Decentralized AI Training? How Sapien is Disrupting the Industry

This episode explores the convergence of Web3, AI, and human intelligence, offering a glimpse into a future where humans and AI collaborate to create more sophisticated and useful technology. Get ready to explore the Edge of Sapien and understand how human expertise is reshaping the AI landscape.

Key Topics Covered:

• Rowan Stone shares his journey from the crypto space to recognizing the critical role of data in AI, leading to the creation of Sapien. His experience at Coinbase and Base shaped his vision.
• Sapien aims to incentivize and reward individuals worldwide for contributing their unique expertise to AI models. By leveraging blockchain technology, Sapien ensures diverse perspectives.
• Ensuring High-Quality AI Training Data: Sapien employs a "proof of quality" mechanism inspired by Ethereum's proof of stake to maintain data integrity. Participants stake Sapien tokens, earning rewards.
• Rowan emphasizes Sapien's focus on the data portion of the AI stack, particularly high-quality, expert-level data. Unlike companies focused on simplistic data labeling.

Episode Highlights:

• "Essentially, we want the world's experts to be rewarded for helping AI to become more sophisticated and ultimately become more useful for everyone everywhere." - Rowan Stone
• "We can make this much more ethical, we can make this much fairer, and ultimately we can get better quality data while we do it by distributing the work and allowing anyone anywhere to participate..." - Rowan Stone
• "We are incentivizing people to provide good quality data and disincentivizing them to provide bad quality data... If they do good work, they get rewarded... And if they do bad work... then they have skin in the game and we can slash them." - Rowan Stone
• "We're focused on providing the highest quality data. And then within the data sector, we're really trying to focus on bringing high quality expert data." - Rowan Stone

People and Resources Mentioned:

• Rowan Stone
• Josh Kriger
• Sapien
• Coinbase
• Base
• Ethereum

About our Guest:

Rowan Stone is the CEO of Sapien, bringing a decade of technical and operational expertise in the blockchain and crypto ecosystem. Before founding Sapien in 2023, Rowan held key leadership roles at Coinbase and BASE, driving strategic growth for critical blockchain projects. His journey spans engineering, energy, and Web3 sectors, with notable experience as COO of Total and co-founder of Horizon Labs. Rowan is dedicated to building and scaling innovative technology ventures and is passionate about redefining how humans and AI collaborate.

LinkedIn Link

Website Link

Twitter Link

‍

Transcript:

Rowan Stone: Hi, this is Rowan Stone, CEO of Sapien. At Sapien, we're not just building an AI platform, we're creating an ecosystem where human intelligence gets the recognition it deserves. You're tuned in to The Edge of Show, where blockchain, AI, and human intelligence converge to catalyze extraordinary ideas and conversations. Stay tuned.

Josh Kriger: Hey, Web3 Curious listeners, get ready for today's episode to discover how Sapien is revolutionizing AI training by creating the world's largest decentralized network of human data labelers. And uncover Sapien's approach to ensure there are always three chess moves ahead of the bots. Plus, what our guest today recommends is a personal use case for AI that will save you tons of time. All this and more at the edge of Sapien. Cue the intro.

Intro/Outro: Welcome to The Edge of Show, your gateway to the Web3 revolution. We explore the cutting edge of blockchain, cryptocurrency, NFTs, ordinals, DeFi, gaming and entertainment, plus how AI is reshaping our digital future. Join us as we bring you visionaries and disruptors pushing boundaries in this digital renaissance. This show is for the dreamers, disruptors, and doers that are pumped about where innovation meets culture. This is where the future begins.

Josh Kriger: Welcome to the Edge of show featuring a variety of top notch guests and other hosts. I'm Josh Krieger. It's another production of the Edge of company, quickly growing media ecosystem, empowering the pioneers of web three tech and culture and responsible other groundbreaking endeavors like the outer edge innovation festival in LA and Riyadh. Today's sponsor show features Rowan Stone, who brings a decade of technical and operational expertise in the blockchain and crypto ecosystem. CEO of Sapien, he has previously held key leadership roles at Coinbase and BASE, driving strategic growth for critical blockchain projects. His journey spans engineering, energy, and Web3 sectors, and he has notable experience as COO of Total and co-founder of Horizon Labs, demonstrating a consistent track record of building and scaling innovative technology ventures. Sapient, founded in 2023, is building a global network of human data labelers to support high-performance AI models. By combining gamification with blockchain incentives, Sapient ensures high engagement and top-quality data for industries like healthcare, Web3, and education. With a mission to make data labeling rewarding and accessible, Sapient is redefining how humans and AI collaborate. That is a mouthful. Rowan, did I get all that about right?

Rowan Stone: Yeah. I mean, pretty close for sure. About right. Great to be here. Thanks for having me, Josh.

Josh Kriger: Perfect. Yeah. Really, really glad to have you on the show. Been sort of following along closely with what you guys are doing and looking forward to unpacking it here today. Where are you joining us from?

Rowan Stone: I'm in the Netherlands. I've been here since February last year. This is new home.

Josh Kriger: Oh cool, so a lot of organic produce and great cheese and all those good things in your world, huh?

Rowan Stone: I'm originally from Scotland. I think I'm the only Scottish person who doesn't like whiskey particularly, and at least not Scottish whiskey, although I do like a bit of bourbon. But I say this because I'm probably also the only Dutch person who doesn't like cheese. It's just not my favorite. But I am in the land of cheese. They love it here.

Josh Kriger: Well, there's a lot of other good stuff there. Good seafood, and yeah. I've actually done a little bit more vegan cheese these last few years and sort of experimented there. It's not my favorite thing in the world, but it's definitely pretty darn good over there. At any rate, you know, you've had such an incredible journey from engineering to energy to blockchain to AI as we kind of got into a little bit. I'd love to sort of learn more about whether there's a specific experience that made you realize the potential of combining blockchain with AI. think there probably was.

