Exponential: A Nexus Podcast
Exponential is a Nexus podcast about people, code, and capital.
Exponential: A Nexus Podcast
Episode 30: Selective Disclosure
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Hank Korth has watched a lot of distributed systems cycles come and go.
A professor in the Department of Computer Science and Engineering at Lehigh University — with an appointment in the business college and as the lead at Lehigh Blockchain — he has been thinking about networked systems since the late 1980s, when his database research examined how independently administered systems could transact across a network.
On this episode of Exponential, he joined the show to talk about where his students and research team are concentrating their attention now: zero-knowledge acceleration, onchain privacy that can coexist with regulation, and the agentic era beginning to take shape beneath the headlines.
Welcome to Exponential, a Nexus podcast where we talk about people, code, and capital. I am Daniel McGlyn, and in this episode, I talk to Lehigh University's Hank Korth on Zero Knowledge Acceleration, the case for private transactions with compliance proofs attached, and why the agentic era will need both. Hank, uh, welcome to Exponential. It's great to have you on today. Thank you very much. It's a pleasure to be here. Yeah, so I'm super curious to talk to you about your work at Lehigh and um some of the research you've been doing and what's going on and what your students are uh interested in and what they're they're talking about. But I think the the first place to start is maybe you could introduce us to Lehigh Blockchain and tell us about what what kind of organization that is and what kind of work you're doing there.
SPEAKER_01Okay, so it's really a lot of things. Um I mean, obviously I think this conversation is going to focus on research, but we also have a substantial educational mission. We have a course in our computer science department that is hardcore systems aspects of blockchain, the obvious things you'd expect, data structures, consensus algorithms, cryptography, smart contract coding, et cetera. We have another course that is open basically to the whole campus that focuses on application, policy, regulation, um, punctuated with a lot of guest speakers. My um flagship guest this year, I would say, is um Hester Purst, who needs no intro to this audience. And um I started that one back in 2018 and um stumbled upon a Lehigh alumnus who happened to be CEO of Coindesk, Kevin Wirth. Um, since the sale of Coindesk, Kevin has joined me as a co-instructor. And it has been phenomenal to have students have this experience of a tech nerd and someone so immeshed in the industry. Back to the education theme, we have a course in our master's in financial engineering, which is of course a very mathematical program, but then focused on the finance side. So that's the education piece, the research piece. We're going to go into those projects later, so I won't um elaborate in um huge detail here. Um, but um they range from kind of performance acceleration, from benchmarking, zero knowledge, to more application-focused things in the domain of privacy and a variety of other things over the years. The um projects are largely undergrad focused because I need trained people, and they are hard to get in this field. So I have my own students. I have kept several for a master's. Um, convincing people to stay for a PhD is virtually impossible given the nature of the industry and the opportunity cost involved. But the master students have done quite well, and I'm really proud of our alums. Um we have um the co-founders of a decentralized exchange, Dolomite. We have um a uh lead developer of Aztec's domain-specific language for zero knowledge, NOIR, a student on um Zcash's tachyon, students at a variety of other um startups, but also larger enterprises, I mean Google's blockchain work, uh several students in Oracle blockchain, and several in the blockchain data analytics firm Inca Digital. So people out in, I think some some pretty neat places.
SPEAKER_00Yeah, and it's really cool to hear that um, you know, one of the one of the ways that kind of um the whole blockchain industry and crypto industry matures is to really develop that pipeline of of talent and engineers and and even policymakers and people that understand the technology in and out so that you know when they are either joining the industry or they're regulating the industry or whatever the case may be, um, there is that that kind of um that knowledge and understanding uh of what it is. I I do remember in the early days of crypto, like the the biggest um one of the biggest bottlenecks was that, you know, people um making decisions about what should and shouldn't be allowed, you know, just didn't have that that background or familiarity with the technology, which uh you know obviously is is um something being addressed now. And it's really cool to hear uh what you all are up to at Lehigh. Um and one question I I was kind of curious about, because you can I I feel like you can uh you know always tell like maybe what what is the current trend by by talking to uh students about uh what they're interested in and and like you know what they're preparing for for the future. And uh so through that lens, I was kind of curious like what your students, you know, maybe some of the research projects that that your students are leading or are really interested in um right now that you think will have an impact on the industry um, you know, down the road.
