How Polyhedra Network is advancing cross-chain privacy with zero-knowledge proofs

0:00I'm Ashen Addison from the Cryptocoin Show and today on blockchain interviews we have Eric Vand COO of Polyhedra on the show to talk ZK interoperability a artificial intelligence of course and machine learning. Uh and I am excited to dive into the mixture of all of this uh today with someone who has been deep in it for a while. Eric so much thank you

0:23so much for taking the time to come on the show. Thanks Ashton. really uh great to be here uh and excited to uh to talk about all these all these topics. Um yeah, so I guess I'll I'll give a little intro. My name is Eric Vland. I am the chief strategy officer at Polyhedra. And I know strategy can be uh somewhat ambiguous. So uh I was the first kind of

0:45go to market uh hire at Polyhedra. Since then the team has expanded dramatically. We've built out our business development, our community, our marketing functions, but really kind of oversee all of the go-to market side of the business. For those that aren't familiar with Polyhedra, we are a infrastructure uh company and we focus on building uh trust uh and scalable

1:12systems uh specifically on AI and blockchain. Our first product that you uh mentioned was was ZK bridge and it is the leading zero knowledge secured interoperability protocol. While building that we built our own zero knowledge proof system called expander and since then have launched uh EXP chain testn net as well as have uh released a number of developer tools uh

1:39for verifiable AI and ZKML. Amazing. Uh, and I want to dive into the the ZK bridge a little bit more and maybe like a Eli 5, you know, where it's like there's a lot of people that are deep in it, but there's also a lot that are still maybe they're holding Bitcoin, they're holding Ethereum, they see all these other blockchains, um, and they're trying to tip their toes into getting

2:05into other chains and exploring different applications. So with the ZK bridge, how does that work and and what is the the value that it provides in getting into these different ecosystems? Sure. So if you're, you know, you're just dipping your toe in blockchain, uh you probably have Bitcoin, maybe you have Ethereum, but you're hearing about all these other uh blockchains that are,

2:26you know, purpose-built for one reason or another. Maybe they're have uh super fast throughput or therefore, you know, securing IP. um and you want to experiment with those blockchains. So, you need to get assets onto those chains. Um and in order to do so, typically what you'll do is you'll you'll find a bridge. Um and bridges can work a number of different ways. The way

2:48most bridges work because blockchains are built to be very good at securing data within that isolated environment and making sure that that data is accurate and correct. But what they're not good at is verifying data between different blockchains. And with a traditional bridge, usually what will happen is data will be requested. You want to read data from one blockchain

3:12and you want to share it with another blockchain. And what most bridges do is they introduce this kind of validator network in the middle that um you know signs off and says the data on blockchain A is this uh and you can you know record it on blockchain B. And so a typical use case would be like I have burned you know 100 USDT on blockchain A. This uh validator network signs off.

3:39Yes, this this action did in fact happen. Now you can mint or release 100 USDT on blockchain B. Mhm. But you still have this trusted uh intermediary. You have this network of validators that you you know have to trust in order for this system to work. And you know, I think working in web 3, many people believe we want to get to as trustless of an uh

4:05environment as possible. And our founders were studying cryptography and specifically zero knowledge. And then they thought that this was a really good application of ZK. So instead of having to use a validator set, you can create a zero knowledge proof that says that proves the data on blockchain A is this and then you can verify that proof on blockchain B and you no longer have this

4:28third-party um kind of centralized piece of the bridge. And so really the advantages to ZK bridge over a lot of the alternatives are just better security more uh you know less trust assumptions and then there are some kind of uh other properties of ZK that make it uh very powerful. So you can pack a lot of you know transactions into one ZK proof. So uh it is also you know

4:57incredibly cheap. Um, so there there are a lot of benefits to kind of ZK Bridge and and I think we're starting to see zero knowledge really being top of mind in a lot of different use cases because of these characteristics of data privacy, scalability, um, robustness in in terms of security. Definitely. Yeah, I appreciate that explanation and and the differences

5:24there. And with the uh with just say I'm on one chain, I I want a bridge uh or I want to use the ZK bridge. What about the user experience? You know, that I think that's been one of the uh barriers to entry of decentralized exchanges or decentralized finance versus centralized. Um people, you know, they of course there's a there's a majority of people still that are like they don't