Rowan Stone: I think for me, I've always been a bit of a technology nerd. Anything that's net new and shiny and interesting is where I love spending my time to explore. And so crypto was naturally attractive for me in the early days. And as I've kind of progress through my journey of kind of building a variety of different things in the crypto space, it's become pretty clear what crypto really can do. And it's bringing lots of different systems or people together from lots of different parts of the world and creating real digital immutability. And there's lots of powerful things you can do when you kind of align both people and that kind of tech stack. And one of the key things that's a major unlock for where we are as civilization today is assembling people to help train AI. And so AI has been super attractive for me for exactly the same reasons that crypto has, and I've been following pretty closely and an avid user of basically every model that's came out. And over time learned that really these things are powered by data. And the best way to amass net new unique data is by assembling a large group of people and hey, Crypto is pretty good at that. And so that was kind of the moment where it started to make sense for me to be actively involved in something like Sapien.

Josh Kriger: Cool. So if I got this right, you've been sort of following along AI and sort of dabbling with AI for a while. And then you sort of learned about the power of blockchain and sort of decided to match them together. Is that sort of the genesis?

Rowan Stone: No, it's the reverse, but it's exactly correct. It's just the wrong way around. I've been in the crypto space for quite a number of years at this point. AI has been pretty new for me, something that I was interested in personally, and not something I planned to work in initially. I was at Coinbase for a number of years. I took a step back for some family reasons, and my plan was actually to do nothing. But over the time of doing nothing, I quite quickly became bored. an opportunity presented itself to work with Sapien. And when I learned about what Sapien was doing, it just kind of made sense. Everything that I'd been doing with Coinbase and Base was about trying to bring a tech stack that had traditionally been used by all of us kind of basement dwellers to a much more mass market audience. Like let's bring technology that's net positive to people that actually are going to benefit from it. and kind of changing the way we did that. I think a lot of people have put a lot of effort into trying to explain this stuff. And for me, that's just not the right way to do it. Typical person doesn't care how these things work. Ultimately, they just want to get value. Like they want to understand why they should use it and they want to get into it in the cleanest and simplest way possible. And so we're taking this same ethos that we applied at Coinbase and trying to apply it to what we're doing now with Sapien. Let's make it really easy for people to step in. and really easy for folks to start participating. I'll kind of pause there because I'm kind of getting a little bit ahead of ourselves. It's probably better we set some context in terms of what Sapien is and why we exist before I start talking about why we're doing things the way we are.

Josh Kriger: Yeah. I mean, I, I, I, you got kind of into it, which we were going to get into, but that's fine. Um, I think, I think that was, that was fine. And, um, uh, cool. So, so, uh, let's, let's start here. So yeah, that all, that all makes sense. And I, and I appreciate sort of the journey that got you there for me, I'd experienced some data visualization sort of, um, work, uh, in a former life and in government and then got into blockchain. And, you know, data visualization, which is sort of a lot of it has to do with AI and quantum machine learning, kind of took a backseat. And then, you know, chat GPT pops up. And before you know it, you know, I'm an avid sort of AI user, it's like my best friend. you know, and if I have a question or a problem, it comes to the rescue, and I don't have to pick up the phone and call someone, right? So, and our whole team has sort of adopted it now. It's been pretty amazing, and it's becoming ubiquitous in society. So, appreciate sort of how you saw this great opportunity here, There's more to it, though, in terms of what Sapien is and how it's sort of evolving into this decentralized protocol. Can you walk us through how the blockchain technology enables rewards for human expertise in AI training? Absolutely.

Rowan Stone: So taking a step back and setting a tiny bit of context, the reason that Sapien exists is to Figure out the right incentive structure to have humans all around the world, everyone everywhere, from creatives through to engineers, lawyers, doctors, mathematicians, you name it. lean in and actively provide their nuanced understanding of the world, their unique expertise in whatever field it is they've spent their time, through models, so that models can go deeper in conversations that they have with everybody around the world. Essentially, we want the world's experts to be rewarded for helping AI to become more sophisticated and ultimately become more useful for everyone everywhere. Today, there isn't a proper protocol or real incentive structure to make that work. Typically, data is amassed through very centralized facilities in varying parts of the world, everything from the Philippines and Bangladesh through to Africa and beyond. And these are essentially giant call centers where you have thousands of people behind desks manually doing either data structuring, data capture, or any variety of data creation type works. And then this data is sanitized and checked for quality to make sure it's accurate and it's truthful Obviously, the models don't therefore take inaccurate information and base their logic upon it. And once they've figured that out, the data is then used for inference and training. And that's where you kind of get into the algorithm and the compute side. But our core thesis here is really simple. It's 2024, 2025 now. we have technology to do this in a much more efficient way. We can make this much more ethical, we can make this much fairer, and ultimately we can get better quality data while we do it by distributing the work and allowing anyone anywhere to participate, which means that we're not going to have local geographical biases or cultural biases or anything along those lines. We're going to be able to source any information that any enterprise or model or AI agent needs in order for it to advance. And so that's exactly what we're building here. The big problem is that when you're doing it centralized, it's really easy to make sure the data's of high quality. You can literally have people breathing down their necks and making sure, like manually watching, that the data capture, data entry, structuring, labeling, annotation, whatever type of work it is, is being done correctly. When we allow everyone everywhere, like pick up your phone, do it on the bus, do it between Uber journeys, or if you're a doctor, perhaps you're doing it in between patients to earn some extra money, it's much more difficult for us to make sure the work is being done correctly. And so that's why we employ on-chain technology. And the simple way to think about this is just aligning incentives. We are incentivizing people to provide good quality data and disincentivizing them to provide bad quality data. How do we do that? Well, we borrow from one of the largest on-chain networks in the world, Ethereum. They've figured this problem out for us through proof of stake. We're calling ours proof of quality, but a lot of the mechanism is the same. Essentially, everybody that's participating needs to have some skin in the game. That's Sapien tokens that are staked in our system. And if they do good work, they get rewarded. They get their base compensation for the work that they're doing. They also get a multiplier for kind of having that skin in the game and being good, being kind of a high quality, contributor. And if they do bad work, if they cheat, if they try and Sybil or anything like this, then they have skin in the game and we can slash them. So they kind of have that economic alignment with the protocol and deliver only good work.