SPEAKER_01Yeah. And in your whole topic of students and the influence, my alums have been very involved and um will advise students coming along so they have a broad understanding of the industry. So we've tried to do, hey, here's some little narrow academic thing, but rather to do work that people would care about. And you know, a lot of our work has been around the area of zero knowledge. As I would expect many of your listeners know, zero knowledge proofs are not only very important now in the business, but are computationally hard to generate, though very efficient to validate. Right. And so we've done a bit of work on acceleration of the generation of proofs using modern parallel architectures. I have some colleagues doing research in that space. So we have the um domain expertise there. My students get to collaborate with those students, and that has been a uh a very fruitful thing. Um interestingly, there, while you see here, you know, GPUs, parallel architectures, you think parallelism. But it turns out that memory is a huge barrier here. These proofs are memory consumptive, and we actually can't always exploit a full GPU in terms of parallelism because we can't fit the stuff in memory, leading to multi-GPU algorithms. So that's one area that you know certainly is having a lot of impact. We've done a lot of work looking at performance benchmarking. Um, initially, some work with an old database colleague of mine who was then at the National University of Singapore on benchmarking layer one chains with standard workloads to avoid some of the self-serving claims that otherwise come out in industry uh exhibits and so forth. We then took that to a next level by looking at layer two and asking what seems like a simple question, but it isn't. When is a layer two transaction final? So if you're a you know a dogmatist, it's not final till it's on Ethereum and final there. But if I trust my L2, as soon as it accepts it, I may be willing to um you know make the business decision to take the risk. Right. So what this means is there are many metrics, and then we look at the question of all these metrics we can imagine, which ones can be measured. And students have you know worked with several systems, you know, all along those lines. Interesting. And then a current thing, um, and I suspect we'll dive deeper into this later on, is the whole issue of privacy. Yeah. Um, which is I think a huge issue in our field. And we have a a quite cool project in that space, but maybe that can come a little bit later on, because I have a lot to say about that.
SPEAKER_00Yeah, no, I'm I'm curious. Um I think um privacy is is clearly at top of mind right now when we're talking about um regulations. And uh obviously that that seems to be gripping the headlines every day when we're talking about the crypto or blockchain space of like uh clarity with regulations. And so one of the reasons I wanted to talk to you today is is because of your research and privacy and and just um to ask the question like are is privacy and and this uh push towards clear regulations, are are those compatible? Um what do what do you think about that?
SPEAKER_01Um for me, definitely compatible. Um you know the um new regulatory regime clearly states that privacy is not inherently a crime, something that was questionable in the US not that long ago. Um but that doesn't mean privacy is absolute. And that doesn't mean I should be necessarily you know free to um you know send stable coins to my favorite terrorist or something like that. But how do you um deal with those things? And of course, you know, one answer is the absolute privacy of zero knowledge, but then we have the issues around things like tornado cash, then the flip side of all of this is that if everything is indeed totally open, it's open to the world. It's not just that somebody can find out. If I pay my neighbor with crypto, they know my entire financial past and can follow my future. And so the project we're working on is um a framework, the high-level goal is that I provide a transaction that I keep private, but attach to it a proof of regulatory compliance. And compliance here is intentionally a parameter to our framework. It could be, you know, something like membership set of a set or non-membership in a set, which is along the lines of what Privacy Pool and Rail Gun do. But our vision here is that we try to incentivize regulators to publish regulations in logic or for trusted institutions to make that translation. Now, with a regulation in logic, going through some various um translations and optimizations, we can wind up with code in a domain-specific language for zero knowledge. Our choice there for our project is noir for a number of reasons. A, it's a good choice. B, my master's graduate is one of the lead developers, which means I have a fantastic help desk. In our internal slap. But it is a phenomenal choice. So then beneath noir, many provers can be attached. So again, our framework is not tied to a specific proof framework. But the the idea here then is that I can submit a transaction to an application proving whatever that application demands. And so presumably it would know that you know I'm good, the recipient's good, we're following all the rules of you know, say the US, or if it's global, the US, the recipient domain, etc. And then it knows that it has something clean to go forward with, but it might not know details about me. For example, I might prove I was KYC'd by some bank. I won't tell you which one. You don't know where I banked. You know, and um so there can be different levels of privacy, and obviously I hope for very strong levels of privacy, but the premise here is that if we as um blockchain people demand absolute privacy, we're not going to interact with the traditional business and finance world. And I talked finance, of course, this pertains too to business, supply chain. You and I are supplying the same thing. We want our deal private, but I want to know that you're not, say, using slave labor or something like that.