5:51mind using a trusted third party or a verified other source if it means that it's more secure or if it means that they have to do less stuff because you know it's it's it's faster, it's more convenient and like convenience is key in you know the age that we're in right now. So, have you guys streamlined that that process of actually making it sure it's not you don't have to rely on third

6:16parties, but it's actually a lot more work. Is that the case? It's not. No, it's not more work. From a user experience standpoint, it is um you know, probably the same or better than uh the alternatives that are out there. really at at Polyhedra although we do have our own kind of token bridge what we're really focused on building is that infrastructure layer. So while you can

6:39build on top of what we call kind of uh our general message passing framework uh you can use this for all sorts of use cases. So you can use this for fungeible token transfer, NFT token transfer. That's actually one use case that is very um or NFT transfer. This is one use case that's um very well handled by ZK. You can also use it, you know, to um build crosschain oracles. So it it's not

7:08just limited to just token transfer. It can pass any arbitrary data between blockchains. Um, and the user experience should be, you know, no different or or better than what what already exists out there. I think, um, you know, one of the big kind of roadblocks that we had to overcome was when people are bridging assets between blockchains, it it can be

7:35very anxietyinducing. they they click the submit, you know, bridge assets and then the assets leave one wallet and they're in the Ether and they're waiting for it to show up on, you know, on the other blockchain and and especially if you're transferring large amounts of funds, you can have a lot of anxiety uh because you're like, I hope it worked. I

7:57hope I didn't do anything wrong. And uh so the reason it took so long for a zero knowledge secure bridge to come is because ZK proofs take a lot of compute and generally or until recently took a lot of time. And so what we had to do was really start at the very bottom of the stack and build a proof system and our proof system is called Expander and it's now open source that had incredibly

8:26fast um proof generation times. And so we can now generate a proof of the entire Ethereum full consensus in under 10 seconds. And so once we kind of overcame that hurdle and zero knowledge proofs were no longer the bottleneck for these cross-chain transactions, now we can give the benefits of security and scalability um without the downside of you know

8:54having to increase the time for the data to move between blockchains. That's great to hear. And could you give some examples? You know you mentioned Ethereum there. If I wanted to move some stable coin USDC from Ethereum main chain to some layer twos like optimism, arbitum base uh but and then and in another example maybe I want to move to other blockchains that aren't EVM

9:20compatible as much Salana or or Binance. Is all of that possible or are you mainly focusing on the Ethereum ecosystem? Yeah, so we definitely started with EVM. We support about uh 35 different networks now. Most of them are EVM. We do uh support you know Binance uh B&B chain um OPB&B um and we also are doing work on nonvm as well. So we've been working on

9:50Solana for a while. There are some other kind of nonVM uh chains that we're uh talking with or integrating with. So our goal is really to enable interoperability across the entire blockchain ecosystem and not just EVM. Um but there are complications that get added in uh when you're moving from EVM to non EVM. And so it just it takes a little bit longer from a development

10:15standpoint. Definitely. That's good to know. And you know in the last year there's been such an uprising of actual functionality and utility of artificial intelligence into blockchain uh with AI and machine learning. Um is there a play that adds extra value with AI into polyhedra as well? Yeah. So I mean these uh you know AI agents have been kind of

10:40a hot uh topic for conversation and there's a lot of teams that kind of see blockchain as and AI agents as uh this area that kind of blends the two technologies or works really well because AI agents especially if we're going to have them you know go out and do commerce for us or pay for things they're going to need some sort of financial rails to do so and crypto is

11:05very a very good system for that. But these agents also need to, you know, if you have an agent isolated on one blockchain, it's not going to be very powerful because like you said, there's all this other stuff going on and there's all these reasons and and benefits to different blockchains. So, it is important for AI agents to have, you know, verifiable data on different

11:27blockchains. It is important for AI agents to be able to execute actions across multiple blockchains. And so we see ZK Bridge as a really powerful interoperability tool for AI on blockchain. There are all you know we can't uh have you know all of these different kind of isolated AIs on different blockchains. We need them to have kind of a multi-chain or omni chain

11:49set of functionality and access to all of this data. And ZK Bridge really um can empower that. And so we are talking to a lot of AI teams that want to tap into um ZK Bridge for that. I would say one other area that we're really focused on is because we had to build this uh highly performant proof system that was incredibly fast at generating ZK proofs.