Josh Kriger: Makes sense. So I appreciate the differences between this model and some of the traditional web 2 models. However, you guys are not the only company in the blockchain space to be thinking about this problem and how to solve it. There's also... layer ones like, like, I don't know what near is anymore. Is it a layer one? It's it's changed. It's evolved. But they announced a few months back that they were sort of building a massive blockchain, you know, open AI product. I'm curious, a how do you relate to that product? And is there a potential relationship there? And then B, how do you differentiate what you're doing relative to other companies that are trying to solve the same problem in the blockchain industry?

Rowan Stone: We are hyper-focused on the data portion of the stack. And so there's a whole bunch of different pieces that come together in order to make the foundational models that we talk to every day. And data is where we live. There's a bunch of other companies, some of them are doing some really, really cool stuff. It's an exciting space to be in. However, not actually that many are focused exclusively where we are. And I think the way that we differentiate even further is not just the protocol and kind of trying to create an open system to allow kind of self-serve and anyone to participate, but also where we're focusing in the kind of complexity curve. And so data labeling is a very well-known industry term within the kind of AI world. And what data labeling traditionally was, was like draw a bounding box around the AirPods, for example. And this for us is like very simplistic annotation type work. It's really low value. And in our opinion, it's going to go to zero very quickly. And the reason for that is that computer vision and just the technology that we've built already is very much capable of doing the vast majority of this with a super high degree of accuracy. And so humans are just not really needed there anymore. And where we see the puck going is kind of towards much more sophisticated kind of expert level data. And that's where we're really focused. So A, we're focused on the data space. We're not focused on compute or algorithms, inference training, or building our own models. We're focused on providing the highest quality data. And then within the data sector, we're really trying to focus on bringing high quality expert data. And so that could be things like helping train a math GPT for children by pooling in teachers that understand how to teach mathematics. Or it could be pooling in oncologists and radiologists to help figure out how to teach a model to recognize cancer and radiography images. It could be pooling in lawyers to help kind of make a model more robust or more updated in terms of latest nuances and cases and law from particular regions in the world. Lots of different things. However, hopefully it's given you a flavor of like where we want to be in the complexity curve. And this is why it's so important for us to build a giant network of people. Because the only way we can be truly useful to someone that's building a model or an AI agent that wants to improve itself, I kind of see that as being where we're going, is for us to be able to service essentially any need they may have. And the way to do that is to amass a giant group of humans, and then to be able to lean into that group and understand who's reputable and who isn't, and understand who has the correct credentials for the tasks that you want to have done. And so that's exactly what we're building here.

Josh Kriger: That's fascinating and it really does clarify things for me and I think for our audience. I guess the natural next question is, let's say I'm an expert in a rare field or aspect of a particular industry like nanotechnology. By sharing my expertise in training this model, are the incentives that are in place for this, uh, long-term or short-term or, or it depends. And in my sort of front running, creating my own version of a model based on my expertise that I have full IP rights to, and the ability to sort of deploy on my own, um, where these rare experts are thinking through the pros and cons of working with companies like yours versus other opportunities to sort of clone their expertise and have it live on forever. So I have another question, but I'll pause there for your thoughts.

Rowan Stone: I think the missing piece of the description of what Sapien is, is that we're essentially building a marketplace. And so, If there's an enterprise company on one side that needs a specific piece of data, Sapien's mission is to connect them with the right person, the right member of society, the right expert, whatever you want to call this person, to provide that knowledge. And it could be as simple as draw some boxes around an image, although maybe recognizing where the box should be could be quite specialized in the case of like cancer screening, for example. Or it could be something completely different. It could be kind of audio speech recognition. It could be helping with translation in really obscure languages. There's kind of a little bit of something for everyone. But the key here is that it's a marketplace. And so we're connecting supply and demand. And we're taking a small cut to facilitate that kind of connection. And we're providing the protocol and the logic to facilitate that kind of connection and that knowledge transfer. And so what we're not doing is actively training models ourselves using knowledge from users. And so we're not saying like, come and provide medical expertise, and then you will have a medical GPT that you can then go off and monetize. We're simply saying, come and provide your expertise to this particular task. We will pay you for your expertise and your time, and you'll be rewarded based upon contributions and time and kind of accuracy and things like this. I can get into that in more detail. And so that's really how this system works. It's a protocol to connect experts with enterprise.

Josh Kriger: That makes sense. Yeah. I, I'd like to know more about how you're rewarding high quality expertise and, and also, um, you know, how, how the pricing and sort of set up works. I would assume for rare expertise, whether there's a finite amount of individuals that can provide it, uh, that the market dynamics would require, um, paying more of a premium for, than more expertise that a larger population has, and there's some kind of gamification elements there. Could you elaborate on that? Yeah, happy to.

