BBW24 – Day 3 – KEYNOTE: From MIT to the Fed: Pioneering the Future of Digital Currencies

uh hello Boston hello quiny thank you for having me uh it's a beautiful day outside um and I I appreciate you all kind of hanging out here it says something about you know your interest level if you want to hang in check this out I promise you though this will be interesting uh this will be this will be uh something I think that you'll you'll really uh want to hear um so as mentioned I'm from uh uh radius and um uh we're working on some technology that I'm going to jump into but um a little bit about myself early on I I I worked at Circle from uh pretty much the beginning uh did a bunch of things there most most interestingly the cold storage uh system and then I started uh usdc so I was a team of uh three of us that that did that so uh that that's been now uscc is like grown but like it grew without me uh they they've done a much better job uh that I can't claim that I you know I I started USC anyway but I'm very very interested in digital dollars like how do you get a digital representation of a dollar and how do you put that on the internet so uh I went to the Federal Reserve next uh and uh I realized that you know you have uh dollars and there's 350 million-ish citizens of the United States but that's ridiculous right there's actually 8 billion users of the dollar potential user the total addressable Market is 8 billion not 350 million so what would it really look like to build a digital dollar that can really scale to that kind of a level and so um I did some work there there's uh this project called Hamilton that uh that I worked on we'll talk about that and and then a uh sort of the evolution of that called parac we'll go into that currently I'm at the uh MIT digital currency in itive the DCI I am a senior researcher uh senior research adviser I should say um and uh and then also as I mentioned radius uh radius is a startup started in the beginning of this year and we are working on productionizing uh some of this technology so let's dive into uh the the problem that uh that you know we've we've basically I've been hinting at here right so there's a lot of dollar denominated transactions in the world today right there's cash credit cards debit cards a other like lot lots of things turns out you know at the Federal Reserve there's like a you know thousand some odd PhD economists and you can ask them a question like hey Black Friday most transactions ever going on how many cash transactions are there and they have this deep network of of uh you know information that they can go into and they can actually tell you what the answer to that is we all kind of know what how much credit cards do and how much a does and uh this is what they said 20,000 transactions a second of cash debit cards are 60 you know credit debit and then 13 a basically you know if you all boil it down you round things out 100,000 transactions per second at Peak all right that's just payments okay so 100,000 trans transactions second let's talk about that for a second that's ridiculous right why is that well it needs to be orders of magnitude more when you have a bot that has some money in an account and it needs to move it uh you can see that happening a lot faster what about micro payments right if you could pay a penny to read a web page without ads you might do that and you do that a lot more than buying coffee uh and um you can imagine iot devices with with accounts Etc you get the idea basically you know tens to hundreds of millions of transactions per second are really what the world is going to need right and again that's just payments there are many other things you can do um we can get into that later so uh at the Federal Reserve and and MIT uh we came up with a solution to this uh we we worked on this project called uh project Hamilton the result of it is an open source codebase called open cbdc uh if you go to DCIM edop cbdc you can see that uh you can see this project we wrote a couple of papers about it it's been presented at uh conferences and it is uh also connected with an open source codebase called open cbdc that's on GitHub you could just go download it and run it uh so this was a system that we basically modeled after generally after how how Bitcoin works it's utxo based so I'm going to like skip a lot of technical things but there are few you need to know utxo is uh unspent transaction output forget what that means basically there's a bunch of bills and each bill has a name on it and the name is the person who owns it and you can take a collection of these bills and update the name to whoever else and now they own it that's the way utxos works that's different than the way your bank account Works your bank account balance is whatever and it can go up or down right there are positives and negatives to that um we built this system we scaled it out turns out it kind of ran into a wall uh around 100 100 like 120,000 transactions a second that we could not get Beyond and then we realized hey the the way to do this is just to not materialize a Global Order of transactions I'll keep coming back to that materializing a Global Order of transactions basically commits to If I buy a hot dog in California and you buy one in Boston does it matter who bought it first not really it matters if we use the same money so we need that guarantee but it doesn't matter if you bought it or I bought it first if we basically bought it roughly in the same second kind of doesn't matter so that realization by the way if you can come up with a reason for why that that is not true like why