is sort of just like in order to enable counterparty Discovery you will need to R some amount of information but this is what like programable intents allow you to do where you can say I have all these State keys in my intent and I'm only revealing two of them and the rest are committed to in ZK um yes and it also moves us away from the centralized counterparty Discovery Model to more of a decentralized counterparty discovery um where and I

didn't speak too much about this about composability of generalized intents one of the big reasons why you have to deeply care about like when you think about like we need an intense system we need an uh composable intent Centric system because if we if intents aren't composable like if I have a for B and someone else has B for C and A solver can't collapse into a4c it essentially means that there must be a mega solver globally that settles that like does all the solving whereas if you have composable solving it means that we can have thousands of solvers that each spe each like specialize in their own individual Niche to do like the step to do the collapse from Like A to B B to C collaps into a to c and then send it off to someone else um um yes and you can really like you're

are not limited here you can describe whatever you want uh you can say I would like to only trade some ether Bitcoin if the weather is nice in Paris and if the weather is bad in Zurich I would like to dump everything and go long like CHF um you're allowed to do this like this is like honestly this is up to you as a developer as a user what kind of complexity you would like to express um yeah and so what does this mean um intense inaps written with can be ordered solved and settled anywhere this can be settled on ethereum main chain this can be settled on any L2 this is fundamentally it's a user Choice um maybe a slight side note here as like protocol Architects we can describe actual protocols and like describe how

they work um it is very difficult to enforce a specific topology and so as a result Ana is very topology agnostic um like if everyone wants to run this on one single server run by the One World Government this is still the right architecture for this this is fine um you can run this Al like essentially like topology of the underlying Network ends up being much more of a user choice on whether many people have many different trust models for example um or like have many different local um latency preferences but in the end like AA is very topology agnostic uh users pick how they want to deploy it um yeah and this is what this ends up looking like you have users and developers on the left um then you have a bunch of P2P notes um that gossip these intents

around they can require some sort of payment they can allow users can stake here and then so for example get guaranteed bandwidth over time for example like 10 free transactions a month this is all like something that you can build on the application layer so that user experience has actually become very uh much more valuable and like much easier to do um then you have a bunch of solvers that say well we have all these intents what are the intersections here and then we those are settled somewhere uh according to where the intents sit um yeah just very quickly at the end um what kind of Novel daps can you build here well the thing that I'm very excited about is next Generation decentralized exchanges um that are fundamentally multi-dimensional so order books aren't predefined you

just have like a n-dimensional matrix of order books where every new intent adds a new sort of like slice into the order book and you can solve across that entire like nend dimensional Cube um you get fully privacy preserving stuff with this this can be fully decentralized because there doesn't need to be a single Global solver that does this uh you can build very cool things like credit clearing like if you like this check out cycles of money um rather than each of us wiring money around we can just like net this out at the end of the day I mean this is way more efficient this is why the banking system actually does this in practice at the moment um and then you can also build things like public signal uh to rather than being uh Supply driven be demand driven um yeah we just launched defet like a couple of

weeks ago um if you're a builder and you're interested uh hit us up it's cool uh the entire software is open source so please check it out also it's written in Elixir which is way funner to write than any other language if you haven't checked out that language please check it out it's cool um and yeah then test Nets are I think coming in the next couple of weeks like public test Nets and then main net uh towards the middle of this year um yeah this is it you can find us on Twitter you can read the white paper it's cool thank you very much [Applause] [Music]