GRTiQ Podcast: 43 Sam Green

The following podcast is for informational purposes only. The contents of this podcast do not constitute tax, legal, or investment advice. Take responsibility for your own decisions, consult with the proper professionals, and do your own research.

Sam Green (00:00:24):

I see so much competence and passion in the people and teams involved in The Graph. I think that we’re going to see this protocol be everywhere in web3.

Nick (00:01:07):

Welcome to the GRTiQ Podcast. Today I’m speaking with Sam Green, co-founder and CTO at Semiotic. On December 9th, The Graph Foundation announced that Semiotic would be the fourth core dev team added to the protocol, joining Edge & Node, StreamingFast, and Figment. Shortly after the Semiotic announcement, The Graph Foundation announced it would be adding The Guild as the fifth core dev team working on The Graph.

(00:01:34):

During my conversation with Sam, he shares incredible insights about Semiotic’s world-class expertise in artificial intelligence and cryptography. With incredible clarity, Sam explains important concepts during the interview, including cryptography, reinforcement learning, and zk-rollups. The topics and insights Sam shares here are illuminating, not only for the future of The Graph in web3, but to the frontiers of technological innovation The Graph Protocol is exploring. As always, I started the discussion with Sam by asking about his background.

Sam Green (00:02:09):

Well, let’s see, so I’ll go back to my undergraduate days. I started out working on a CS degree, computer science degree, and I wasn’t going to classes. And my dad was a professor at the university and he said, “Hey, are you going to classes?” I said, “No.” He said, “Hey, you should withdraw so your GPA doesn’t get wrecked.” So I withdrew, and then I spent a few years, waiting tables, construction. And one day, I was working at a golf course, and I was digging holes, doing maintenance, stuff like that, and I said, “This is pretty hard work and I’d rather work my brain hard than my body, and so I’m going to learn web development and I’m going to get a job doing web development.”

(00:02:53):

So I basically quit that, the maintenance job, and I teach myself Adobe Flash. And then a few months later, I get my first web development job. And then I’m doing web development, and then I start learning HTML, JavaScript, PHP, Perl, mySQL. And during that time, I realize, actually I’m a dummy and I don’t know any basic knowledge about computer science. I don’t understand how these basic algorithms work, database design, all this basic stuff. And so I realize, okay, maybe there is something to be gained from formal education. So then I go back and I start taking classes on the side. So I’m working, taking classes.

(00:03:38):

And near the end of my bachelor’s degree, I have this amazing professor, he is a math professor, and he really got me interested in math, especially applied math. And this guy, you should do a Google search for “grandpa mathematician builds castle Reddit”, and you’ll see this guy come up, he’ll be the first result. While he was a professor, he was building these amazing bridges. He was analyzing these bridges, real bridges that he was using. Then after he retired, he kept honing his skills.

(00:04:16):

So after I finished my bachelor’s degree, I go work for a company called Acxiom, and they buy your credit card transactions and they use those transactions to help people market toward you. So it’s kind of the traditional evil corp sort of company. I do that for seven months, and then I just can’t take it anymore. And I decided to go back and work on a master’s degree in applied math, which was, that’s where my passion was at the time. But kind of in between, when I got the job and started the master’s degree, I’d also gained a new mentor. He was a physicist who had turned into an embedded developer. So an embedded developer is basically someone who uses small computers called microcontrollers to make gadgets do intelligent things. So this was like a precursor to the internet of things sort of guide.

(00:05:07):

He taught me how to design electronics, how to design print circuit boards, how to program these little devices. So I spent a lot of time with him and learned these skills. So when I went back and worked on my master’s degree, my advisor, he was a fluid dynamics guy, and so I basically used my background in computer science, my interest in embedded devices, and I specialized in using field-programmable gate arrays, FPGAs, to accelerate fluid dynamics computations. Now it turns out that that’s actually a horrible application for FPGAs, but I did learn an interesting skillset of designing these accelerators. And it turns out, FPGAs, by the way, were the first devices used to accelerate bitcoin mining.

(00:05:58):

So after I had finished my master’s degree, I had decided at that point that I would never do another boring job. I would rather, well, just go back to construction or whatever, as opposed to doing a boring or soulless job. So I moved to San Diego where my brother was. He’s like, “Hey, come move out here. There are a lot of good jobs out here.” I was previously in Arkansas. All that backstory was in Arkansas. So I moved out to San Diego, and through my mentor, my embedded electronics mentor, he got me a job out there. And that job, I worked for what’s called a micro controller design house, specifically Pic Microcontroller, it’s a brand. And small companies who were building gadgets and having problems getting their gadgets running, they would come to us and they’d be like, “Can’t get this thing running. Can you help us fix it?” So it was fun. I would help debug people’s products, get their products working, and hopped from product to product.

(00:06:52):

So I did that for about a year. And then my brother-in-law got a job in New Mexico at Sandia National Laboratories. They’re one of the Department of Energy’s nuclear engineering laboratories. And my brother-in-law started working there. And of course, he knew me well, and he’s like, “We are doing some awesome things, and we’re hiring, and you should apply.” And I already had high regards for Sandia Labs because I used their research during my master’s research, so I knew who they were, I knew what they were up to, and I would really wanted to work there. So I applied and I got the job.

(00:07:27):

And when I got there, they basically said, “Hey, you have this combination of computer science, applied math, and this embedded development knowledge. We have a project that focuses on cryptographic hardware assessment. So basically, what you would be doing is you would analyze cryptographic hardware for weaknesses, and you would help us build better cryptographic hardware through your efforts.” That really aligned with all of my interests at that point, and I couldn’t think of anything more exciting to do at that point. And so I said, “Yes, let’s do it.” And so I spent five years, really working with this amazing team of electrical engineers, mathematicians, high performance computing engineers, and we did some really nice work on these cryptographic hardware sort of attacks.