you actually do need to know globally I would love to hear it I've been like racking my mind trying to figure out why that is oh hot hot doging contest well you know you still need to buy you probably buy in bulk so that's one transaction I don't know but they they're all sort of local you just can't use the same bills you know um so we built this system that doesn't materialize a Global Order of transactions and that turns out to be able to scale linearly so what that means is like you want to go faster you throw more computers at it uh and and we we basically did that as some negatives not really programmable and uh it's fairly inflexible we said okay great we have this system that goes really fast I think we did 100 1.7 million transactions a second with this one and then we said you know what we kind of want to be able to do other things uh why don't we try this and why don't we try that and like every time we looked at something we said okay we have to destroy the thing cut it all the way down to its studs and rebuild from scratch basically the whole system sounds like a waste why don't instead of us doing that let's build this new thing which we call parek which is a virtual machine Runner so you can run any virtual machine right so we picked the evm why not right it also runs uh Lua and you can do other things as well you can like literally any VM you could do x86 if you wanted um but we stuck with the ethereum virtual machine turns out there's all this you know all these contracts that are already written wouldn't it be great if we could like run those at scale uh and so sure enough we did that with uh parek and you are actually able to take a smart contract from the ethereum uh ecosystem off the main net or whatever and run it unmodified with parek literally the only thing you do is change the url in metamask from public ethereum mainnet to parseek and it would work um it doesn't commit to a global order just like the other uh system that we built uh that is a big deal because un let's rewind back to remember I said utxos these are like little bills and each bill has a name on it and we're just updating the names right there is no limit to that you know as long as you're swapping one Bild from one person to another you don't run into a problem but what if you're trying to use accounts right so we have a balance we're going to push it up or down let's say we're we're paying uh you know payroll at some point like if we're going to run out of money I got to stop and that's a problem you know you're not going to be able to know when to stop if you kind of parallelize everything so we we have a solution to this um this system as as uh as it it shows we got up to 118,000 transactions a second and then we kind of like say okay it's fast enough it's fast enough to like do the things that we want to do so we'll we'll try some other things we'll get back to that okay so how did we do that well um I'm going to skip a bunch of this because I kind of want to you know get to questions as soon as possible because those are always interesting basically we have sets of shards all the data is you know put into like a a shards it's broken up to shards when I say shards don't think blockchains these are not blockchains these are key Val Val servers and then we launch these agents and they operate they run the VM that's like they would run the evm for example so your user wallet sends a transaction to one of those agents and then the agent finds whatever shards are involved in this transaction and coordinates a two-phase commit to uh settle that transaction it can do that critically in parallel with thousands of other agents if you want um as I said I'm going to like skip a bunch of these things the only thing I want you to look at on this slide is the the stuff all the way on the left so there a transaction it starts you uh you know it gets a ticket number and that's a way to resolve conflicts not going to go into that and then the evm is running and it's just basically doing read and write operations every once in a while if it likes it can take a read uh lock that it has and upgrade it to a right lock but basically it just does that till it gets everything set the way it wants and It prepares that says okay this is the new these are all the new rights I want to do and then if you know if we haven't run into a a lock contention it commits and then you know we discard or we discard and we're we're done so uh so that's how that works and I could go into it more deeply believe me I can uh but I won't uh right now so as I mentioned uh I'm working uh at this startup called radius and uh radius the the idea here is to rewrite that for productional like how do we really you know do that and the the code base that we have by the way again Parc is also open source it's on GitHub you can just download it and run it um you know that codebase is a research codebase meant to run a research load so if you want to try you know I don't know like uh an amm like Unis swap at a million transactions a second you can go and do it you know find that you're going to have a problem doing that but but you you could um so in rewriting for productionize migration right we have to do we have to add all these things that it you what is key migration well if you have like 10 shards and then you know you get a lot of load and somebody you know starts to put a lot of data in your system you need to add shards so you need to be able to move shards around and you need to be able to delete them and stuff like that you need to be able to autoscale the agents add more of them when they're