Rowan Stone: Again, let's use the AirPods as an example. Draw a box around the AirPods, you're not even gonna get minimum wage. Like this is something that is no longer valuable to the majority of these models. However, some people still require it. And so we still occasionally have customers that want us to do what seems to be relatively simplistic kind of image-based recognition work, but it's kind of a minority of the work. And really the customers are setting both the requirements and they're setting kind of the premium for the task. We're providing feedback. And so if we have a customer that rocks up and says, hey, we want some super senior board certified oncologist and we want to pay $20 an hour, like clearly we're going to recommend that there's a massive disconnect in terms of their expectations and that these people are not going to engage for that amount of cash. But typically, customers coming in, they're setting the demands, they're speccing out the data that they're looking to secure. Whether it's, we have a data set, we want you to help us structure it and label it, or we want to improve our model in this particular area. And to do that, we need this net new data set with these types of qualities. And then we would go out and source that for them. So that's typically the way it works. And you're right. It's very simply, the more sophisticated the work, typically, the more expensive it will be.

Josh Kriger: So I guess at the end of the day, is there a moment in time where human expertise becomes obsolete or extremely rarely needed just based on the acceleration of these models and how much information they're soaking in? I'd like to think that I have unique expertise in the world that will continue to be valuable forever. But part of me wonders, is that really true? You know, eventually these models have enough information to be able to sort of do anything I would be able to do where, you know, my time is best spent hanging out on a beach or spending time with family. What are your thoughts there?

Rowan Stone: a really interesting question and I think anybody that gives you a definitive answer is just talking nonsense. The truth is that we don't know. The speed of development across the entire industry is incredible. The demand for data is absolutely nuts. Is that demand for data gonna be there in perpetuity? Like, we don't know. I don't think anybody really knows. I think there's certain areas where the likelihood of demand for data being consistent and kind of stable is super high. And there's other areas, like I said, like the draw box around this, I expect to go to zero very quickly, if not already. But I think there's lots of things that require nuanced human understanding, that require human opinion, and that require us to kind of steer from an ethics perspective or steer from a right versus wrong or any number of types of things. One good example here is autonomous vehicles. It's a big chunk of the work that we currently do. There's a ton of different companies out there. It's basically an arms race among all of the non-Tesla autonomous vehicle companies to try and catch up with Tesla. Realistically, they've absolutely smashed out of the park. They started moving before everybody else. People have realized that electric vehicles are actually viable if we make the infrastructure. And they've also realized that autonomous driving is like a bit of a future unlock in terms of productivity and safety and a variety of different things. So there are tons of companies out there who are super innovative and have thousands or hundreds of thousands of cars out in the wild doing millions of kilometers every single year. And these cars are all encountering weird situations on an almost daily basis. where they have no idea how to deal with the situation. They've seen millions and millions and millions of miles of driving, but they're always going to encounter something new, and they're very likely always going to require a human to help them understand what they should do in that particular scenario. Autonomous driving, I think, is one area where data from humans is going to consistently be needed. But coming back to my initial opener here, we just don't know. And I think that's the right way to frame this. If you look at a long enough time horizon, it's anyone's guess. If you look at the next three, five, maybe even ten years, I think high likelihood that humans are going to be required for the vast majority of things. And my reasoning here, without going on a tangent, is that the AI space is kind of repeating what the crypto space did in that everybody is incentivized to build their own thing and to build their own thing in isolation. And what that means is that we don't have one giant technical standard that we're all working to improve and we're pouring data and compute and different advances into to get it to the point where it just does everything. Instead, we have tens of thousands of completely independent models that don't talk to each other and don't share data with each other. And because of that, we're going to need to do the same work tens of thousands of times. And that in itself is going to create huge demand for data for the foreseeable, for absolute sure. But it's unfortunate. And I think it's something that we need to tackle. And I think the obvious analogy here, if we think about a world in which early days internet, we didn't all settle on one technical standard. We decided TCP IP is dumb, let's make 50 versions of it. I don't think we'd be having this conversation in the way we are now. It would have been a very different world, but that's exactly what the crypto industry has done. They've said, Bitcoin's not good enough because I can't put an application on it easily, and it's difficult for us to control because the miners have control over what happens. got my own. And then everybody's just joined that bandwagon because the incentives are such that it makes sense to do so. And now we have thousands of chains that are completely separate and don't connect and don't talk. And so I think a level of standardization is going to be a big part of the future story. And it's actually something that we at Sapien are actively trying to engage with. A part of my work when I was at Coinbase, helping launch base was specifically pointed at. Let's try and create technical parity in the industry. Let's bring open source technology to market that everybody can get behind and make better. And if everyone starts building on this shared technical standard, then hey, we've got a much cleaner path to interoperability. And then all of these kind of user experience pain points that everybody's trying to design around that are literally just because we've created this fragmented monster of an industry, start to go away, because we can just have things all communicate together. And I think AI will eventually come to that same conclusion, and we will see a kind of amassing of models. And we'll also see a daisy chaining of things, which is what we're seeing with agents today, whereby you can kind of start with a prompt in one place, polish a prompt, perhaps use a different model to go out and source and do research, and then perhaps use a third model to collate or do some actual tasks. I digress. My point is, we don't know. But for the foreseeable, data is the fuel that is driving the innovation forward and the key unlock that makes this stuff useful for us.

Josh Kriger: Yeah, I appreciate all those points. And I didn't necessarily anticipate you'd have a definitive answer. It's a very sort of It's an evolving space and I think what may happen in this area remains to be seen. This is good fodder. I'm actually moderating a panel, a closing panel at the LEAP Technology Conference in Riyadh with a electric vehicle CEO among some other folks. So you've given me some things to think about there and to sort of steer that conversation. So I appreciate your perspective. I wanted to talk a little bit more about the gamification element you mentioned previously. And I'm sure you've thought a lot about potential risk and sort of ensuring fair competition. fair compensation. I can think of the possibility that someone tries to use an outside AI tool to sort of provide expertise that you intended for a human to provide. What are sort of the challenges you've run into as you've built this product and how are you ensuring, you know, quality?