(00:08:19):

During that time, we would go to conferences to stay up to date with cutting edge research. And I had read a book called Cryptographic Engineering by Çetin Koç, K-O-C, and he had co-founded the largest cryptographic hardware conference. And we went to this conference where I was at one of these conferences and I met him, and I just wanted to say, “Hey, I really liked your book and I liked your past research.” And that was really the intention, just to say hi and meet someone who I respected. He said, “I’m going back to UC Santa Barbara, the University of California, Santa Barbara, and I’m looking for PhD students. You should apply. Come work with me.” So I said, “Yeah, Santa Barbara is a great place. It’s right by the ocean on the coast of Southern California. And I love your work, and I would love to do that.” It took a few years, but then in a few years, I was there in Santa Barbara, working on this PhD with him.

(00:09:13):

So my PhD was funded by industrial money to look at electronic and internet voting systems. So that was really interesting to me, especially coming from the military side of things at Sandia Labs. And my academic interest was in reinforcement learning. During my PhD, what I focused on was efficient or low power reinforcement learning. So I did that for several years, and when I was getting ready to graduate, about two years ago, I got a phone call with Ahmet Ozcan, he’s the CEO of Semiotic, and he said, “We’ve been trying to work together for several years. We’ve been collaborating for several years. What do you think about starting a company together?” And at that time, I still, at that time, really just wanted to do something interesting and meaningful. And even though I loved doing R&D at Sandia Labs, which is where I was actually working, as I finished up the PhD, they were supporting me generously to wrap up the PhD. I could work and publish. I said, “Hey, I want to do something bigger and more ambitious, and starting a company really fit the bill.”

Nick (00:10:28):

Sam, thank you so much for that background and overview. What an interesting career and the threads that kind of come together to where you are today. For listeners that aren’t a hundred percent clear on what is meant by applied math, how would you describe what applied math is?

Sam Green (00:10:42):

Well, there are two types of math. There is pure math and applied math. Pure math is focused on proving some sort of statement or some sort of theorem. And applied math is about taking the tools of math to do something with those tools. So for example, if you’re trying to train a neural network, that would be applied math. If you’re trying to prove, let’s say, the space between prime numbers, that would be an example of pure math.

Nick (00:11:13):

Makes sense, and I appreciate that overview of that. I want to also go back to something you said about your time at Sandia National Laboratories, because I think this’ll play a lot more into what Semiotic and what you are doing as core devs. You are cryptographic engineer, and obviously, the crypto space seems to have some name origins with cryptographic. What is meant by cryptographic engineer? What’s the type of work that you are doing?

Sam Green (00:11:38):

Well, at the time, cryptographic engineering would apply to the art of building cryptographic systems that would implement some sort of cipher. Let’s say, for example, your cell phone. When you connect to a cell tower to make a call on your cell phone, you actually establish some sort of cryptographic handshake between your phone and the cell phone tower. And that handshake is actually pretty heavy duty mathematically. It takes a lot of computations. So you have a chip in your phone that will accelerate those computations that are needed to establish that handshake. And so that chip would be designed by a cryptographic engineer and the whole software stack to make that work. That would also be designed by a cryptographic engineer. One would be more hardware oriented. One would be more software and protocol oriented.

Nick (00:12:40):

So then, Sam, how would all of that relate to the crypto industry and the crypto space?

Sam Green (00:12:45):

Well, it’s actually very related, and it’s an extremely exciting time, for people who are interested or have backgrounds in cryptography, especially cryptographic engineering.

(00:12:57):

When I was at Sandia, the way I perceived cryptographic engineering was as a fairly kind of fixed field, and most of the problems had been worked out, and we were just trying to fine tune problems and just make sure that implementations of well understood problems were done correctly and securely. But there weren’t really a bunch of open questions. What’s happening with crypto and blockchain is that, we have all of these brand new problems. And these problems are going to be solved in the long run by cryptographers, when I say cryptographer, that’s more of a mathematician, someone that’s more of the pure math side, and cryptographic engineers, people who work with cryptographers to figure out how to apply the output of cryptographers or the math, the ideas that cryptographers come up with, to solve real world problems. And so this space, the space of crypto turned cryptography and cryptographic engineering, from something that only really mattered to, let’s say, the government, or if you needed to fine tune something for some well understood application, into this really blossoming new field of problems that opens the door for many people to have careers in this field.

Nick (00:14:14):

Well, I love that explanation, and I’d be curious to know then, as you think about the problems that cryptography solves in the world, what a short list of those problems might be.

Sam Green (00:14:24):

Sure. At a high level, cryptography solves the CIA problem. So that is confidentiality, integrity, and authenticity. So an example of confidentiality would be you and I wanting to communicate, without a third party being able to know what you and I are saying to each other. That’s confidentiality. Integrity is, you and I are communicating, and then a third party or some sort of external random influence can’t corrupt the messages between the two of us. That’s integrity. Authenticity is you and I wanting to communicate, and let’s assume we’re not face to face, and our messages are being sent to each other over a long distance, you want to know that the message that you received that was purportedly from me is actually from me, and vice versa from my perspective. So that’s the CIA problem, and that’s what cryptography solves. And we can unpack this, but this is where we’re using these principles. These confidentiality, integrity, and authenticity principles are also being applied in the crypto blockchain space.

Nick (00:17:02):

Well, I love that CIA acronym. It lays out some of the problems that cryptography seeks to solve. But I guess a follow-up question I would have for you, Sam, then is, I always understood that blockchains were inherently or embedded with cryptography. So I guess I’d be curious, what do cryptographic engineers or people with your type of background have to do in the blockchain space?