SPEAKER_00Right, right. Yeah, I think um the the component there too is once you have that proof attached to a data piece of data, then it can follow that proof like through its life, right? So then you're like you were the point you were making before is you know, the the cost of validation um is minimal. The the cost of creating the original proof uh is the computational cost, but then you know validating again and again um is not. And so then it becomes interesting when you have that chain of you know, that chain of proof, I guess. Um and and and we're not duplicating computation or duplicating um, you know, trying to comply with multiple regulations. It's like once once you can show that you know whether that transaction or that data is regulatory compliant, it um, you know, you can you can just go about your business from there. Exactly.
SPEAKER_01I mean that's right on target as an observation, but it's actually non-trivial because if you imagine these rules and logic, there's no saying that the EU and the US, let's say, will write in exactly the same way. And so there are optimizations at each level of the stack. Optimizing over regulatory logic for what I need to comply with and what I've already proven. Yeah, and then optimizing over the set of proofs that we have to minimize what I need to prove incrementally. And then for what I need to prove, optimizing at the level of the um domain-specific language for ZK. So there's an awful lot of stuff that goes into the stack here, which makes it fun.
SPEAKER_00Right. And I also like that point you were making of like um, you know, we talk about um sort of crypto primitives or blockchain primitives being programmable and composable so that, you know, we can interoperate and anyone's free, you know, you can have permissionless building and um, you know, features of uh self-custody and all these things that make blockchain so important. But um the the point you were driving at was um you know, why can't regulations be also programmable and composable? Um and if you had a system in place like um, you know, uh a proof that was attached to data, maybe that it makes that more feasible. Um but your your point is taken. That it's one of those things sounds simple, but is actually probably pretty complex when you take it.
SPEAKER_01And even just getting those regulations. I mean, well, it may just sound like hope, but um we've had some conversations, you know, like with the uh enterprise Ethereum allies. Like, yeah, this is something we could get behind. And um the there's interest in this, not only for say my use of it, but also in an AI domain. You know, if I'm going to have AI helping me comply with the law, well, if I can have well-defined input, that's going to be a whole lot better setting in that regard. And ultimately, we can imagine a competitive ecosystem here. You know, so a jurisdiction that doesn't publish regulations in a proof-friendly format will be a a less um desirable place to do business. And so we could imagine jurisdiction jurisdictions competing, but also I could imagine stablecoin ecosystems competing on privacy. You know, after all, they're not going to compete on the price of a dollar.
SPEAKER_00Right. Yeah. Yeah, that's interesting. Um yeah, privacy is a feature set, I guess, right? Um exactly. So the the other thing I wanted to ask about um kind of your your research related to to layer one architectures. I think the the another trend we're seeing kind of at a high level is um new kinds of layer one architectures. That's what we're building here at Nexus. So it's kind of top of mind for us. But um I'm kind of curious from your research and your perspective, like what what are you seeing um in that domain right now?
SPEAKER_01Well, I have some new things, some old things. I mean, I'll start new. That um when you have a clear application, you know, as you know Nexus does, you know, then having a custom environment for that can be you know tuned, fast, and you know, wonderful. Obviously, you have to ask the obvious question, well, are you sufficiently better than you know Ethereum or whatever else to justify it? Where things get challenging is when there's interoperability involved. Um, and I mean we're seeing that now in um the Ethereum ecosystem, that it was thought there was this ideal solution with layer two. As an application, I pick my assets, pick them up, move them to layer two, and then work there, no interference from anybody else, and bring them back. But this very new development that they're calling the Ethereum Economic Zone is looking to create interoperability across all these things. And this is going to run into some really hard challenges. If the domain of interoperation is well curated, it can work. But there's work, this goes back to databased research I did back before my hair turned gray, like the late 80s, early 90s, where we were looking at database systems that were connected over the net, but separately administered. You know, you know, imagine, let's say, a travel ecosystem with each travel vendor having its own database, but allowing cross-system transactions. And in that environment, we found that it is very, very difficult to ensure proper transactional properties, proper data consistency, and do so without having very draconian restrictions on the kinds of transactions that could be tolerated. And you know, we have papers back from then about both possibility and impossibility results. And so I'm reading a lot of the current stuff in this space and even earlier stuff from the ETH ecosystem on charting, with a, I'll say a slight chuckle, like I know where this is going. And um, yeah, and now I'm seeing that again. So um, you know, really excited about the challenges here, you know, the strong possibilities, and um, you know, yeah, love what's going on there. I mean, for for you, you're doing it in the finance space. Um a number of L1s are doing this now in support of AI ecosystems. And that also looks very promising. You know, we were talking before about payment systems. You know, my AI agent can't open a bank account. It's going to be paying, presumably with stable coins. Now, there's a lot that has to be done there. And, you know, my agent has to prove to your agent that it truly represents me, that it has the resources to do what it's going to do. And then we have to be careful about how much my agent reveals about my personal information. It may need may need more overall than it's going to reveal to you. And this again falls into this whole idea of um attaching a proof to a transaction, now not for regulatory purposes, but for agenda AI purposes. So all these things flow together in what I think is a truly exciting way.