Rowan Stone: Huge question. Okay. Where to start? Let's start at the absolute core. The very first thing that we need to know when we're engaging with a net new person is that that person is human. And so a big part of the onboarding is trying to figure that piece out. What we don't want is tons of friction. We're using on-chain tech for quality, but we all know that on-chain comes with tons of complexity and friction and most people just don't want to care or don't want to learn. And so we need to kind of get rid of all of that. But at the same time, we need to get to the point where we're confident that this person is a human or we treat them as kind of a bot until proven otherwise. And so what we do, we onboard somebody, shout out to Privy. Privy are awesome. They enable the spinning up of a wallet very quickly just with an email address. And so that gets rid of the complexity of like connect a wallet, learn a seed phrase, all these types of things. And then when the user first engages with our app, we kind of assume that they're a bot. And so when they're providing inputs, they're going to have a Cloudflare pop-up that's just verifying that they're not a bot every single time they're doing something. And it's looking for types of behaviors and things like that. If they can then give us a bit of confidence that they're not a bot, i.e. click on a little icon at the top right of the screen that says verify with Worldcoin, or verify with World, I should say now, then we start removing some of these preventions and we kind of make the interface a little bit slicker. For anyone unaware, World is a global network, working on proof of humanity, and working on a variety of other things around that. And we're leaning upon them to help verify that users are individuals and not someone running 5,000 accounts. And this kind of tiering in the world system in terms of kind of surety that this person is a person. The first level is just clarifying that they have a unique device and so they can scan a QR code in our app and it clarifies that for us. That's like the basic verification. The second tier is they've done some KYC and so they've provided like a passport or an ID card two worlds, nothing to do with us. And so they can just basically flash up their world passport to the app. Again, privately, they're not sharing their details, they're just proving that they have done that. And so that gives us more confidence than a person. And then the third tier, which we're not interested in yet, it's kind of a little clunky at this stage, like world are super ambitious and we'll make this slick in the future, but they're not quite there yet. And that's like iris scanning. And so there are what they call orbs in different countries around the world and users can rock up and they can scan their iris. And that RFIs within the world ecosystem that they're a unique human and deploys a wallet for them. And so we won't get into too much detail there, but ultimately step one, are you human? Yes, no. And then step two, is to monitor the quality of what the person is doing. And so we have a QAQC system. We work on a mixture of kind of manual review, some automation, but then also some peer review. And ultimately QAQC feeds, as in like the quality of the inputs for a user, feeds into what we call the reputation score. Now reputation currently is centralized off chain in our systems, but future state and what we're working on just now is bringing reputation on chain and making it something that we can kind of pull in from other ecosystems so that your reputation is not just the sum of your kind of impact within Sapien, but it's also perhaps your impact at Coinbase or how you've transacted on chain or whatever. And so reputation is kind of the next big part of the story. And it's kind of a layered approach to quality as you'll kind of see as I explain. And so reputation feeds into what we call qualification. This is on a task by task basis. And ultimately, once you've achieved a certain level of reputation in the system, we see you as a trusted person. We know you're delivering good quality data. And so we let you get access to more kind of premium or sophisticated tasks where the compensation opportunities are much higher. And obviously the risk to us is much higher that you do bad work. And so we need to make sure that you have good reputation.

Josh Kriger: So it sounds like you've given this no thought whatsoever, and it's just like survival of the fittest. And if the bots win, so be it. No, I mean, that is quite a novel approach to the problem. And we've explored various technology around media over the years. And of course, creating more natural engagement is the goal, but these bots, they're ubiquitous and they're getting smarter and smarter as well. So I really appreciate the nuance of what you're doing there and how you are sort of solving this problem, or at least trying to stay three chess moves ahead of these bots, right? Which is the goal there. It's very cool. Just on my the other question in terms of, you know, ensuring fair compensation for expertise, is it your market dynamics? Or is there any kind of like, flooring that or sort of suggesting that you include in the process?

Rowan Stone: So we were originally going to have a floor and then realized that it just wasn't wise. And the reason it's not wise is that it prices out a bunch of work that some people might actually want to do. And I think it's unfair for us to just determine that nobody in the world wants to earn less than $3, for example. The reality is tons of people in the world want to earn less than $3 and are very happy doing simple tasks for little amounts of money, because to them, it's not a little amount of money. The US dollar is a very strong currency, and you take that to other countries, and all of a sudden, it's very meaningful. We do a ton of work with a ton of different people in a lot of different places. We have 195,000 contributors today, and they're coming in from 100 plus countries. So I think it would be unwise of us to put a minimum. However, the market is 100% in control here. And so if nobody is willing to do the work for the price that the customer sets, then clearly the customer needs to pay more and they'll just keep notching that up until somebody bites and somebody does that work. So every single contributor has the option to participate or not. And everybody will make their own decision in terms of what their time is worth and kind of what they want to engage with.

Josh Kriger: Makes sense. So we've hit on a lot of the challenges that were top of mind for me in terms of what you're embarking on. I just want to make sure we cover the full gamut of challenges and opportunities that you see here in terms of decentralized AI training and how everyday folks can be involved, 200,000 is a very large audience and there's many more folks that I think are gonna learn about what you're doing and jump on board over coming months and years. What keeps you up at night and what are you most excited about?