Sam Green (00:17:25):

That’s a great question. So at a high level, what I would like to say is that, you can actually get a degree in cryptography, which means that, really, cryptography can be used to solve many different types of problems. If we go back to the OG blockchain, Bitcoin, there were going to be a couple of types of cryptographic techniques used. There was hashing used. And the hashing provided integrity, going back to the CIA acronym. And people had, of course, their keys that they were signing transactions with, and that gives authenticity. So crypto or cryptography has been used in the past and is a key component for crypto and for blockchain.

(00:18:10):

But what we’re looking at, especially moving forward with techniques like zero-knowledge proofs, zk-SNARKs, is that we’re going to be introducing and developing brand new techniques that weren’t previously used in other aspects of crypto.

Nick (00:18:29):

Am I right in assuming, Sam, that a lot of the fundamental nature of cryptography is in math, that it’s not a piece of software or it’s not a piece of hardware, but cryptography is really applied math, to borrow a term you used earlier?

Sam Green (00:18:45):

Cryptography is this special field. It sits in between the realms of theory and application. If we go back in history, the two mathematical pure math fields that really influence cryptography are number theory and abstract algebra. And really, what we’re talking about, where cryptography comes from, is that, for a long time, people have considered numbers in just this pure sense of the word and the relationships between numbers. For example, a common one with cryptography is divisibility. So if you take one number, you divide it by another number, will you have a remainder or not? That’s divisibility.

(00:19:29):

And all of these special properties, all of these studies of the relationships between numbers actually resulted in the discovery of these hard problems, which cryptography is based on. So cryptography is then people realizing, oh, I can actually take this relationship, this hard mathematical relationship that previously had no application to anything, and I can now protect information with this hard mathematical relationship. And so that realization right now is what transitions us from the pure math, the theory, into application, into the applied math realm.

Nick (00:20:13):

Sam, I can’t help but think, as you’ve outlined your history, coming up and not necessarily enjoying school, so doing some manual labor for a time, figuring out that that’s a lot more difficult so you decided to go back into academia. You pursued math. Then you got employed by a company that you didn’t necessarily agree with the nature of its work, found yourself to the West Coast, found yourself into cryptography. I mean, all of this seems to be a vortex, a path that led you to where you are today. Is that how you see it as well? Everything’s led you to where you are today?

Sam Green (00:20:45):

Yeah, absolutely. I’m very happy with where I am in my career right now, and I couldn’t have predicted back then, as I was taking this kind of following my passion sort of path, that I would collect the skills that I needed, that would be, I believe, were very applicable and useful in the blockchain space. So I think of crypto as being the combination of economics, computer science, and mathematics. And I have a deep background in two out of three of those fields. And of course, as I spend more time in crypto, it’s inherently an economic system. So I think all of us learn more of the economic aspect of it, become more interested in economics, because of our involvement.

Nick (00:21:33):

So then, Sam, how did you find yourself in the blockchain space?

Sam Green (00:21:37):

Well, it’s a bit of a story. I was wrapping up my master’s degree in 2009, and at that time, the best source of technical news was Slashdot. And I was reading Slashdot and I saw Satoshi’s paper posted. And I’m like, okay, so this digital currency based on cryptography. I download the paper. I read it. And I’m like, huh, I understand the math that he’s talking about. I also, because I’m doing this FPGA, field-programmable gate array, accelerations, I understand that I could accelerate the mining of Bitcoin. My skills are kind of at that peak skills right now because that’s all I’ve been doing is accelerating math. And I think about it and I’m like, but I don’t know really what this is. I don’t really understand the implications of this technology, so I have to pass. I’m going to go back to tinkering with electronics, instead of building an accelerator and plugging it into a server to mine crypto.

(00:22:39):

Then I get this job at Sandia Labs and teaching cryptography courses. People talk about, Bitcoin, at that time, back in 2010, it was around $300. So it’s getting some attention and I’m like, okay, we’re studying it. My group, my cryptography group, we’re like, we understand how this works. We understand why it’s secure. We understand why Bitcoin is a scarce resource, but the only application we see is Silk Road, these black markets. And I’m working for the government and I’m like, this is a super interesting thing, but I don’t really want to be involved in this thing. So yeah, that’s about 2010 to 2015.

(00:23:19):

So 2015, I go back to grad school, UC Santa Barbara, and I get there. And a few years after I get there, Ethereum starts taking off. And of course, these students, they’re more exploratory. And I’m in a cryptography lab, so there are students in the lab talking about Ethereum, and that’s the first time I come across smart contracts. And this was the first time that I’m like, okay, so this is interesting, but I see there’s just this ICO boom, and I don’t really know what these ICOs are about. I don’t really see their applications. I see there’s something happening, but I still can’t pull the trigger yet and get into it.

(00:24:05):

Then let’s fast forward until 2020. So in 2020, I’m starting Semiotic. And at that time, I moved to the Bay Area, Palo Alto, and I’m friends with the lead cryptographer at Edge & Node. And I’m starting my own company, but it just so happened that the mathematical techniques that I was doing, which was not crypto related, it was more cryptography, old school cryptography related, the mathematical techniques overlapped with Jackson’s, the techniques Jackson was using, for his work for The Graph. And so we would talk every few weeks and I would be thinking, I was thinking, wow, the stuff that I’m doing for Semiotic, this is pretty cutting edge stuff that we’re doing. But when I’m talking with Jackson, I’m thinking, wow, this actually sounds like the future. I mean, I thought I was doing some cutting edge stuff, but Jackson’s actually working on the future.

(00:25:02):

And so come, let’s say, the end of 2019, that is when the GRT becomes public. I’m thinking, I really like Jackson. We’ve spent a lot of time talking about The Graph. I understand The Graph actually has a utility. So if we come to the end of 2020, I think to myself, okay, so now actually I know quite a bit what’s going on, at least with one protocol. I understand there is this utility that Jackson has been working on and it sounds awesome. And so I’m going to get involved with this utility, The Graph, and in addition to that, I’m going to learn everything I can about crypto during 2021.

