Guy Itzhaki I FHE Rollups and Co-Processor in Action I Fhenix I Pragma San Francisco 2024

[Music] without further Ado let's welcome guy from Phoenix to tell you all about FHA and co-processors please give him a big round of applause thanks how is everybody doing um I'm the last presenter before lunch I hope to keep this entertaining so by the time we end the conversations are still here um so as introduced I'm the CEO of Phoenix we provide L2 uh rollups and co-processors uh based on FH I'm based in Portland every time I come here I envy your weather and your basketball team um and with that I'm going to start with the basics and we're going to start building it from that f is kind of a complex topic and yet um I'm I'm I'm sure you'll enjoy it so um starting with the basic and I'm sure a lot of you know but there is one thing that's common to all blockchains out there which is the fact that they all need to do two things one is they need to verify that a transaction is valid and the second is they actually need to verify that a contract is executed correctly and in order to do so they need to reach consensus over public data but what it actually means is that if you think about it what blockchains does best is provide correctness of computation and there is literally no privacy in the blockchain space all data out there is is public so what does privacy even mean and I think many people have that a bit confused when we think about privacy we think about two things one is anonymity or pseudonymity and the other one is confidentiality so thinking about anonymity it's the ability to hide the identity of the person that does the transaction whether that's the sender or the receiver there's been several projectss trying to solve it um like tornado cach Etc and they not all of them ended up well but what we're mostly focused is on a part which I think is is more interesting and that is confidentiality which means the ability to hide the ident the content of the transaction so if you think about uh Bitcoin it's the ability to uh send an amount and have that amount actually uh encrypted but in the more interesting use case of ethereum it's the ability to do confidential smart contracts meaning that the smart contracts can work on arbitrary data so uh confidential smart contracts have been researched for quite some time there's been a lot of resource invested in it and typically they're being sold by using what's called privacy enhancing Technologies or pets for short I will say that even though there's been a lot of research invested up until recent L there hasn't been yet a silver bullet to solve that but we definitely see them becoming widespread and practice and what does a confidential smart contract actually means it means that from a user standpoint they can encrypt the data they send the transaction the transaction is being executed on the blockchain the state gets updated and the output is returned to the user and the user can actually see the output of the computation so I'm going to give an example and the example I'm going to give is like seal bid auctions if you think about the marketplace and you want to make a bid uh you want to make sure that the bidding is happening in a confidential way so let's see how the confidential smart contracts Works in a SE bid auction example think about a user that wants to bid on an nft or a user that wants to bid on uh um an llm model that will be a Marketplace for LM models at some point so we have Alice Alice will send her bid in our case 5 e but with a confidential smart contract the bidding itself is actually private and then we have Bob Bob can't see what Ellis bid so he'll bid his offer in our case it will be 10 e and so on and so forth and when the contract time is done and it's being executed there is a computation that's going to be done and the winner will be announced on the public and that's how we achieve a financial benefit for the higher bidder to get um to to win so it all sounds great why hasn't it been sold by now and there are several cryptography that have tried to solve it I think you've all heard about zero knowledge and NPC and tees will walk briefly about them so uh zkps clearly are widely being used in our space today mostly for scaling they've been proven for quite some time but zkp we have to remember they're a proving mechanism they're not a computer ation mechanism which means that they can only offer single user privacy the use case that we've just described of like seal bit auctions cannot be done with uh zero knowledge uh or it's very complex to execute it with zero knowledge uh MPC multi-party computation sorry um enables you to run computation from multiple uh U participants without them sharing the data with one another the benefits of MPC it's very fast but MPC requires is a significant amount of communication between the different parties which is why you're seeing MPC widely used with custodian wallets when you have a limited amount of participants but it doesn't really help you and it's very slow to use when the number of participants become big and the last one is trusted execution environment tees you can compute inside a secure Hardware the it's it's fast it's programmable but the problem is that you need to trust the hardware and they're prone to side Channel attacks and when there is an attack on a hardware it becomes quite complex to fix it so what is the solution what is the endgame H it's it's FHA F morphic encryption um yesterday I was in Salt Lake City and I met the guy that invented fol morphic encryption Craig Gentry first article was done in September uh 2009 so I think we're almost like over a bit over 15 years for that uh FH is amazing what FH does is enables you to do computation on data while the data is encrypted everything that we've talked about encryptions we know that when you encrypt data it's Ming L and you have to decrypt it if you want to execute on it with f you can actually execute and compute over the data while the data is encrypted and the reason that F works is because what F does it maintains the mathematical structure of the data that you encrypt so you can still compute on over that and the response will be encrypted and only the user who encrypted the the data originally can actually decrypt the the the output of it so F has been out there for 15 years always been associated with two significant problems the first one is complexity um we are at this point in time where FHA is no longer complex it's very easy to build on FH uh we have FH VMS developers can build in solidity use the same tools Foundry whatever they use today to build an ethereum they can use they just decide what they want to encrypt and decrypt very easy to use uh f um companies like third web that we just saw support uh Solutions around FHA uh the second part of it that you can write contracts that part of it will be encrypted and parts of it won't be encrypted and that's on its own is is quite simple so I say like somewhere like 20 minutes you can already have like a developer starts to build a contract with f it's the the the up is is super quick so I said two problems the second problem that's been associated with f is performance FH is complex it's complex it takes time to compute however there's been a lot of progression around FH in both fronts one is just