Rowan Stone: What keeps me up at night is probably where I'll start for me. I put myself in the shoes of a new person that's just listened to this and they go to game.sapien.io, they sign up and they jump in, I'd be a bit like, oh, because the first thing I'm going to see is five or six tasks and that's it. And the reason for that is that the vast majority of the work that we do, we have 17 paying enterprise customers including a bunch of very large businesses that, candidly, a new company like us has no right to do business with, like the Majornes, the Amazon Zoox, the Alibaba, the Baidus of the world. But majority of that work is in private today. And it's in private because we haven't fully fleshed out and matured the four key systems that we need to bring that work into the front end. And so what you see when you go to GameSapien.io is a really small kind of glimmer of what it will be in the future. And these are basically the simpler tasks that we are comfortable putting in the front end, or that the customer is comfortable having in the front end. Now imagine you are Amazon Zoox and you're training autonomous vehicles, you may not want other autonomous vehicle companies to see exactly what types of data you're requesting. And so you don't want to have that in the public eye. And we solve for this through these kind of four systems. It's basically make sure people are human, Make sure they have a good reputation, as in they're delivering good data. Qualify them for the specific task. Part of that qualification is like, do the NDA process really onboard them into the type of work that they're doing? Make sure that if they need credentials for the work, those have been screened. Make sure they're capable of doing the job. Perhaps there's some kind of pre-job training. All of that happens in this kind of qualification system. And then the last piece is a matching engine. So once we are able to do these kind of four things, the mission becomes matching the right demand with the right supply. And we're not there yet. And so that's what's keeping me up at night today. We have a little bit of work to do to get our QAQC system to the point where it's truly scalable. We're currently pre-token. And if you remember what I was talking about with the kind of QAQC, a big part of that layered approach is users having skin in the game. And so pre-token means we have to do a bunch of stuff that doesn't scale because we don't have that added incentive and disincentive. And so, short term, it's mature QAQC, it's get the token into the wild, and it's fully build out reputation and qualification, and then have them fully deployed, rather than being hidden away in a semi-closed product, like in the public. and for everybody to use. And that's going to be the big unlock that brings us from five or six tasks in the front end looking a little bit meh to everything that we're working on available to everybody, providing they've put in the work to demonstrate they're capable of doing it.

Josh Kriger: Yeah, it makes sense. So, you know, folks that join the Sapien journey have to have some patience and understanding that there's a back and forth tango here with their expertise and sort of the opportunities that will come on board. But being early pays off because you're going to sort of set yourself up for success in the long term if you buy into the vision. Are you able to share in any level of detail sort of the current maximum earning potential that you guys are tracking and the future potential?

Rowan Stone: It depends on your area of expertise is the kind of main thing I'll say here. But we've had people earning by hundreds of dollars per hour to do complex medical work within the system. So it's definitely not all low paid work. there's a ton of different nuanced understandings. And actually a kind of fun analogy that I kind of thought about a couple of nights ago that I'll throw in here. I presume everybody's seen The Matrix. It's one of the coolest movies ever. I mean, the first one, the rest of them, I'm going to pretend don't exist, but the very first one's awesome. And there's a question in that where they say, do you know how to fly this? But like, they're going to jump into a helicopter and they pause for a second and they kind of grab the data from The Matrix. They kind of look up and just go, I do now," and they jump in. I think that's the world that we're eventually going to get to, whereby an agent might be tasked with doing something, and it's going to have autonomy to go and figure out how to do it. It might hit a roadblock in that it just doesn't understand what it's looking at, and it might need to hunt for data to better understand that thing. I think a future state here could very well be a world in which we have demand popping up left and right from different sources. Having reputation, having some kind of skin in the game early means that you have access to whatever requirements are there. The other thing I'll say is that we're not just going to be purely reactive. A big part of the strategy here is to proactively compile datasets using expertise from specific functions that we kind of have some level of confidence are going to be useful. And so you'll start to see that coming live in the system in the relatively near term. And we're going to start with simple stuff like audio speech recognition in tons of different languages. And I think it's important to do these types of things first because it means that basically everybody can lean in and immediately be useful, which I think is really important.

Josh Kriger: Absolutely. My panel at LEAP is about a hopeful future. I for one am hopeful. What I've learned from our conversation is that as humanity has the opportunity to evolve faster and companies have an opportunity to make a bigger impact in society with using this type of AI training technology. I've noticed sort of how much faster the edge of company has evolved over the last year using AI to its advantage and oftentimes we're doing new things, providing new services where we're figuring it out with the help of AI and you know, sort of, you gotta stay with the times. I come from like a lean startup background where, you know, you're doing these two-week agile sprints and the goal is to fail faster, but faster could mean, you know, learning that what you're doing doesn't make sense over the course of one or two months or six months versus several years. And now we have an opportunity to engage AI and learn these lessons and make these choices, you know, in the moment and have, you know, basically a full sort of business life cycle of micro pivots and adjustments along the way that certainly is going to lead to new things for society, I think, in a very rapid manner. Any closing thoughts on my comment there before we move on to the next segment?

Rowan Stone: I couldn't agree more. I think the way I view this is collaboration over competition. And the reason I say that is that we've seen a ton of positioning around AI is taking human jobs, AI versus humanity, AI will wipe out humanity, all of these like pretty crazy thought chains. And I think the future I want to see and the future that I believe is much, much more powerful is humanity and AI, which is a tool at the end of the day, collaborating and using the kind of knowledge that we've amassed to augment our work and to accelerate progress, to accelerate sewers for disease, or to slow down aging, or to accelerate development of new technologies. I am a sci-fi geek, right? Behind me you'll see some Star Wars stuff, some Dune stuff, some Star Trek stuff. I want to see a future that looks like this, and I feel like we're starting to see the green shoots of growth in terms of what's possible when we dump humanity's collective intelligence into a learning algorithm and create a foundational model with it, and then iterate constantly, and then be able to have anyone and anywhere have access to essentially humanity's knowledge, For me, that's the way we're going to get the kind of sci-fi nerd dreams of Back to the Future hoverboards and all these types of things. It's not by competing, it's by collaborating. I think the same thing is true within industry as well. We need to work together to make this work rather than trying to butt heads and do our own thing.