Nick (00:25:43):

So, Sam, for listeners that are interested in learning more about cryptography and some of the things we’ve talked about so far in the interview, could you point them in the direction of any books or resources that they might be able to get a nice introduction into the field?

Sam Green (00:25:56):

Yeah, I’d love to. There are two resources that I’ll recommend, and neither of them actually require a high level of mathematics. You don’t have to be an engineering or a math major to get this stuff. You really just have to have basic knowledge from high school, and time. I’m not saying that the techniques are trivial, it will take time, but they’re definitely accessible mathematically.

(00:26:19):

So one book is called Understanding Cryptography by Christof Paar. This book goes through more of the fundamental cryptography algorithms that are very ubiquitous. These algorithms are going to be the basis for how Bitcoin worked, for example. And then if you want to look at something that I would say is more cutting edge, but everybody, I believe, can still tackle this with time and energy is, if you want to jump straight to zero-knowledge proofs and zk-SNARKs. The reason that I even bring this up is because I think this is where the unexplored boundary is. This is the frontier. And there’s an amazing book by Justin Thaler, T-H-A-L-E-R. He has a book on zero-knowledge proofs, and it’s freely available on his website. And it takes you from the very beginning, he assumes no prior knowledge, and he’ll get you up to speed with everything you need to know to understand the cutting edge, what’s happening in this field.

Stefan, The Graph Academy (00:27:23):

Hey, this is Stefan, from The Graph Academy, announcing The Graph Academy Grants Program, a program that is intended to foster the growth of our community-driven platform, The Graph Academy. We are offering grants to contributors to incentivize the creation of high quality educational material. If you want to apply, simply visit thegraph.academy/grants. That is thegraph.academy/grants. You can also find more information in the show notes. Thank you very much for listening.

Nick (00:28:12):

Sam, I have a lot more questions for you about how you became acquainted with The Graph and some of that utility that you recognized early on. But I want to come back to Semiotic. As we track your story, from working at Sandia National Laboratories, you connect with an old friend and you launch Semiotic. So I’d be curious to know, when you started Semiotic, what were some of the problems, some of the things you wanted to initially address in the world?

Sam Green (00:28:35):

When we started Semiotic, we were focused on encrypted computing, specifically a type of cryptography called fully homomorphic encryption. And let me give an example use case to give some intuition of what we were going after. Imagine you’re wearing your Apple Watch, and your Apple Watch actually has the ability, as you probably know, to collect your heartbeat data. So with homomorphic encryption, you can collect your heartbeat data on your watch. You can then encrypt it with your secret key. You can send your ciphertext, also called your encrypted data, to the cloud. And then on the cloud, if you’re using fully homomorphic encryption, which is what we specialized in, we could then use a neural network and input your heartbeat data into the neural network and do something like detect whether or not you have COVID. We then take that prediction, that detection, send it back to you, and then you would decrypt the result.

(00:29:40):

Now the magic is that, on the server, even if the server had been hacked and was compromised, or if my company wanted to know the answer, we couldn’t. Your data is never decrypted. That’s what fully homomorphic encryption gives us. And maybe more as a mathematical concrete example, imagine you had two numbers that you encrypted separately. You encrypt A where A is a number, and you encrypt B where B is a number, and you have these two now encrypted numbers, these two ciphertexts. With homomorphic encryption, you can add those ciphertexts, you can give them to me and I can’t see what those ciphertexts are, but I can add them. And I can also multiply them. And then only you, once I send the result back, can decrypt it and see what the answer is.

Nick (00:30:31):

So you started in web2 with this really great technology, and then I got to assume, you made a pivot here at some point. You talked a little bit about this in your relationship with Jackson at The Graph, but what was the pivot moment for Semiotic then moving away from this original idea to what you’re doing presently?

Sam Green (00:30:47):

So we made a lot of progress on the fully homomorphic encryption technology stack. And we had some interests, specifically, we got interest and funding from the National Science Foundation, the NSF, and from DARPA. The DARPA funding actually came through this great program that’s run out of Berkeley National Laboratory called Activate. And what Activate does is it supports hard tech startups. And so we got this early support from the government and it was very critical to us, to our existence.

(00:31:24):

But we did not get much traction in the business world and in industry. And it turned out it was because, the decision making of the industry, and specifically, I should say, we were targeting healthcare and FinTech, decision making in healthcare and FinTech is driven primarily by regulations. And because homomorphic encryption is such a new field, there are actually no regulations that cover it or compel companies to use it. In fact, the way most of these industries work is they just simply trust the cloud. They trust Google that Google won’t let anybody hack their servers, or Amazon won’t let anybody hack their servers. So we were having some hard time getting traction there. So how did we pivot?

(00:32:15):

So I was talking with Jackson Blazensky, he’s the senior cryptographer at The Graph, and he came to us and he was like, “My boss, Brandon Ramirez, is interested in, not homomorphic encryption, but reinforcement learning.” And reinforcement learning was my PhD topic. And Jackson said, “Would you be interested in having a chat, just a high level chat about how reinforcement learning could be applied in The Graph protocol?” And I said, “Sure, yeah, that’d be great. I would love to just have a chat.” Reinforcement learning is a topic that I’m interested in, just generally, and so I was happy just to chat about that.