needed and shut them down when they're not some of this is uh you know relatively well known but the key migration stuff isn't uh there's some interesting ideas about uh flexible key replication like you could do like keys that are heavily hit on a very a server that only does heavily hit keys keys are extremely large on other uh servers that are optimized for that there's a bunch of things you can do um critically we can can also operate this system as an L2 on ethereum that brings some unique challenges uh for example one of the unique challenges is well how do you create a fraud poof if you don't have a materialized Global Order of transactions so we we have a way to do that um so so there's some interesting uh things there um so I said this would be interesting okay well let's talk about it like back uh you know if you if you look at what's avail ailable right now like I could imagine salana is relatively fast where they 60,000 transactions a second let's just say that they could do uh you know a million with the with fire blocks or something like that they or um fire dancer rather uh you know Mega eth could go a lot faster probably you know maybe reach uh a million at the tops right uh monad is another one they they could these are all you know basically scaled evm same same idea but they're doing it a very different way uh that could probably get into that range too but they can't go to 10 million or a 100 million or a billion transactions a second so there's really only one way to do that and that is not to materialize a Global Order of transactions I keep saying it right well what do you do well you actually materialize a local order of key updates and so that's that's how we handle that uh problem so now you would imagine if you're Unis swap right you want to run a amm you're going to have to lock a root key over and over and over because you got to make sure there's enough Assets in the pool to handle whatever transaction is going on you can't just like drain the pool down to zero and say whoops that wouldn't work so you do need to order those um so we have a way to do that but there's also this implication that as smart contract development gets more and more uh you know uh like con concentrated on it's going to focus on well how do you parallelize the load and so I think a lot of that has to happen um uncontended load like you know this is kind of you know what what we're basically doing any number of keys can be written in parallel right so what we're what I just tell you about a test I did this morning so uh we fired up a test this morning where we ran an erc20 contract that sent money from one account to another account pretty much a random account they're not they know don't contend and when you do that you know you can spin it up we spin up uh I think uh for that was about 4 and a half thousand computers all around uh the United States and uh we are able to I think that test was like 1.2 million transactions a second um we've done tests that get a lot larger than that um up to the you know 20 25,000 computers uh all running together and and we can reach much higher loads as well it it just literally scales linearly by adding machines um the last thing I'll talk about is contended load I kind of hinted a little bit of this don't want to get too too technical but like heavily contended Keys have to be sequenced we have a way to do that we can pre-execution [Music] don't block uh they can go they can be uh parallelized while while that's happening um and then uh we as I said again we don't commit to a global order uh you know we we uh certainly do not want to do that for non-c contending transactions uh because that that would really you know create an issue um and then as I mentioned before you can have shards that are that are really fast or shards that are really big or shards that are you know uh kind of transient in nature um shards that are private even which is which is a cool idea um some of this some of these slides came got a little small but um you get the idea um so there's some interesting ideas and I'll jump to questions because you don't have a much time um like I said permission keys what if a a smart contract could actually hold a private key so it could actually sign things that would be really cool yeah you can actually do that um we can run any VM not just not just ones made for smart contract when you run VMS for smart contracts they're made really interestingly they're made to you know maintain a just a collection of keys that just that smart contract and can control and that's what you want because you don't want some random person to just update their balance to whatever so that's a valuable you know uh uh design space but you don't you can violate it you can actually go out of that you can use things that don't normally happen like you can you you can Harvest Randomness you can directly read the internet without an oracle there's some fun things you can do um the last one that I'll mention here is a utxo system and a really essentially a balance based system can use the same state if you think about it a utxo is a create read delete that's it you can make the bill you can you can uh you know read the bill and you can delete it and then it gets recreated somewhere else that's it there is no update with a balance that's a create read write delete um so you could have two systems that use the money in this in different ways one thinks of it as a utxo and the other thinks of it as a balance um this is actually possible uh with systems like this those are some interesting ideas thank you very much I'm going to