making FH itself faster and faster different schemes that are out there are coming and improving FH performance and the second part is FH is computation bound so better Hardware means better performance and there are already optimizations for gpus fpga and even as6 that are being built for FH coming up in early 2025 so even with what we have today FH is definitely ready for applications today and just gets faster and faster so the talk is about F rollups in co-processor so let me briefly walk you about how we use FHA and an FHA rollup and it's no different than a general generic rollup that we have out there today the users they encrypt their transaction inputs and they send it to the rollup we then have the sequencer the sequencer Aggregates all of the transaction and executes them using fully morphic encryption and we have the settlement that is done on ethereum so you have all the ethereum security guarantees that associated with it if there is a request to decrypt the data there is a component called a threshold Network which is taking the network key and sharing it among multiple participants for security the threshold Network computes decrypts the data and if there is a need to do Post execution validation like fraud proofs there is also a mechanism that enables you to run fraud proofs on top of FH so FH rollups already out there uh you can you can test it um and and and check the the Phoenix test net now what if I have an L1 or an L2 or an L3 and I want to use FH computation and I don't necessarily want to use an FH rollup the solution for that is a co-processor for those of you who may not know what a co-processor is co-processor enables you to do computation offchain so um high level flow is that you take a computation you send it to the co-processor the coprocessor runs the computation sends you back a response and and the reason that you would trust the co-processor is because the co-processor would also send you a proof that the computation was executed correctly and these proof can be in a shape of a ZK proof or some kind of a te proof or a crypto economic proof that shows that this was executed correctly so co-processors are coming uh and FH co-processor are actually coming in a couple of months out there and the beauty of co-processor is now any L1 L2 L3 that wants to implement encryption and run computation using polymorphic encryption can actually do that and the flow is that the chain itself would get the contract would send the contract with a computation to the co-processor the co-processor itself would run the computation that's basically step four that you see out there if there is a need to decrypt it will be sent to the decryption Network the proof itself in our case is done through igen layer and then the result and the proof is sent back to the network itself and you get encrypted computation in a very easy and and and and efficient way so uh what I'm going to do now is take a couple of minutes just to talk about a couple of applications because it's all great we have technology technology is is is transformational but what can we do with FHA uh so some of the use cases I would say are what I call like the immediate suspect so very easy to implement clearly confidential transaction is a mechanism that works very very well you want to send a transaction from user a to user B you can use f to make sure that the the transaction itself is uh indeed encrypted similarly with voting for Dows that you want to maintain a voting uh that is private and not public seal bid auctions as as I shared for Marketplace or or for me uh um the right to do me extraction is something that uh is is has generated some interest uh compliancy pool but for me these are all like applications that make a lot of sense but we're also starting to see interest in getting into the more complex and interesting use cases one example is decentralized Identity the ability to get identities on chain and the reason that you can put a user identity on chain is now because it's private and once you have decentralized identity you can actually now start doing on chain credit scoring and that's something that's uh missing today and when you have credit scoring on how well the identity and the user behaves now you can do efficient Landing uh um solutions for that so all of that really generates like this um this Snowball Effect that really brings maturity and innovation and use cases into applications out there today and then uh lastly I'm going to take a few minutes just to talk about uh confidential AI just because I know there is like a lot of interest in it and most of the conferences I talk people ask me about like verifiable Ai and Ai and where AI is with blockchain and with FHA so um when we think about blockchains and AI I typically like to look at it as we have blockchain and Ai and we have ai and blockchain um BL chain and AI or where AI can make blockchain better we're starting to see uh solutions that actually enable through machine learning applications to build decentralized applications um remember this is still a small ecosystem and the more developers and even if developers are using machine learning based solutions to build application that makes a lot of sense um audit smart contracts build smart contract using AI uh uh is is something that we can see AI helping blockchain but where things become more interesting is we're actually using the mechanism of FH and blockchain to help AI based applications so a lot of what's happening today in in the industry is we have what we call corporate AIS right all of the the the machine learning models are maintained by a couple of of big participants um but there is a trend and we're starting to see the shift for decentralized models where we're going to going to see more and more models starting to build by uh um de centralized users how do you build the models how do you bring data which is private data but you can still put it on chain and use that um that's a vehicle that will become super interesting about either making the models decentralized but also making the models better by bringing data that you cannot bring today because the data that you have is private so how do you bring the data on chain and how do you use that to train the models f will play a significant role in that space as well uh another space that's kind of interesting is what I call agent to agent transaction and the future and I don't think it's like in the far future but maybe in the near future users will have agents that would help them end their day-to-day uh online tasks and having these agents to agents conversations can happen as transactions and these transactions can run on the blockchain and they can be protected through FH and then lastly is data authenticity as we grow more and more and we as we see more and more AI based applications it becomes challenging to know what data is real was generated by human or was generated by Machine learning blockchains do a phenomenal job in doing authenticity and we can start S Marketplace for authenticity of data and S bit auctions for Marketplace and just protecting it with f uh makes a lot of sense so uh quite an exciting uh future uh there is a lot of traction around FH um if you're interested you can come talk to me or Kyle or just follow us on Twitter that's it [Applause] [Music]