(00:32:57):

And so the context of that chat wasn’t about us doing work for The Graph. It was just to bounce ideas back and forth with Brandon Ramirez, about what is The Graph trying to do with reinforcement learning, and where could it be fit in, and what sort of skill set do you need to apply reinforcement learning, I’ll just start calling it RL, to The Graph? And so that was a great conversation. And then a few days later, Jackson said, “Hey, there is this grant program from The Graph Foundation. You should consider applying for a grant and see if you can do anything with RL on The Graph.” And so, we talked among ourselves and we’re like, okay, so do we want to do this? Do we want to pivot basically, or at least spend a significant amount of time delivering on this task if we were to get it? And we chat and we’re like, yeah, let’s give it a shot.

(00:33:55):

And so I have a few more conversations with Jackson and Brandon, trying to understand what is needed. And we spend a lot of time, trying to write a detailed proposal for a grant. And then we’re interviewed by The Graph Foundation during the grant process. Then eventually, we get it, and that would be our entrance into the crypto space, into web3.

Nick (00:34:21):

As one could expect in talking with someone like you, with such a diverse background in some really kind of technical and specialized fields, you bring up some new terminology. I want to spend a minute here, before we get back to the Semiotic story, and that’s reinforcement learning. What is reinforcement learning?

Sam Green (00:34:37):

Yeah. So we’ve been talking about cryptography, and now we’re going to switch to AI. Reinforcement learning is a branch of artificial intelligence, and it is focused on learning through trial and error. Specifically, it’s focused on AI agents that learn by themselves, through trial and error. And what may be the most well known example is the AlphaGo agent from DeepMind, which learned how to play the game of Go by itself. And this was a game that computer scientists speculated would not be solved by AI agents for a long time, for decades, because of how many moves are possible in the game of Go. But reinforcement learning was used to then solve it and then beat the world’s best Go player.

Nick (00:35:29):

Well, I’m actually familiar with that game. I watched it online one day and I was blown away by it because you can watch this machine actually learn the game and pass the game. And even as somebody who knew that it would pass it, I still watched with suspicion. And I think the thing that really struck me was the benefits that machines have of going through cycle, cycle, cycle, and learning something new. What’s the background of this discipline? I mean, how did we ever come to the idea that machines could learn and do these types of things?

Sam Green (00:35:58):

So reinforcement learning is basically an algorithmic instantiation of some ideas that come out of behavioral psychology. And specifically, these are the ideas of learning through trial and error, and reward and punishment. A very intuitive example is, imagine you have a new puppy and you want to teach the puppy how to shake or roll over. And it takes some time, practice and treats, and then you can get the puppy to do that.

(00:36:30):

But you can actually extend this idea to some fairly extreme ends. And I think you really highlighted the fact that, with computers, especially if you can simulate the sort of task that you would like to solve, like you can with the game of Go, if you can perfectly simulate your task, then you can run billions of trials and errors, and you can find, or your agent can find, the thing that it needs to do or the things that it needs to do, to achieve very high performance in very complex tasks.

Nick (00:37:07):

I’m sure a lot of these themes, Sam, will reappear here when we talk more about The Graph and the work Semiotic is doing now as part of the core dev team. Before we get there, I’d like to learn more about the team at Semiotic. What can you tell us about the team at Semiotic?

Sam Green (00:37:20):

Semiotic has four co-founders. Ahmet Ozcan is our CEO. He is formally trained in physics. He has his PhD in that field. He previously worked in silicon device manufacturing, before transitioning into machine learning where he became the manager of the Machine Intelligence department at IBM Research. We also have Alexis Asseman. He is formally trained in nano engineering, and he worked for Ahmet at IBM Research, doing AI hardware acceleration. And then finally, we have Gokay Saldamli. Gokay is also a professor at San Jose State University where he specializes in cryptography. And prior to becoming a professor, Gokay worked at Samsung, where he designed Samsung’s hardware accelerators for their cryptographic chips that then went into billions of cell phones. So those are our four co-founders.

(00:38:22):

What’s unique about our founding team is the mix of deep AI experience that we have, and the deep cryptography experience that we have. And we’ve been blending those two fields since the beginning of the company, and we’ll continue to blend and apply those two fields, maybe sometimes separately for some tasks, and sometimes together for some tasks.

Nick (00:38:52):

We’ve already got Figment working on the core dev team. We’ve got Edge & Node. We’ve got StreamingFast. So how should listeners think about the way that Semiotic will fit in?

Sam Green (00:39:02):

Well, first off, it’s a big honor to be included in this company of the other core devs. These groups are highly skilled. As we all know, many of the team members have been interviewed on GRTiQ Podcast, so these teams are doing great work. So how does Semiotic fit in now?

(00:39:21):

Well, kind of at a high level, we’re the first research and development team that has been added to the core dev group. Our history of our company is, all of us were industrial R&D researchers, prior to starting Semiotic. And now what we’re doing is we want to improve. We want to improve the protocol through R&D. We’re going to be doing that on two fronts. One front is going to be related to AI and reinforcement learning, and the other front is going to be heavily focused on cryptography, and especially, zero-knowledge proofs and ZK-SNARKs.

Nick (00:40:03):

So let’s take both of those for a minute here, if you don’t mind, and just dive a little deeper. We’ve already covered some of the fundamentals and your background in both of these verticals, if you will, but how would you describe what you’re going to be doing in terms of AI and reinforcement learning for The Graph?

Sam Green (00:40:18):

We’re going to be applying AI and RL in two ways. Well, one way is going to be to test the current protocol for its strengths and weaknesses. And we’ll do that by basically training agents to interact with either the existing protocol, or propose changes to the protocol, to see how the mechanisms of the protocol react under, let’s say, the load of an agent who may be trying to exploit the protocol for its own ends. That’ll provide feedback in fine tuning the crypto economic mechanisms that really underlie the rules of the protocol. So we’ll apply AI and RL for that use, for testing the protocol and proposed changes to the protocol.