12:16We started to think you know what other areas are is real time proving a necessity and one of the uh areas that we are spending a lot of time and resources on is verifiable AI. So how do you know you know that an AI agent uh belongs to a certain person or a certain entity? How do you verify that the mo uh you know the AI that you're

12:43interacting with is you know actually utilizing the AI model that you expect it to uh and that it's not just you know a human behind there um kind of feeding it outputs or how do you how do you know that the AI uh output generated was actually generated from the inputs that you gave it. So these are all you know black AI has been a black box and we've

13:08kind of accepted at that you give it inputs it gives you outputs and you can choose to trust those or not. Um but with zero knowledge machine learning we can add some transparency to AI so that you can have confidence that you know this output was generated from this model based on these inputs. uh and that opens up a whole new kind of world and we think really uh

13:34increases the trust between humans and AI as AI, you know, starts to become used in everything that we do. Yeah, that's a great point, Eric. Especially because in blockchain, these permissionless blockchains, you can just go look at the source code, you can see everything. You know, there's no back door. and with the the LLM systems that most people are using, you know, you

13:58don't really know what the input and then there's something in the middle and then the output. Um, and I feel like if we're going to make a decentralized version of that or at least we're going to use it with blockchain systems, this is not transparent at all. No, I mean, and and you know, it's not efficient for us to run these AI inferences often times onchain. So,

14:20you're going to need to run these inferences offchain. you you you know you you give your inputs it goes to some server or some you know a cluster of GPUs those inferences are run and then those outputs are come are are brought back on chain and then used in you know whether it's some DeFi protocol or you know whatever it is and you know anytime you kind of move offchain and then come

14:45back onchain there's this period of time when you don't really have that full kind of supply chain or you know of what happened. And so ZK proofs really help bridge the gap between onchain and off-chain because you can still do this compute um offchain when necessary, but you still have this proof that you can verify and post onchain to prove that everything, you know, was legit.

15:13Yeah. No, that's a that's a great uh use case. And so when you for example the the ZK bridge using the ZK technology there and then using uh ZK for verifiable AI did you guys use you know part of the same uh like principles for it or there's feel like the use cases are slightly different. Yeah, I mean at the end of the day it's kind of proving

15:40um that data or computation is uh is what you say it is. So from a blockchain perspective, you're proving that you know the data exists on blockchain a uh and you're proving that cryptographically with ZKML. you're proving that the AI inference happened um accurately and and as expected and that it wasn't tampered with. And so at the end of the day, it just comes down

16:07to kind of proof of computation or proof of data. Um the the the key theme and I I think you're right like these are two kind of very they can seem very far apart in terms of uh applications. The reason that we we picked them is because we we believe that these are both applications where that speed is very critical. If it took,

16:36you know, uh 10 hours for you to use ZK bridge to move assets between chains, you'd probably go and find an alternative. Similarly, um, if it takes, you know, days to generate a proof of accurate AI inference, it's probably not going to get used. And that used to be the case. It used to be the case that it would take, I think, uh, for, you know, two

17:03and a half days worth of some of the benchmarks to generate a proof of of AI inference. And by applying expander and GPU acceleration, we've now gotten that down to, you know, an hour and a half. And with optimizations uh and improvements that we're making in our ZKML stack, we're hopeful to get that down to, you know, sub second by the end of the year. And so now you can

17:30actually generate an inference and in the in that same you know output when you give it the when you give the user the inference you can also give them a proof at the same time. And so it's these use cases where you know real time uh proving is necessary where our technology really uh is beneficial and so you know there are other use cases where ZK is is really popular. ZK

17:56roll-ups are one of them. ZK roll-ups, you don't have to post the transactions from the layer 2 to the layer 1 in real time. You know, if you did that every 12 hours or every hour or every 30 minutes, it would probably be satisfactory because, you know, it's not super important for that to be real time. But for use cases like bridging AI inference, and there's, you know, a

18:18whole self-driving cars, there are, you know, use cases where you really do need that real time verifiability. Mhm. Definitely. And with the verifiable AI, you gave an example that the inference uh LLM that that we're using, it might be offchain, you know, like most of 99% of what people are using, but I've also had a lot of discussions in the past months about decentralized

18:45AI, and maybe that's not completely onchain, but they're trying to, you know, pry uh the the the all the people that are using uh open AI, you know, into something that is a little bit more respectful of their privacy and um the the the information that it's collecting. Um is the verifiable AI technology that that Polyhedra is creating is it applicable to both you

