Can Crypto Make AI More Trustworthy? Sreeram Kannan on Verifiable AI and EigenLayer

Transcript

AI is intelligence accelerator. Crypto is coordination accelerator. The Microsoft CEO Satya is asked in a podcast, is AI kind of like its own species? And then he laughs and says, no because at the end an AI cannot own property, cannot take liabilities, cannot enter into contracts, cannot start companies. But that's exactly what a blockchain is intended for. You can have fully self-sovereign agents. This is, you know, a very sci-fi forward view, but I think it's not that futuristic. I'm talking about less than a 3-year time horizon. Sriram, [music] you made a transition from academia to becoming a startup founder.

What was that transition like for you? I would say it's a continuous transition over many years from curiosity-driven research, which is what I used to do in academia, to impact-driven research, where once I realized I could be curious about any number of things, why not find the thing that has the most impact and then work on that. And so that over time naturally led me into a startup-oriented journey. Specifically, I got into blockchains and and crypto in 2018 and then I decided that I have to actually go and build this in the real world and make an impact.

>> For people who are not academic, how would you describe your curiosity-driven research? What were you investigating? Yeah, I was um I used to have this long-term research program. It's basically how is information stored in living systems. This was my research area. I was working on computational genomics. So you have a DNA and you have bits of information in it. And this somehow translates into this complex moving three-dimensional human being or other living beings. What is the steps involved in mapping this sequence of bits to this complex creature? If you can crack this mapping, you can make all kinds of new synthetic

organisms, you can cure diseases, you can understand costly what's going on in biology. So, this was my curiosity-driven >> [sighs] >> research program, which I actually ended up having quite a bit of interesting collaborators. I I wrote a paper with David Baker who won the Nobel Prize in Chemistry last year. So, this this was an interesting academic journey that I was on over many years before I got into blockchains and doing a startup. That's amazing. Tell us what you're building at EigenLabs. EigenLabs is building a suite of products called EigenCloud.

And when you think of a web application, you don't want, you know, developers don't want to provision the infrastructure themselves. They use this cloud in which they can just send their software, it executes correctly, and then brings them the results. So, that's what happens in the cloud. What we found is in crypto, you need something similar. When you want to develop a decentralized application, today you have to go and find nodes all around the world and then create a decent distributed community and then make sure they're all like running your node software and then coming to consensus. There's a whole bunch of steps that go underneath starting a new decentralized protocol.

What if you could build a cloud for decentralized applications? Of course, different from the regular cloud. The cloud itself has to be decentralized, right? So, you can't run on a centralized data center. So, the base layer of EigenCloud is EigenLayer. What happens in EigenLayer is anybody can come and put up some money in terms of ETH or whatever crypto tokens and then promise that they are going to run a certain software correctly. And we have a massive amount of stake, nearly $20 billion is staked on our protocol. And a global decentralized operator network where lots of nodes around the world participate. This goes all the way from home stakers operators to professional operators like Google Cloud

or Coinbase. So, you have the whole spectrum. And that becomes the kind of base base layer of Eigen Cloud. And so, anybody who wants to develop a decentralized verifiable application can write their application and then deploy it into Eigen Cloud, where there is a series of abstracted services, you know, for data, for compute, for AI, for inference, all these things. You mix and match these uh services to build very powerful crypto applications that were simply impossible before. So, how do you describe that to somebody who maybe doesn't get crypto, doesn't really follow this industry, or why that might be important?

>> When you think about uh what is crypto, there are different answers, but the answer that resonates deeply with me is it's a technology for trust. How do you trust something? And if a program is guaranteed to execute cryptographically correctly, that's the highest grade of how you know that something actually does what it says it does. And crypto is a technology for trust, and if you think about, you know, societally, one of the other biggest things happening is AI today, right? And the question I think in a lot of our minds is how do you make sure you can trust AI? And that when you're delegating more and more agency to these agents, how do you ensure they have the right set of constraints and guardrails and they're enforced

in a way that protects the behavior exactly like you want. And that's where I think the set of things that we are building can make a big impact because you can make AI verifiable, that when you give your agent a certain task, or when you give a trading agent an a certain action, you are guaranteed that this piece of code interacted with this piece of authenticated data to actually make this set of actions, as opposed to you know, uh somebody running one of these models felt differently and then changed some parameters, and then suddenly you're now left holding a bunch of Dogecoin that you don't want.

So, a a lot of people are excited about this convergence of AI and crypto, but how much of that is real versus projections of what may be to come? In any industry which is not yet fully formed, a lot of people trying to find different aspects of it. But, I think if you abstract it one step up, you can see what is AI is just programmable intelligence. Right? So, it's it's intelligence that is programmable digital, right? And if you have programmable intelligence, what you need is you can't work with soft squishy legal slow regulated institutions. You have to work with digital programmable automatically

enforced, you know, instantaneous institutions. And that's what crypto lets you build. So, we believe that this intersection that programmable intelligence needs programmable institutions, smart contracts, and the ability to like engage in contracts, own property, all of these things are really powerful. In terms of what specific ideas and projects are there, I think people are trying all kinds of different things. Some people are saying, "Oh, we need to get GPUs all around the world to participate." So, that's one angle. Some other people are saying, "Oh, I'm going to incentivize data collection because, you know, you have tokens. That's an incentive mechanism." Which is an interesting angle. What we are working on is making everything verifiable. So,

verifiable trustworthy. And so, I think some of these pieces will come together, but I think the meta thesis that programmable intelligence needs programmable institutions. And crypto money as the first example of an institution that crypto built, I think most people anchor on that as like, "Okay, that's the scope of crypto. It's money technology, money Legos." But, actually what we believe is much bigger because, you know, cryptographic proofs of correctness is a technology for trust and verifiability. So, yeah, a lot of attempts right now. I think the area is still very very early.