jump to questions because they're always fun and we're always running out of time so uh let me uh do that thank [Applause] you hi Anders uh great talk you never disappoint thanks for coming um I earlier in the week we heard a presentation from um uh from a a prior presenter about the cbdc uh work at the Fed so I want to kind of Channel your thoughts on this one um they really thought that when the the the when the the US government launches their cbdc it'll be the ultimate kind of spying and like there will be no privacy uh almost what I think we hear what CH the China cbdc would be like I'm sure you've heard this before it's not my bingo card that cbdcs would become political but here we are in 2024 they are I'm curious given your your work at project Hamilton and other places what's your view on that type of perception that's out there both that's probably myth maybe some fact you're on the inside what can you tell us thank you yeah so uh good question I will of course preface this by say certainly no question about it I do not speak for the FED I never uh just just so that's clear these are my personal opinions um uh yeah so it turns out that the technology can supply unbelievable things unbelievable things like you would not think that they are possible however that probably is not common knowledge um and people rightly uh are concerned and so I don't think it's a question of whether or not the technology can do it in my opinion it's probably possible to make something far more private than than you know than what what we really even have today even with private uh uh cryptocurrencies however I don't think people will believe that you could do it so I don't think you can um you know that's my personal opinion I don't know I I uh I would love to be proven wrong yeah uh great conversation what are your thoughts uh maybe are your inside knowledge about developments of offline capabilities with stable co uh not stable coins But Central Bank digital currencies and have you heard anything as well too collectively from different nation states in terms of how they're going to um use maybe a basket of currencies or Commodities or something else to actually um back up the Central Bank digital currencies with the underlying asset yeah so uh I think two separate things one is uh you know what how how do you do an offline currency um you know there's there's generally two ways to do that one is to have trusted execution and you know that kind of thing and the other is more like ecash chiai and ecash where um the actual transfer it you know just happens person to person but you need to reveal at the end to to uh collect your money in sense you can't resend it so respendable offline money you know you're you're stuck with the hardware and if the hardware gets hacked you know we've seen some pretty big Hardware hacks very recently um you know UB key and and there's some other the the I think one of the Intel uh trusted execution environments recently uh got hacked as well so there are problems with that you can limit size of transactions and size the amount that you're likely to hold to to mitigate that but it's imperfect um so uh offline is is particular ularly difficult uh you don't actually know you have the money until you you go back and check online generally um so so that's a hard one about backing it I don't know do you need to um you know that's essentially you know tokenization stable coins you know the they're all exactly the same thing there's a big pile of something there and then separately there's this token that we track uh you know you kind of have to trust that the the bond is maintained and that's that's the weakest link in that chain so yeah yeah I you know I'm the technologist not the I get to I get to get out of jail free card because I'm the technologist um yeah I don't know I mean you know it really does depend on uh you know the the reputation of the issuer uh I I'm I'm not an economist I'm not I'm just a technologist so I don't really have very much for you yeah uh we have one or two others and then uh and then we got to cut it I think so yeah yeah um hey can I do this one first oh there sorry yeah hi um so one of the last parts of it you basically were talking about um uh a key inside of a smart contract or basically the ability for a smart contract to sign transactions yeah it's something we think about a lot and uh wondering in that context do you believe that that would be the future on like a usability side or is there any room for eoas kind of coming in the in the next the next generation of all this um what's EAS uh externally owned accounts like a like a normal yeah yeah yeah yeah um right so um uh I don't I don't know we'll see you know sort of which way it goes the uh having having some private information sort of requires that you have made a centralization uh tradeoff that you might not be comfortable with um so you know I guess it depends um one of the interesting things about uh the the parseek system that that we built is you know in a sense you can think well we're sort of running all these shards but we don't have to we can take one and like have another entity run that you know like let's just say DTC runs it and all the Apple shares are are on that Shard so they could automically update some let's just say usdc transfer for one share of Apple atomically oh it's possible I I think that's likely how it'll play out less less likely that the the um uh you know that something would hold a private key directly and that would you know because you kind of have to trust whatever that service is for anything it signed yeah