Apr 10, 2024

p0x Labs co-founder and Manta Network core contributor Kenny Li joins "First Mover" to discuss the intersection between artificial intelligence (AI) and blockchain.

Video transcript

I'm not too concerned about developer adoption of A I from the complexity perspective, but I am a little bit concerned about the practicality of all of these, you know A I apps that will be floating around. Um But I'm sure that we'll find uh you know, specific use cases that would be hopefully valuable in the space. The intersection of Blockchain and A I has become all the rage with the two emerging technologies being at the forefront of discussions about innovation. I'm on a journey to figure out how these two technologies can work together to solve real human problems that we are experiencing today. Kenny Lee is the co-founder of Pox Labs and core contributor to Manta Network Manta Network has partnered with Celestia Aura and Hyperbolic Labs to bring fully on chain access for artificial general intelligence models to Manta Pacific, the native layer two block chain of Manta network. Now given how early we are in the development of both Blockchain and A I and how many challenges there are still to solve when it comes to implementing these technologies into existing business models? I wanted to ask Kenny how long he thinks it will take for this future to come to fruition and what's the real benefit that it's going to bring to developers and ultimately end users. Let's take a listen. The protocol is presented by the Stellar Community Fund Accelerate your web three project with stellar funding, Kenny Lee. Welcome to first mover. Thank you very much for having me here, Jen. Really appreciate it. I'm really excited to have this conversation because we're talking about the intersection between Blockchain and A I and that is just the conversation that has taken over the headlines recently with all the new A I developments. Why are we talking about this? Why is this important? Yeah, I think it's extremely important because I think regardless of if you're developing building applications in uh you know, traditional centralized sectors or decentralized sectors, right? You definitely want to make sure that you have the right tools at your disposal. And right now those tools are all related to A IAG I specifically, right? And it's, it's just like why does everyone use chat GP T? Right. And so I don't think that uh having these tools limited specifically to web 2.0 is any benefit to, you know, the Blockchain ecosystem. So the more we can kind of drive this adoption, I think the more the better off we will be, you know, I've heard a lot of folks say on the back of this, you know, uh surge of A I um that A I is what is going to help Blockchain ecosystems become more mainstream. A I has the potential to solve some of the challenges that developers are experiencing right now. Can you unpack that a little bit more for me? Is that actually the truth or are we, are we just like hanging on to the hype a little bit? Yeah, I think that we're getting a little ahead of ourselves there. Um This kind of reminds me of when Ethereum first started making waves and the promise was decentralized Uber. And I think, you know, up until this point, we still haven't really seen decentralized Uber. I I do think that, you know, we we do have to um kind of condition or temper our expectations here with A I, right? For from my perspective, you know, A I is definitely an important tool to have at a developer's disposal in order to build the applications that they want to build. Uh But at the end of the day, right, this type of adoption um on both the developer side, as well as the user side, it boils down to user experience and value, right? And it doesn't matter what tools you're actually using as long as you can deliver that value, as long as you can deliver that experience. I mean, I think that's what actually defines the success of an application. I think A I tools are there to help bring that benefit to some extent. But it's not necessarily the case that every single application will benefit or even needs and all these sort of A I implementations. Let's talk about your news. Now, Manta Network Celestia aura and Hyperbolic Labs is bringing on chain access for A G I models. Now, before we get into the specific news, just set a foundation for me. Give me a brief explanation on what artificial general intelligence is. Yeah, sure. So, I mean, that's just a very fancy way of saying um sort of generalized A I and I guess I'm just kind of throwing the words, they're scrambling them around a little bit. But the idea here, uh and the, the key difference here is, um you know, I was actually in the A I space uh a previous generation ago and previous generation actually was like 2019. And so the, the interest that was built around that generation of A I tools was all around image recognition. And so uh specifically like facial recognition and even things like, you know, um Amazon shopping where you go and buy groceries and they scan your items and you know, exactly what you're buying based on image detection and all that stuff. And so, you know, that wasn't generalized A I because it had one specific purpose, its purpose was to look at something and tell you what it is, right? It's like the Silicon Valley episode where it tells you whether it's a hot dog or not. A hot dog, right? But A G I has extended beyond that purpose and can fit basically in multiple different personalities, multiple different shoes, multiple different use cases. And I think the most, um, the most common example right now is chat GP T at the end of the day, is this A I powered chat bot? But if you're able to train it with the right sort of data sets with the right sort of information, then it can become an expert in any specific industry that you need it to be in. But at the end of the day, it's still, you know, more of like a jack of all trades before you make it into a master of some specific topic. And so that's, that's kind of, you know, a G I versus more traditional sort of A I built purpose specific. All right. Now let's segue right into bringing this fully on chain. What does this unlock for developers? Yeah. So bringing this fully on chain, I think is extremely critical because, you know, there's, there's kind of two ways to approach this and we are approaching it more