The scope of what this can do, I think maybe projecting to the most extreme version. >> Please do, yeah. What's What does the future hold? >> The future is fully sovereign intelligence. So, when you think about AGI, right? And like artificial intelligence, maybe a half a trillion or a trillion dollars spent this year like to actually like grow this industry. And you ask what is the extremal version of like an artificial intelligence? It's an intelligence that's not, you know, your agent or my agent. It's its own agent. You know, the Microsoft CEO Satya was in a podcast and about his AI kind of like its own species.

And then he laughs and says, "No, because at the end an AI cannot own property, cannot take liabilities, cannot enter into contracts, cannot start companies." But that's exactly what a blockchain is intended for. A program can own property and take limited liabilities. It can start companies DAOs. So, it it can actually like you can have a fully self-sovereign intelligence, an intelligence that comes packaged with a unit of money. It goes and does work and earns money whether as an influencer, as a scientist, as a you know, as a programmer, as a teacher. It just has all these jobs and then it does those jobs and then earns money and then it it it uses that money to fund its, you know, future training and improvement.

And you can have fully self-sovereign agents. They can also raise money for themselves by issuing, you know, income shares from their future. Everything programmatically and on chain. This is, you know, a very sci-fi forward view, but I think it's not that that futuristic. I'm talking about less than a 3-year time horizon to to this happening and you can already build elements of this on Eigen Eigen AI today. But when you think about the the other socio-economic aspect of AI, one of the things people are missing is the ability to actually invest in some of the biggest successes that are happening in AI, whether it's an open AI or a cloud.

They're privately held companies. 800 million users, but none of them can invest in, you know, what they know and like love. I think this is a very different world from 30 years back when Microsoft IPO'd at $100 million and rode the public markets up to $3 trillion. That as these kind of like fully on chain programmable intelligences can get, you know, users invest in them at a very early stage because they're actually like fully autonomous sovereign. So, I think there's a lot of convergence of the technology and the economics that makes it really powerful and interesting. If we're on this trajectory where we're building these fully autonomous, fully sovereign agents,

new species maybe, what ultimately ends up being the purpose of you and me, of humans? What do what do we do in this world? >> What do we do in this world? I think the first thing to realize is that when we have abundance, right, in terms of the technology creating outcomes and outputs, it lifts up everybody's like standard of living. So, that's a base base level. Now, as a human, what do we do? There are so many dimensions of being a human that I think what has happened in modern society is we're very over indexed on our mental prowess. If you went back 300 years, right, you know, before the Industrial Revolution, if you're physically strong, you have a huge edge.

And like with machines, being physically strong doesn't have that much of an edge. And with AI, you know, basically another way of thinking about it is democratizing intelligence and expertise. So, have being like mentally super agile is not specifically an edge relative to like an AI, but I think when you think about what it makes for us to be uniquely human, it is the human interactions, it is actually working with each other on positive sum outcomes. I think these are uniquely human attributes. Leading humans, working with humans. I think this is there is going to be a uniquely human set of traits. And the way I think about it is just because humans exist, it doesn't mean all the other species have

vanished. Right. So, we have you know, we have >> Although some people are worried about the humans making the other species vanish. >> Making some of them vanish, for sure. The large scale of evolution, humans one species along with all these other, you know, biological species and I think digital species which will exist and coexist with us. I think the question is how to make it synergistic, symbiotic with humans existence as opposed to counter-aligned. Because the same power of smart contracts, trust and verifiability can also create binding conditions that these AI agents have to obey or they'll lose their position, lose their money, whatever the thing is. So, you can actually build super additive society where where like, you know, these niches of humans and

agents can exist. Very cool. Sriram, you have time for a lightning round of questions. What's the worst advice you've ever received as a founder? Worst advice I've ever received as a founder is listen to investors in deciding how to build a product. [laughter] I feel I feel seen. What book should people read? I love Sapiens. Mhm. Yuval Noah Harari. I think the first time I got this thing that humans superpower is not intelligence, it is coordination. I think he makes this click case very clearly there. Mhm. You can even think of like AI is intelligence accelerated.

Crypto is coordination accelerated. So, that's one of the reasons I switched all the way from computational genomics to come work on crypto. Mhm. Super interesting. That is a great book. And Sapiens meaning like what, the ape who knows or something? Isn't that like the translation? Yeah. Um Homo sapiens. Yeah, the wise ones. The wise ones. Okay, there you go. What's your biggest productivity hack? A clear mind. When you have big problems, the instinct is to like rev yourself up. I think when you clear your mind, you actually, you know, give time for the subconscious to process out all the complexities that might that you might not see clearly. Do you have a ritual, a practice, a regimen? Different things I try, but meditation generally.

I think that is the That's the one that I would index on. Okay, and what is the smallest hill that you will die on? Coffee. Coffee? What about it? It is a mug which says like a mathematician is a machine that turns coffee into theorems. >> [laughter] >> Well, some people say it's good for you or not good for you. I'm like, I love coffee man. I can definitely get on board with that. Thank you so much for joining us, Shriram. >> Thank you so much, Robert. Really, my pleasure.