(00:41:07):

And we’ll also apply AI and RL for building tools to increase the efficiency of the human users of the protocol. So for example, one of our focus areas right now is building a new tool for Indexers to help them automatically price their incoming queries. Now, why do you need that? Well, it turns out that when you get a subgraph query, the query can be very lightweight in terms of how many resources it takes the Indexer to respond to it, specifically the Indexer server that is running The Graph node software. It could take the Indexer server a trivial amount of time to respond to the query. You can imagine, let’s say, a Uniswap query that says, give me the token prices for the two tokens. That’s a very lightweight subgraph query.

(00:42:01):

On the other hand, you could have a very heavy duty query that basically requests that the Indexer dump their entire database. And then really, you can have anywhere in between. You can get queries that are anywhere in between these two extremes. And the Indexer has to, of course, pay for their infrastructure costs. And so how do they pay for it? Well, they need to charge for each of the queries that they’re getting. And the tools that we’re working on will let them automatically learn their costs for queries, and then automatically charge the appropriate price for their queries. And that will basically free them up from having to manually set prices for all of the different query styles and complexities that they may receive. So automatically pricing queries is an example of a tool that we are currently working on to increase the efficiency and improve the ease of use for the human users in The Graph.

(00:42:58):

But in the long run, what our goal on this front for AI and RL is to make tools for wherever they would be useful, in all parts of the protocol. If there are tools we can build to help Delegators, Curators, or dapp developers, we’d also like to eventually build tools to help these people too.

Nick (00:43:17):

So, Sam, The Graph is already so cutting edge, and web3 is so exciting. I mean, I often feel like we’re right at the frontier of something really spectacular in terms of technology, for humanity. And then to learn what you are adding through Semiotic with AI, I mean, this is really incredible. Given where we’re at in time, and your understanding of technology, AI, and cryptography, this is really cutting edge.

Sam Green (00:43:43):

Yes. It’s super exciting for us to think about, if you were to think of the physical economy, let’s say the US economy, there are so many moving parts that it would be very difficult to apply AI holistically because of all of the complexity that it has and the real world, the physical touch points that the US economy has. But what is amazing about The Graph, it’s also actually touching a real economy, but this economy is kind of like a computer science economy, and we can actually encapsulate much of it and capture much of it algorithmically, and now, for the first time, tackle these economy-wide problems holistically with these AI algorithms. So that’s very cutting edge. We expect, I mean, I’m echoing something Brandon has said to me in the past, but we expect that what is happening right now in The Graph is going to be studied by other people.

(00:44:41):

You could imagine, for example, economists studying The Graph protocol, and learning and making observations and testing hypotheses on The Graph protocol itself. And they could test things that they would never be able to test, let’s say, in the US economy. So I really think we are just getting started with the cutting edge R&D that we’re going to see come out of making improvements in The Graph protocol.

(00:45:08):

I think we can see some sort of analogy and get some sort of hints at what may be coming for The Graph protocol, by going back and looking at some other utilities of the past. My favorite utility to study is Bell Laboratories, or Bell Telephone who then created Bell Laboratories. Bell Telephone was simply a way to let people do voice communication. That was the purpose of the company. And then they created Bell Laboratories to improve their voice communication utility. And what’s awesome about Bell Labs is that, through that pursuit of improving the voice utility, they invented so many things. They invented information theory, which is heavily used by cryptography, especially when you’re analyzing the strength of cryptographic systems. They invented coding theory, which is also related to cryptography. They invented the C programming language, the Unix operating system, the integrated circuit, and then hundreds and hundreds of other inventions, just through improving a utility.

(00:46:17):

So The Graph represents a brand new utility, just like voice communications represented a brand new utility. And we are going to see, I believe, so many unexpected problems. And when I say problems, when an R&D person says problems, that’s a good thing for an R&D person, because they want to solve, that’s what they live for, is getting problems out of the way. So The Graph, as we want to improve The Graph, and as we improve it, we’re going to get to some level. And then once we get to some level, we’re going to say, “Oh, okay, now we would like to get to another level. And to get to this other level, we need to solve these problems.” And let me tell you, from an R&D perspective, I am convinced that this is going to be one of the richest problem spaces in crypto is going to be found within The Graph, and that’s going to attract more and more talent to The Graph because people are going to realize, there’s a lot of good work for us to do here.

Nick (00:47:48):

Incredible stuff, Sam. Thank you so much for sharing. And I’ve caught the vision for a lot of what you’re saying there. Turning more towards the cryptography side of your core development work, how would you describe that for listeners?

Sam Green (00:48:00):

Well, before I say anything about our cryptography efforts, I’d like to give recognition to Jackson Blazensky at Edge & Node. He has been mapping out the direction that we are now pursuing for several years, and he has really laid the groundwork for our team to come in and start working with him. I believe that this topic of cryptography, this story deserves further expansion, and I think Jackson himself would be a great person to tell that story on the GRTiQ Podcast.

(00:48:29):

But in terms of what we’re doing at Semiotic, specifically, we’re working on zero-knowledge proofs. I’d like to give some intuition about the types of problems, or the type of problem, that ZKP can solve. So at a high level, ZKP lets us do two things. It lets us prove that somebody else knows something that they say they know, or that somebody else has done some work that they claim that they have done. Well, why is that important in the context of The Graph or Ethereum?

(00:49:05):

Well, I’ll start with a more general use of ZKP, and then I’ll tie it to Ethereum, and then we’ll tie it to The Graph. Imagine you had some tiny computer, some very small low-power device, and you wanted it to do some sort of heavy duty computation. But your device just didn’t have the horsepower to do it. Well, with zero-knowledge proofs, you can outsource this heavy duty computation to a machine that has more resources than you have, like a heavy duty machine. And the heavy duty machine can take your inputs that you want processed, and it will send you a result. You could do that today, without any cryptographic technique like ZKP. But if you don’t have ZKP, then for most applications, you won’t really know whether or not the outsourced computation was done how you needed it done. They could have skipped steps, or they could have manipulated the result. ZKP solves this problem.