of that second way. The first way to look at it is essentially OK, let's build all these models on chain. But the reality is that, you know, the, the existing sort of models that are out there already have had such a head start that, you know, it's not impossible, but it's highly improbable that we would be able to replicate these models perfectly to that level of accuracy, to that level of intelligence uh directly on chain, especially in a cost effective manner. And so you know the the question now is how do web three developers and web three applications take advantage of all these really powerful A I tools that are available to the rest of the internet, the rest of the world. And so by by bringing those on chain, right, I think the possibilities here are extremely, more extremely but limitless because what you are really seeing now is you are putting these types of developers in the web three ecosystem on a same level playing field as the developers that are building these A I applications outside of web three, which I think, you know, isn't entirely possible if you wanted to purely build those models in house. Now paint a real world use case for me, paint a picture for me, what kind of decentralized apps could be developed that solve real human problems today by bringing A G I on chain. Yeah, that's a really good question. And I think that it's also a matter of um timeline, right? I think a lot of these sort of problems that we really want to solve are not going to be solved overnight, right? So for example, one of the sort of um use cases here is around health care and medical diagnosis. Um And II I worked in the healthcare space as well. And you know, from someone that's been in the healthcare space, I know that the problem with technology adoption and health care is one of operations and compliance. And so, you know, I'm only preface this by saying that, you know, these are definitely future use cases in a world where, you know, all these other sort of issues are resolved that aren't necessarily technological bottlenecks, but rather human and compliance bottlenecks. Uh But with that being said, right, I think that one of the most powerful use cases here is being able to actually own the model, own the data that can recreate the model. Um So let's say, for example, you're a um a, a AAA book publisher and you want to publish A I generated books, right? And the uh and I'm giving a very contrived example because the, the the industry itself is not important here. You want to publish very uh you want to publish A I generated books and you have this specific formula for publishing it so that the books themselves read like the, the Harry Potter series, 80% Harry Potter, 10% Twilight. And yeah, all all all the classics, right? So, so you know, having that perfect formula. Now, if you built this on Chat E BT, right, who owns that data, who owns that model, who owns the actual implementation of that? Right? If Chat EPT decides to say, oh, we're going to cut your API A access one day or chat EPT goes down et cetera, et cetera, right? Like how, how do you recover that? Right. How do you build your next series without trying to retrain that entire model from scratch? Um Whereas in Blockchain, right, you have that ability to keep AAA fully self custodial record of every single piece of data, every single action, every single interaction that has led the A I model up to a certain point to be that expert in that field that you need it to be. And so in the event that these types of models go down, or there's some type of question of ownership because it is fully self custody, you're able to have that business continuity that you may not necessarily have in the web two space. And so that's, that's a little bit more broad. But I think that highlights sort of a um uh a scenario that extends beyond one specific use case and one specific industry and just talks more about the power of having that on chain in general for any sort of purpose. OK. And now a question that I've had as I kind of contemplate A I and Blockchain is that the training of these models requires a lot of data. And when we look at multiple different blockchains in the ecosystem, there is this big scalability issue. When there's a lot of data to be processed, we see the chains slowing down. Um processing slowly gas fees going up. Uh How do the two technologies work together in the future, do you imagine? Oh, yeah, that's a really good question. I think there is um a lot of sort of approaches to try to resolve this, right? I think, you know, when it comes to training uh A I models, there is a lot of factors at play even on the hardware level and especially around the hardware level, you know, like we see projects like hyperbolic and other, you know, uh deep in, I guess category of projects that are specifically focused on lending out GP US through a decentralized ecosystem of peers. And uh you know, that that's kind of how they're resolving these types of training issues by, you know, sending it directly to the GP US themselves rather than having the data on chain and process on chain, which I, I personally don't think that's going to be happening in the next like, you know, 5, 10 years. Um But, you know, even these solutions, I think it still takes a lot of um uh very heavy architectural design thinking because, you know, when you have all these GP US, it's great. But when you have all these GP US separated across the world, then latency becomes a huge issue. And so there, I, I think like we're still very much in the early stages of trying to figure out how to do training in a decentralized manner. I think inference uh which is, you know, actually using the model is a little bit more um or light on the compute, which makes it a bit more practical. Um not without its own scalability issues, but uh definitely on the training side, you know, I I don't think that this is going to be purely happening on chain anytime soon. Alright. And Kenny, I know that there are a lot of challenges to solve before we we reach a future where like I said earlier in this interview, both of these technologies are solving real human uh challenges. But what's the next big milestone you're looking forward to when it comes to the convergence of Blockchain and A I? Oh yes, that's a really good question. I think probably the next big milestone in terms of this