Josh Kriger: Those are some great closing thoughts. Thank you so much for sharing what you're up to at Sapien. I for one am very hopeful and excited about what you're doing and where it can lead to next. And we'll have to have you back on the show for some updates in the future about the latest sort of Roblox challenges and advancements of your product. All right, Rowan, since you have such a unique background, I'm excited about our next segment, Edge Quick Hitters, which is a fun and quick way to get to know you a little bit better. There's going to be 10 questions. We're looking for just a short or single or few word response, but feel free to expand if you get the urge. You ready?

Rowan Stone: Let's do it.

Josh Kriger: All right. And you cannot use AI to help you. This is a human task.

Rowan Stone: I'll do my best.

Josh Kriger: What is the first thing you remember ever purchasing in your life?

Rowan Stone: Uh, music. I don't know exactly what, but for sure music.

Josh Kriger: Okay. Was there a genre that struck your fancy first?

Rowan Stone: I used to skate when I was pretty young. And so for me, it was rock and metal all the way.

Josh Kriger: Nice. So maybe some ACDC or some Aerosmith or Guns N' Roses or something like that.

Rowan Stone: Yeah, I mean, something like that, for sure. My mom and dad were Led Zeppelin and like Deep Purple, stuff like that. So I've always enjoyed things along those lines.

Josh Kriger: Nice. My first concert was Better Than Ezra, which is kind of, you know, a little bit milder version of All The Above, but I just liked them because it was my first concert. So I got into them for a while. What is the first thing you remember ever selling in your life? A pair of rollerblades. OK. Did you get another pair or did you hang up the skates?

Rowan Stone: No, this was upgrade. This was, yeah, evolving in my journey, I guess. I did a bit of street skating and then I did a bit of park skating and it was a ton of fun. And yeah, I sold a pair of skates to buy a new pair.

Josh Kriger: Any major wipeouts along the way?

Rowan Stone: Tons. Tons. Less so skating than I have in motorbike land, which is kind of my adult passion, if you like. But yeah, I've had a few run-ins with tarmac. Not ideal.

Josh Kriger: You know, it's a good analogy for business, right? It's not all roses. There's good days and bad days and challenges. So I think sports and business go hand in hand. What is the most recent thing you purchased?

Rowan Stone: Most recent thing I purchased? A bag, very uninteresting. Like a hold all bag to take on vacation week after next.

Josh Kriger: Cool, where are you going?

Rowan Stone: Tanzania.

Josh Kriger: Wow, that's exciting. Some safari action?

Rowan Stone: Absolutely, yeah. I've never been and always wanted to, and so gotta be done. Can't wait.

Josh Kriger: Great, I'll look forward to those pictures. What is the most recent thing you sold?

Rowan Stone: I've actually got a small pile of things that I'm supposed to sell and just haven't had time to. I haven't sold anything in a very long time. And if anybody wants to buy stuff, like I have a bunch of stuff. Nothing sold recently. I should really pull my finger out and get it done.

Josh Kriger: All right, well, just do an X post and anyone who's in the Netherlands can pop by and clean out your storage unit. What is your most prized possession?

Rowan Stone: Candlesticks from my nana is what I would call her, but I guess grandmother to the majority of people that are listening. Worth absolutely nothing, but the sentimental value to me is huge. Couple of wooden candlesticks.

Josh Kriger: Beautiful. If you could buy anything in the world, digital, physical, service, experience that's currently for sale, what would it be?

Rowan Stone: Ooh, blank check style. Cool question. I'm going to be very specific. A Pilatus PC-12, a turboprop airplane. I'm learning to fly and it would be my dream plane. Yes, please.

Josh Kriger: Cool. How many seats does that one have? Eight seats. Nice. I, I went on a plane last year from Utah to LA that almost was like autonomous. So semi-autonomous and it self-corrected its, um, its flight path, uh, in such a way that it was like the smoothest flight I've ever had. I expected, you know, for a four seater, I think four or six seater, it was going to be bumpy, but it was like, I was sleeping like a baby most of the flight.

Rowan Stone: Nice, you must have been lucky with good weather. I mean most planes have autopilot now and so once you're up in the flight levels the work for the pilot's pretty low, at least in the bigger planes, even with some of the smaller ones. But yeah, very cool.

Josh Kriger: The next question of course we all have to decide is would we fly in a plane where there was no pilot?

Rowan Stone: I'm a pretty crappy passenger, car, motorcycle, whatever. I feel like I'm a pretty good passenger in a plane. I'm not convinced I'm ready for pilotless planes just yet. I did the Waymo thing in SF recently. That was pretty weird. I mean, I was genuinely nervous, even though I fully understand the tech and I fully understand how safe it is.

Josh Kriger: Yeah, I was crossing a crosswalk with the Waymo coming at me and, you know, a little bit of anxiety there. Like, I'm sure this thing is going to stop, but what if it doesn't? It's just, yeah, I'm not quite ready for that. All right.

Rowan Stone: This is why we should be nice to the models. We should be nice to AI as it slowly becomes intelligent. We don't want to annoy. We don't want to have future things attacking.

Josh Kriger: 100%. Yeah. If you could pass on one of your personality traits to the next generation, what would it be?

Rowan Stone: It's a good and a bad. I'm pretty stubborn. It means I don't give up easily, but it also means I'm a pain in the ass. But I think it's important people are stubborn. Like if they want something, they want to make it work, they'll find a path and they'll make it work.

Josh Kriger: Persistence and curiosity. That's what it's all about. If you could eliminate one of your personality traits from the next generation, what would it be? Oof.

Rowan Stone: My complete lack of patience. I think people should be more patient generally, and I think we'd be better as a species if we were more patient in almost everything that we do.

Josh Kriger: Yeah. I'll cheer to that I'm with you there. Let's, uh, let's Chuck, Chuck our impatience out the window, uh, and try to leave it there. Um, what did you do just before joining us on the podcast?