SPEAKER_00Yeah. I think we're in this like interesting moment of history where, you know, people have been concerned with and talking about privacy on the internet for as long as the internet's been around. And um and then zero knowledge proofs came along, um, you know, first as a privacy technology and then also as a computational scale technology. And then um, and then you know, blockchain came along, and uh we were using it as a force of decentralization and um, you know, permissionless um access. And it's kind of interesting how they're all blending together, and then you know, this other wave of um we're just seeing the rise of AI coming incredibly quickly, um, you know, generative AI, um, consumer finance AI, all kinds of um artificial intelligence, and and they're all combining into this um this kind of push-pull or this like sort of um this interesting tension, I guess, uh where they you know potentially there's there's a way to hold everything in balance and still um see progress that's meaningful. Yeah. And so I'm curious how you think about all that stuff that's in the L1 thing.
SPEAKER_01Oh, sorry.
SPEAKER_00Yeah, yeah. Yeah, no, go ahead. That's that's where I was gonna like just how how you see that combination of these these new L1s, this kind of privacy technology, kind of all shaping the the future of AI. What is that?
SPEAKER_01So I mean we talk about AI in terms of agents, but of course, all the hype about AI is largely around LLMs. And this is another layer one story. Um, at the moment, these LLMs are these monolithic systems, you know, we take a FARB and turn it into a data center, and that's my um you know new LLM. Um, but you can imagine a situation of having small language models that fit on relatively compact machines. And interoperate through an L1. Now, an example of that is um BitTensor. Um, and um, we do happen to have a research project with them looking at um their infrastructure and their incentives in getting the players in this space to both collaborate and compete with appropriate sets of rewards. And then, you know, like any such complicated L1 ecosystem, you have to look at not just how it ought to work, but how you protect against some players who um might not necessarily have the best intentions. Right. And that I think is uh a very important direction, you know, circling back to the whole privacy story. Um we're looking to have with LLMs a repeat of social media where a handful of tech companies basically own everything. You know, my Facebook post is not mine, it's Meta's. It's about me, but it isn't mine. Yeah. And with a decentralized AI, we could have a situation where the knowledge I share remains mine, and when I get responses back, I know something about the source. And not just some net ID, but something about the nature of the process that got there. You know, the the biases or focuses of the model, et cetera.
SPEAKER_00Yep. It's interesting too. It seems like we haven't learned that lesson from the first, from like the web two world where all of this information we're creating um through the process of using these AI tools, uh, how valuable that is. And and like we haven't we haven't figured that out, like how to uh, you know, that was one of the promises of of uh crypto and blockchain technology is is being able to, you know, yeah show that digital ownership and and make that happen.
SPEAKER_01But outs of other domain for that. And one that will become bigger as we start to have AI models doing medical diagnosis. Now, those may necessarily be outsourced by the medical practice, but now how do I know that that outsourcer actually ran the model as opposed to running Hank's Stupid Model? It just says you have cancer and doesn't bother reading the input. Um and so I now need to have a proof that the provider actually use the FDA approved model and do so while keeping the patient data private and keeping the um service provider's proprietary parameters private. Um this happens to be a um research interest of something that I'm looking at with um of um uh very surprisingly our business dean who actually is knowledgeable in this space, working on this space with another colleague um who's now at UCSD. Yeah, that's cool. But I do find it remarkable. You're asking at the beginning about Lehigh blockchain. We've got a business dean who actually um understands what's going on and is an active participant.