convergence for uh Blockchain and A I, at least for myself is being able to see. I'm trying to figure out how to describe it. It's like um fully decentralized censorship resistant models. And I think we are starting to see this censorship resistance uh appear in the web two space um for better or for worse, right? There's definitely tradeoffs here. But the the issue here is that there is not that optionality, I think with web three, right, having that sort of permissionless censorship resistant model that people can access if given the option allows for a lot more flexibility, a lot for a lot more um entertainment of various use cases, right? Uh One very simple example was I remember for a brief period of time um cha G BT I think was um uh was, was a little reluctant to write code for people, right? And it's not because CG BT couldn't write code. It's just because, you know, what does this imply from like a uh a societal perspective as well as from a legal rights perspective, as well as from like what happens if it breaks and there's slas broken and all that stuff, right? So I understand the complexities there, but as a result, right, like this, this privilege was essentially taken away from the entire population of users, some of which are just trying to, you know, maybe learn how to code through chat GP T using it as tutorials, right? And so I think looking forward to these sort of censorship resistant models that allow for people to have open access globally. Um I'm sure that there is going to be political issues as well with A I models in the future. Um And so, you know, being able to kind of circumvent and continue to have this sort of global access is something that I think is going to be very powerful. You know, it's not easy for developers when web three became popular, it was difficult for web two developers to bridge over to web three. And now we're introducing A I to the mix. Uh What advice do you have for developers who are developing in web three, who want to start experimenting with A I implementation. I think it's a little bit tricky because right now the the trend, the the buzzword right is A I. And so I think that there's going to be a wave of attempts to kind of fit a square into a circle peg or a circle into a square peg. Um not exactly sure what the saying is. But um in, in that attempt to bring A I to use cases that may not necessarily be um you know, required, right? Super superfluous essentially. And so, um from the developer's perspective, though, in terms of integrating these models, it's actually not too complicated because, you know, if you even look in the web two space, a lot of these models right now are just basically chat GP T bots that have been optimized for specific use cases through API S and you know, subsequent training. Um But, and in the web three space, I think we're going to see a lot of the same things. I think that's the power of being able to access web two inference models in web three. And that's exactly, you know what we're bringing to the ecosystem through this partnership with Celestial or in Hyperbolic in order to make this all happen. Um And so I'm not too concerned about developer adoption of A I from the complexity perspective, but I am a little bit concerned about the practicality of all of these, you know, A I apps that will be floating around. Um, but I am sure that we will find, uh you know, specific use cases that would be uh hopefully valuable in the space. Well, that's exactly where my mind went, you know, when web three, well, web three still is very popular. But in the heyday during the bull markets, there are all these Web three applications that pop up that maybe don't survive that long. Um They raise money off of the excitement, they build off of the excitement, but that, that problem that they're solving is not there and they're not able to talk to the ultimate end user. And I think that we're probably gonna see this again with A I. So it sounds like your advice is to developers or people who are building an A I product is to really think about if we really need A I to solve this problem. Yeah, I mean, it's easier said than done, right? Like it's uh but, but I think, you know, the, the problem that you raise is it's not specific to A I. Right. I think it's more of this prevalent issue in web three in general. Right? And, and what is this issue? I think that the issue is not necessarily that we haven't built the killer app or these use cases that would drive that mass adoption. But we're a step ahead of ourselves in trying to find that use case because what we haven't really recognized as an industry is that we are using the wrong incentives. I think that, you know, crypto and Blockchain was born out of this need for peer to peer finance. And it has never really escaped that identity, even with the advent of, you know, ethereum and decentralized applications and all these other use cases. But at the end of the day, it still kind of flows back to this sort of financial incentive model. And that is great when you're using financial instruments, it's great when you're using defi products, but it's not necessarily great when you're using Social Fi or Gamey or all these other apps, right? Like, I mean, you know, I don't think anyone uses Twitter to, I, I wouldn't say anyone that's a little bit extreme, but I don't think most people use Twitter because they can make money, right? I, most of my friends, they don't even use the monetization route because their incentive of using Twitter has nothing to do with the money. Whereas in social fire, we see a lot of these sort of incentives kind of flow back towards, you know, the money and all this other stuff that are financial reasons, which I, I don't, I think are necessarily the right incentives to drive this adoption. And so, you know, thinking about just general web three adoption, I think is a bigger question. Um that isn't unique to you know A I decentralized applications, but just decentralized applications and the underlying incentives in general, Kenny. It's been such a pleasure chatting with you. Thanks so much for joining the show. Thank you very much for your time, Jen.

Learn more about Consensus 2024, CoinDesk’s longest-running and most influential event that brings together all sides of crypto, blockchain and Web3. Head to coindesk.consensus.com to register and buy your pass now.