(00:50:11):

So the awesome aspect of ZKP is that, when you outsource a computation to the more powerful node, more powerful computer, in addition to it doing the computation that you’ve asked it to do, it will also generate a proof. And it will return both the result of the computation to you, and a proof of the computation, a proof that it did the computation correctly. And the key thing here is that you can verify that the prover did the computation like they were supposed to, and you can do it in a way that requires you less work than it would’ve required you to just do the computation yourself to begin with. So that is actually the key to the power of ZKP. Now, how does this tie into Ethereum?

(00:51:01):

Well, as you probably know, it is extremely expensive to run computations on Ethereum. And so now, Ethereum is our resource constrained computer. It’s the computer that doesn’t have the time or bandwidth to do the computation that you want done. So with ZKP, we can do computations outside of Ethereum. So you could do the computations, or some stranger could do the computations for you, and they can also make a proof, and then all they have to do, in the end, what they’re going to do is submit the proof to Ethereum. And now, Ethereum only needs the result of the computation and the proof, and it can more efficiently check that the outsourced computation was done correctly. So with ZKP, you can outsource a computation from Ethereum to some untrusted computer. It can do some heavy duty computation, and then submit the result of that computation, and the proof of the computation, to the Ethereum virtual machine, which can then just do the verification of the proof. And that is much cheaper than having the EVM do all of the work to begin with.

(00:52:14):

So now let’s tie it to The Graph. So The Graph is an application, on top of Ethereum and other blockchains. And with The Graph, we have many moving parts, and we want all of those parts to be done correctly. One of those parts is payments, payments from consumers that are using the dapp or dapps, to Indexers who are serving queries from the dapps. So what we want now is a trustless way to track who is owed for queries. And we’re currently working on this, we’re currently building this ZKP, zero-knowledge proof, based micropayment solution to do this tracking.

(00:53:00):

And I would like to tie this more concretely to the Ethereum virtual machine. Imagine you and I were making many transactions with each other. If we had to track these transactions in a trustless way using only Ethereum, then it would only make sense if we were doing thousands of dollars in transactions for each time. Otherwise… I mean, even then, the gas is so expensive. But when we’re talking about paying for queries, we’re talking thousandths of a cent for these microtransactions. So it doesn’t make sense to do this on Ethereum, and that’s why we need to do it in a more efficient trustless way, using zero-knowledge proofs.

Nick (00:53:43):

Well, Sam, this is really exciting stuff. To learn more about this core dev partnership with Semiotic, and for you taking the time to describe a lot of this, I mean, again, I said it before, it feels like a lot of cutting edge, new age technology. That’s so exciting. I want to ask a broader question about web3 now. Given your perspective of everything that’s going on, how do you think The Graph fits into that?

Sam Green (00:54:07):

I view web3 as the next version of the internet, and I think that’s a fairly common viewpoint. I would like to maybe take this question more from a philosophical angle first, and then I’ll come back around and specifically answer your question.

(00:54:28):

When we think of what the internet, how the internet was originally created, originally, the internet, it came after what was called ARPANET. ARPANET was the Advanced Research Project Agency Network. It was envisioned by the military to be this robust communications network, where people could still communicate, send data, even if one of the nodes went down. And then that kind of transitioned into the commercial internet, the web2 internet.

(00:54:57):

And then web2 internet was, as we all know, this is also a very common thing to say, it’s a very centralized thing. You have these silos of control of all of the big players, the big industrial players. And that’s not robust. For example, it’s easy for Google or Amazon to pull the plug on you if they don’t like what you’re doing, or Twitter to pull the plug on you if they don’t like what you’re doing. And so really, what that’s going against, that’s like anti-robust. That’s an anti-robust mechanism.

(00:55:29):

And so really, what’s interesting to me is, web3, it’s actually instantiating this sort of original vision of the internet as being this extremely robust thing that is incapable of being taken down. And from a engineering perspective, I’m so fascinated by this idea that we now have these programs that are out there, on Ethereum and other blockchains, that these programs are there for anybody to use. And there’s never been anything like this in the world. And what this does is it allows, in addition to anybody being able to use them, anybody can also audit them, if they’re open source, and most of the best ones are, and people can analyze them for weaknesses to see if they’re robust. And we basically can gain trust in them through this way.

(00:56:21):

And now we have The Graph sitting on top of this. And The Graph, of course, provides this fundamental utility of providing data. And data is a cornerstone for communications, for modern communications. And it’s clear to me that The Graph serves this fundamental utility for communications. And I think this was really driven home recently by Vitalik’s blog post on state expiry, and this is related to the EIP-4444 topic. In that post, in these posts that Vitalik made, he mentioned two possible solutions to the state expiry problem, and let me summarize the problem quickly.

(00:57:10):

Right now, if you’re running a validator node and the validator nodes are the ones that provide security in Ethereum, you have to sync a lot of data and a lot of historical data. And this causes a burden for people, new validators, who want to come into the network. And if it were possible to separate the responsibility of syncing historical data, separate it from the job of validation, then it would make it easier for new validators to come online, which would then add security and scalability to Ethereum. And in Vitalik’s recent blog posts, he only mentioned a single protocol as being a solution to that, and that was The Graph. And so to me, I’ve already considered The Graph to be a robust fundamental utility that’s going to be important in the future. And then when you see Vitalik mentioning The Graph, well, then it’s like, for me, it’s case closed. Let’s build. Let’s make this happen.

Nick (00:58:09):

Well, Sam, on this topic of EIP-4444, and it’s definitely been a headline grabber for The Graph community and a lot of people in blockchain, I remember when that blog post got some publicity. And I think you were the first one to break that. So how’d you come across it, and when did you let loose with that news?