Rowan Stone: I was back to back to back calls with the team, uh, calls with an investor. just basically figuring out how we accelerate product and get to the point where what I talked about earlier, all that awesome stuff that's hidden away in the back is front and center for everyone to see.

Josh Kriger: Right on. And what's next after the podcast?

Rowan Stone: It is 7.35 PM here in the Netherlands. I haven't had proper lunch and so it's going to be a very late lunch.

Josh Kriger: All right, I hope it's a good one. I have a bonus question for you. We always like to ask something that is stirred by the conversation. What is one task that you use AI for regularly that is a life hack folks may not realize exists?

Rowan Stone: I don't know if it's a life hack, but it's probably the thing that I use most frequently. And this comes back to the patience thing. I don't have any. And so if I'm presented with a huge document, I'll quite often copy-paste that into AI, depending on what it is, obviously. And if it's not something that needs to be kept private, then I'm going to get an executive summary from AI, and I'm going to basically do the kind of Blinkist version of whatever I'm trying to absorb. It's awesome. It's really good. The quality over the past even three to six months has taken a massive leap. And so now it's very reliable.

Josh Kriger: Yeah, I love that one. And I'll also tell AI to act as my general counsel when reviewing a document and give me all risks and modest and more assertive recommendations for corrections for each clause that is questionable. And then I choose whether I want to go a little bit more light or guns blazing.

Rowan Stone: You know what, I'm going to add one here. For the majority of people, Google is no longer the best search, like it just isn't. They're doing a great job of implementing AI search results, but right now searching via a model is a massive life hack. Pretty much any question you can think of, you're gonna get a more detailed answer. And if you find something that's not super detailed, or like the answer to your question is good, but you know better, the reason for that is that we haven't found a good way of incentivizing people like you to provide your knowledge, and that's literally what Sapien is for.

Josh Kriger: Yeah, and it could be a good idea or inspiration for what to do on Sapien, right? I'm a fan of Perplexity AI Pro. I use that one a lot, and they just incorporated DeepSeq right away, DeepSeq. And just in the last two months, I've seen the sophistication of the information I'm getting back, you know, improve dramatically. I'm sure there's a lot of great tools out there. This is just the one that sort of I've latched onto.

Rowan Stone: Very cool.

Josh Kriger: So one last thing, Ronan, before, sorry. So one last thing, Ronan, before we depart today is we have this segment called a shout out where you get to sort of acknowledge someone in your orbit that maybe doesn't get as much spotlight attention as you think they ought. You are sort of in the midst of doing big things. So I'm curious who comes to mind.

Rowan Stone: I would have to shout out, and I'm not going to use his first name because I don't know how often he uses his first name, so I'm going to shout out his Twitter handle instead. I'm going to shout out Kalos, who is currently leading a whole bunch of ridiculously cool stuff. He is first and foremost leading Parallel, the trading card game, which is using a bunch of on-chain stuff to basically just supercharge a really cool kind of magic style TCG. He's also building a proper platformer, fully three-dimensional game, which is absurd for a relatively small studio given the kind of tasks they're taking on. But more topically, While doing that platformer, they realized that NPCs in games should just be AI. We shouldn't be doing this weird pre-scripted thing. We should have people that can think for themselves. So he layered in NPCs with real AI and ultimately ended up giving them spending power. Long story short, he's now building an AI agent specifically for the crypto space. I'm butchering the work that he's doing, but if anybody hasn't heard of Kalos, check him out on Twitter. Check what he's building. Jump in and see Wayfinder. Jump in and see Parallel. Have a look at the Prime ecosystem. The quality is absurd. The artwork is absurd. It's super, super cool.

Josh Kriger: So Kalos sounds like a really interesting person. What's his ex-handle? Temple Crash, C-R-A-S-H. All right, we'll find them somewhere in the net. That's it for today, Rowan. I hope this hasn't been too painful, and that you enjoyed this conversation as much as I did. I was really fascinated by what you're doing. I think you have one of the coolest jobs on Earth, and I'm really excited about what's to come.

Rowan Stone: Thank you for having me. Thanks for the time. Not painful at all, but a ton of fun. Look forward to doing it again soon.

Intro/Outro: We've reached the outer limit at the edge of show for today. Thanks for exploring with us. We have room for more adventurers on our starship, so invite your friends and cool strangers to join our journey. If you're among the hundreds of thousands following us on Spotify, iTunes, iHeartRadio, or watching us on Maiko or YouTube, please hit that subscribe button, rate us, and make an awesome comment too. And don't forget to pass this episode along to a friend or two who would benefit from it. Doubling back to again, recommend checking out myco.io, where you can watch The Edge of Show and earn for your time and attention. That's m-y-c-o dot i-o. Don't forget to visit theedgeofshow.com, the as part of their domain name, where you can learn more about collaborating with us, and also subscribe to The Edge of Weekly Newsletter for the latest Web3 news, events, and show drops. In addition, connect with us on all major social platforms by searching for The Edge of Show. Join the exciting conversations happening online. Lastly, be sure to tune in next time for more great Web3 and AI content. Until then, keep pushing the boundaries. Thanks again for hanging out with us today. The views and opinions expressed on The Edge of Show reflect solely those views and opinions of the show hosts and its guests. Please make sure to do your own research. Our show is not financial advice. You understand that you are using any and all information on or through this podcast at your own risk. Whenever making financial decisions, we recommend doing your own research and talking to your accountant for financial advice. From time to time, we may feature sponsored content on the show for which we receive value, and we may share links for which we receive a commission if you make a purchase through one of these links. Refer to our website, www.edgeofnft.com, for our full disclaimer, terms and conditions, and privacy policy.

‍

‍