19:13know centralized AI so to say and the the decentralized AI as that grows as well? Absolutely. Yeah, it can be applied to across both uh fields. So, it's really kind of um it it doesn't matter whether it's it's onchain, it's decentralized AI, it's centralized AI. At the end of the day, what we're trying to solve is, you know, is there a need for this this kind

19:42of verifiability and this proof of of accurate execution? Yeah, I love that because most people are not using decentralized AI. at least we're still in the infancy of it. Uh but is there an incentive for you know the the centralized parties that run the centralizing AI to actually implement uh ZK or verifiable AI into it? Absolutely. I mean so right now there's a lot of

20:12talk about you know open- source models and and for our initial um kind of release of ZK PyTorch which is our our tool uh that developers can use to turn a machine learning model into a verifiable model. We we do a lot of our benchmarks around open source models. But you know, one really interesting use case is you think about all of these applications

20:39where the the the model itself is really the special sauce. It's like the proprietary uh piece of technology that a company has developed to to make money. And so that could be like a a financial model that outperforms the S&P. That could be a health care model that is really good at diagnosing particular diseases. Now, in these instances, it doesn't make any

21:09sense for the developer of this model to open source it because that's like that's how they make money as a company. But at the same time, they need to figure out a way to license usage of that model out. And typically what happens is you know their customers give them the inputs and then they run the model and then they give the outputs to the to the customer. Now in this

21:31interaction you really having to trust that company that they are actually running the model on your inputs giving you the outputs. And so we're coming back to this idea of trust right you're you're you're trusting the company that you're interacting with. Mhm. Uh and you could eliminate the need for that trust by turning those models into verifiable

21:54models. And so you know you give now as a user you give the inputs to the company that you're you're hoping to you know use their model. They give you the outputs along with a proof that proves that you know this model the same model that had uh you know these 99% accuracy on historical back tests is being used. And so you don't actually have to give

22:16the details about the model. you can just give a cryptographic proof that's unforgeable and it can give you the confidence that you know you actually are getting the outputs from the model that you expect. And so this is uh kind of an emerging field called machine learning as a service that I think is going to become growing in popularity as we go from these large kind of general

22:39purpose models to uh maybe more niche special purpose models. Yeah. No, I uh I'm excited to see the growth of these small language models as more and more information that you know you could go you could be studying one specific topic for for decades and like I don't think the LLMs that we have right now can get that deep into it but eventually all of

23:05that's going to be uh input into the AIS and and probably tokenized somehow and and all this stuff. So we'll see how that grows. Now with the verifiable AI uh technology where exactly is polyhedra at with that right now? Yeah, so uh we we published a paper on it about two months ago. We had done some kind of preliminary uh you know internal benchmarking and we

23:31just recently are um kind of publishing our recent benchmarks publicly open sourcing ZK PyTorch open sourcing um a lot of the work that we've done on the ZKML side and making that available for developers. So any, you know, we're we're con uh constantly in talks with different teams that are building in AI, both kind of more traditional web 2 uh

23:59and you know, web 3 blockchain AI builders. Um, and we're kind of getting this first cohort of applications that believe in the importance of verifiability and want to integrate that into their stack. Mhm. Yeah. I would love to um have that ease of use of using an AI agent that had verifiable AI to then give it the you know tell it tell it my agent to use the ZQ bridge to

24:28transfer my assets uh to you know the chain that I want so I can start doing DeFi or it can start doing DeFi for me. So exactly that's the that's the key right. I think um you know Visa published a really interesting article the other day about how we're kind of moving towards this uh world in which AI is going to go shopping for us. It's going to handle

24:51all paying our bills for us. It's going to automate a lot of this kind of annoying um you know thing these annoying tasks and and that we have to do every day. At the same time, you know, when we went from a cash society to a credit card society, that introduced a new vector uh for fraud. And then when we moved from, you know, kind of traditional swipe your credit

25:18card at a terminal to online, that introduced a new vector for fraud. And now as we go from kind of credit cards to AI agents that may have access to a crypto wallet or to a debit card or whatever it may be, now we're introducing another uh vector for fraud. And so I think it's important that we continue to focus on while we are becoming more efficient um and these

25:43tools are definitely you know kind of giving us back the luxury of time by allowing us to outsource a lot of these tasks. they're also, you know, making it more complicated and making it more likely that there is, you know, opportunity for something malicious to happen. And so we need to continue to invest in these tools for trust and safety. As much as we are investing in