SPEAKER_00Yeah. It is cool to see how um these technologies start to become important and and touch all other kinds of domains. Um and and we're starting to see that, you know, people always talk about mass adoption. What does it look like? Well, it looks like you know, when when blockchain is kind of uh becoming the rails for um all kinds of other processes, whether it's you know medical records or um you know various other things. So super interesting. Um one thing I I'm curious, and I kind of asked this at the beginning, but um but when you're looking at um kind of trends or things that are underreported, or or even I'm sure at being, you know, having the the the opportunity to work at uh an academic institution um and a research institution looking at these big questions uh before they become kind of like industry concerns. What um if there's any like trends or unreported things or or just wrinkles of interest that you or your students are kind of pursuing that um, you know, maybe they will never make it to industry, or maybe they're they're just like interesting um research questions. I'm just kind of curious uh what you all are looking at.
SPEAKER_01Actually, in that row, I'm gonna start with a um misreported trend. Um academia is that with the media saying there are no jobs in computer science, yeah, parents are telling their kids to stay away. Okay. We are having no problem placing blockchain students in jobs. Um and um the fact that they have a specialty, actually, any specialty would be a difference maker. And certainly here there are relatively few schools producing the set of students we do. And um I think one of the um things that I mean we certainly try to um instill in our students is a broad sense of the industry. And so one of the things that I don't think we look at enough and kind of what shows up on X or other things in the you know, the random feeds that come in, is the interplay of perhaps unrelated things. And so our you know, multi-faceted focus, not just on tech, but also policy and application, is that our students actually understand why they're doing what they're doing, where the impact is, and then can turn that back into um addressing problems. Yeah, so uh all our computer science do capstone projects, and we have several of those. We have one with Oracle Blockchain, we have one with Yuma on that decentralized AI thing. We have a um really cool one with a coffee plantation in El Salvador in cooperation with a local firm, Chainbytes, where we can pay the workers in crypto, because they've got wallets, track the beans through a supply chain. So at the end of the day, here's a bag of beans, and here's a proof that these beans were picked by workers who got paid and got paid a fair wage. Now you can easily imagine all the tech in there, but then you think also about all the interface points along the way to make this happen. And so having students get exposed to this full systems thing is um, I think that's really where some of the big issues come in. That we have underplayed user experience, connecting everything together, and then when we finally have a product, making it something that the end user can find credible. And that that has not been a strength of of this industry to um a large degree. Yeah, I agree with that.
SPEAKER_00I think the the other interesting thing you mentioned is just this idea of systems thinking, it seems like that is going to be the skill set of the future, right? Where we have we have these tools now that can um work with information, manipulate, and do a variety of tasks for us. But I think if you can think in systems and understand how you know different things need to change at different points, um be be a good skill set to have. Yeah. Um okay, last question. We're almost at time here. Is um if you oh go uh if you had uh to make a big uh bet or bold prediction about the future based on your research right now, where you sit at at Lehigh University, and um, you know, also your your years of research uh in in database architecture beforehand and and all of that, like what what what are you looking at or what what do you see on the horizon that's kind of interesting to you?
SPEAKER_01I'll give you a two-letter answer. Okay, okay. And to address the grandparents of your audience, I could um reference a 1967 movie The Graduate and say ZK is the new plastics. Nice. And uh from the smile, I think you know what I'm talking about.
SPEAKER_00Yeah, I I know that. Yep, I do know that reference. I like it.
SPEAKER_01I don't think your audience generally will, but I thought I'd go with that anyway. Just uh hey, I've I've got the gray hair to go, even though I work 20-year-old soldier.
SPEAKER_00Right, right. No, that's good. The the new plastics. Um perfect. Well, that's a great way to wrap today, Hank. Thank you so much for your time. It was great hearing uh what you're up to and and the work you're doing and uh the work some of your students are doing at Lehigh. And um definitely keep an eye on all that. And uh we'll have to talk more about ZK uh soon and and as these issues of of scalability and privacy and regulation and AI all kind of combine into this interesting mesh. Uh so thanks again.
SPEAKER_01And thank you so much. I look forward to continuing to follow your work and your team's work.
SPEAKER_00Um appreciate it. Thank you for listening to this episode of Exponential. We'll be back next week with another show featuring people, code, and capital. Please like, review, and share Exponential wherever you listen to your podcast. And be sure to visit app.nexus.xyz to see what we are building.