Sam Green (00:58:29):

Well, I was on Twitter, as I often am, and I came across Vitalik talking about this state expiry problem. And whenever I have a chance, I like to try to read his thoughts. And so I dug a little bit and I saw him link to Reddit, and I was like, okay, let me read about this problem. And then I’m reading it and then I see, okay, The Graph is mentioned here. It’s only protocol that he’s mentioning, external to Ethereum, by the way. So I’m reading through Reddit and I’m like, huh, I haven’t seen anybody talk about this. And everybody likes to try to be the first person to reveal some alpha. So I thought, okay, let me post something on Twitter and let people know that Vitalik is talking about The Graph.

Nick (00:59:16):

Sam, before we pushed record, we were talking about this very thing. And in addition to finding this really cool blog post and sharing it with the community and getting a lot of hype around it, we were also talking about what Vitalik posted and EIP-4444 relates to this social scalability and cryptography. Can I get you to share some of your thoughts on this really interesting topic?

Sam Green (00:59:39):

You bet. So this is maybe more of a philosophical topic, and it was actually pointed out to me by Jackson Blazensky when I was working on the fully homomorphic encryption efforts. And there’s a blog post by Nick Szabo, S-Z-A-B-O, and it’s about social scalability. And what’s really interesting to me is, cryptography is all about information. We go back to, it’s about confidentiality, integrity, and authenticity. And this concept of social scalability, it’s related to the problem of trust.

(01:00:14):

So as people, we only have the mental capacity and the time to establish trust relationships with a small number of people in our lives. It takes time to develop deep trust between people. And what cryptography does is it provides these tools that allow you to extend trust, without spending the time to develop a trusted relationship. And that’s actually the magic of crypto itself.

(01:00:50):

So for example, let’s look at the most exciting example of social scalability right now, in the context of The Graph. Ethereum has to have a provable history that can be queried of past transactions. And now we see Vitalik saying, “Hey, look, we can use The Graph as a way to offload providing data for these historical queries.” Well, if we had to establish some sort of personal relationship between Vitalik and The Graph, I mean, that’s definitely not what we’re talking about doing. That’s not going to fly. What we’re talking about is a mathematically provably secure way that allows us to plug The Graph in, to Ethereum, and have zero doubt that the data being queried from The Graph is the same data that would be retrieved from the full nodes that would’ve been run on Ethereum. And to me, that’s awesome because we don’t have to trust that The Graph is going to provide data correctly for Ethereum. We’re actually going to have mathematical proofs that the data was provided correctly.

(01:02:06):

And so now, going back to the scalability issue, this allows more people to run validation nodes, because they’re going to have lighter computational requirements, and therefore, we’ll be able to scale greater than we would’ve otherwise.

Nick (01:02:21):

Sam, I want to ask you just two more questions. And the first I want to start with is your long-term vision for The Graph. So as you look out into the future and project The Graph’s impact in the world, what do you see?

Sam Green (01:02:34):

Well, right now, I think we’re only beginning for The Graph. I think that because what The Graph solves is so fundamental, and because the teams The Graph has gathered, and the community that The Graph has gathered, I see so much competence and passion in the people and teams involved in The Graph. I think that we’re going to see this protocol be everywhere in web3. And from an R&D perspective, which I can’t get away from because that’s what drives me as a person, and professionally, I think this is such a problem-rich environment.

(01:03:23):

The Graph in particular is such a problem rich environment. That means they’re going to be refinements upon refinements forever. And through these refinements and problems that we need to solve, we are going to invent some amazing things. And these inventions, these new zero-knowledge proofs, or these new RL, AI agents, they’re going to be… What’s really exciting to me is that, of course, we’re going to apply them to The Graph to make The Graph more efficient, to make The Graph easier to use for everybody, but also these are going to be open source.

(01:04:00):

And so we have this whole open source ethos, and these improvements that we all make, and these discoveries that we all make and learning how The Graph should operate, and how to make The Graph operate better and better, these are going to be open to the entire community. So the whole web3 community, we’re going to see The Graph helping many other protocols in the future.

Nick (01:04:20):

Sam, the final question then I want to ask is, I want to kind of go back in time on your personal story. You shared early on that you’ve done some work, digging in the ditch before. You worked for a corporation where, again, it wasn’t aligned with your values and your purpose. So how has your professional life, your career, your passions, changed now that you’re working on projects like Semiotic and The Graph?

Sam Green (01:04:45):

Well, what’s interesting is that, everybody that is in web3 could be doing something else with their time. And I consider myself very fortunate because the founders of The Graph, and also other founder, the founders of Ethereum and their early pioneers, the other web3 pioneers that actually drive the use case need for The Graph, they came in and they were like super early, when very few other people got it. So now, I feel fortunate now to be surrounded by these people that are very, I think, visionary, very visionary community. And not just visionary, but both visionary and highly skilled. So here we have these high skill, high vision, clear vision teams that are working towards something. And actually, what’s awesome is we really don’t know where we’re headed. We know where we’re headed in the next one or two years. Of course, we have crystal clear technical priorities for the next several years. But we don’t know what the long-term impact of what we’re building will be on the world. And I think it’s going to be significant. And so I’m so excited to be part of this.

Nick (01:06:12):

Sam, thank you so much for your time. You’ve been very generous and this has been very informative, to get to know the work that Semiotic is contributing in the core dev partnership with The Graph, and to learn more about you. If listeners want to follow you, learn more about your work, what’s the best way to stay in touch?

Sam Green (01:06:27):

So first of all, I’d like to thank you for your time. I’m a huge fan of the GRTiQ Podcast and it’s an honor to be on the show. If people would like to follow me, the best place would be on Twitter, @0xSamGreen.