26:07some of these like really hyped um you know technologies that are going to improve our efficiency, we have to make sure that we are taking safety and trust seriously at the same time. Yeah, that's a great example, Eric, because right now in uh in inferencing and AI, you know, you can ask it, you know, plan a trip out for me, but you can't actually ask

26:32it to book the hotel, like take my personal information, take my credit card, book it for me. I still have to go to the website and type that all that in myself. Um, and then that's where the verifiable AI comes in, ensuring that if I give the AI my personal information and my credit card number, someone can't just hack into that and like book other

26:51stuff with my credit card. Um, I think that's really important and I I feel like there's going to be an aha moment where where people realize how easy it is once the AI agents can actually book stuff for you or uh make purchases on your behalf that it's going to go crazy. Yeah, it's going to be great. It's a really exciting time, you know, and and I don't want anyone to think that we

27:17are trying to stifle growth. Like we don't think that it's not like drop everything, stop everything before we need to solve this problem before we can continue to, you know, uh, improve our AI models and our AI systems. That's not the case. I think just what we're trying to do is educate people on the importance of verifiability on the importance of safety on the importance

27:40of trust and just you know similar to I think here's another analogy you know similar to the internet early on before we had HTTPS you know we were interacting with websites we didn't really think too much about security um we probably even put our credit cards into some very insecure websites and very insecure forms Um there were probably some issues with

28:06that. Probably some people lost some money. But eventually, you know, in the background, we were developing this HTTPS technology that allowed for websites to develop uh you know, a form or a way to take data inputs and encrypt it and make it more secure. And now every website that you go to has HTTPS. I think you'd be it hard for you to go to one where where where it didn't have

28:30that that added layer of security. And so we kind of see the progress of AI taking a similar path where there will be these um very kind of high lever um highly uh critical tasks that where if something went wrong it could be you know disastrous. Those will be the first ones to implement ZKML and verifiable AI. But you know 3 5 10 years from now

29:02once you know that the the technology has improved it's become extremely easy to implement. It's extremely fast to generate the you know it's it takes milliseconds to generate these proofs. It's going to be like HTTPS. Every AI you interact with will come with an inference. That inference will come with a proof. That proof will be instantly verifiable and you won't even think

29:22about it anymore. Yeah. And you mentioned earlier about uh the developers with the the ZK PI torch and like how developers can get involved and help build this and experiment. What about for people that aren't developers or you know maybe they have done some DeFi and they experimented with different chains. They're not developers. Is there a way for people to

29:44contribute or participate in in the ecosystem right now? Absolutely. So we have um you know a ton of like I said we're our primary focus right now is kind of onboarding the next generation of verifiable AI applications. So one I would just say you know follow us on on Twitter we're polyhedrazk we are announcing new partnerships on a daily basis. um follow

30:12us. We also do our own kind of live uh stream show called the Polyhedra Show. Usually about twice a week, we often have ecosystem projects on there. So, it's going to be, you know, really leveraging some of the applications that are building with our technology and and experimenting with it in those ways. Um, and then yeah, I mean, if you're a developer, it would be, you know, going

30:40and and reading reading the documentation on ZK PyTorch and and leveraging that for building out your uh your verifiable AI application. Yeah, very cool, Eric. I appreciate your insights into where AI is at right now, how we can make it more secure with ZK and verifiable AI. I'm looking forward to seeing that in the AI platforms. as I continue to get deeper in it and learn

31:06the learn the tricks and uh ensure that everything's safe and secure and I I tip uh h like hats off to to Polyhedra for making something that I feel like is going to be ever more important. People don't really realize yet, especially when we get transactions coming in and personal information. Um it it's it's going to be so important for AI to have this technology in there. So looking

31:29forward to seeing how it all develops. Um, I I I'll leave the link to the socials you mentioned, uh, the developer docs for, uh, ZK PyTorch as well and the Polyhedra site in the show notes below for anybody that wants to, uh, learn more about it and and follow along on the socials. And, uh, I'm wishing you and the team all the best moving forward. And I would love to follow up

31:49again in the near future. Absolutely. Would love to come back on in in six months and and show you guys all the different ways it has been applied. We've got a lot of exciting stuff in the pipeline. And then yeah, I mean, I guess if you if you have any questions that came out of this podcast, you can also just reach out directly to me. I'm Vland uh on Twitter or X. Um I'm always happy

32:11to interact with people in the community that want to ask questions and learn more. Sounds great. Thank you so much, Eric. Thanks for having me.