PrismaX CEO Bayley Wang on How Robotics, AI, and Blockchain Will Power the Next Wave of Innovation

Hello, and welcome to the latest edition of the Techstrong AI Leadership Insight series. I'm your host, Mike Ard. Today we're with Bailey Wang, who's the CEO for Prisma X, and we're gonna have a little chat about the intersection of robotics and AI and blockchain and how all this stuff is coming together through a marketplace that these guys have built.

Bailey, welcome to Shah. Yeah, great to be here. Everybody's talking about AI and a lot of attention is of course, being paid to robotics, but you don't often hear blockchain in the same sentence.

So how do you see all these things coming together and how are they kind of maybe mutually dependent? Right, right. So I, I think it behooves us to give a little context first on what we mean by AI and robotics.

Basically, in the past decade, there has been a ton of advancement on the hardware and low level controls algorithms in robotics, right? You've probably seen all these videos from Boston Dynamics and Unit Tree and all these other companies of robots doing backflips and all these crazy motions. But what you don't see is robots going around and planning the, the sequence of steps to complete long and complex text tasks by themselves.

That level, that part of the control algorithms, which is sort of like a user facing thing, is still done by programming fixed function paths into the robots. And that makes robots really hard to deploy in novel applications. So the ultimate vision of the robotics plus AI movement, at least in the, among the sort of sort of US based AI community, is to build robots with increasingly greater degrees of automation, ultimately resulting in a robot where you can prompt the robot with natural language like an LLM and have it go complete these really complex tasks for hours at a time.

And in order to do that, uh, we need to, we need to obviously scale the distribution of robots, scale training, data scale, the diversity of trading data, as well as the number of hours. And that's where blockchain comes in. Blockchain is a really good way to coordinate enormous amounts of, uh, enormous amounts of human work very quickly.

And that's exactly what robotics needs right now. The, uh, AI, native robotics industry is facing sort of a crisis when it comes to scaling the human neighbor behind the behind robots. Uh, we need millions, tens of millions of hours of trading data, but we maybe have tens of thousands of hours of training data right now.

We need robots everywhere in the world in order to make these models robust. We currently have robots in four Airbnbs in Silicon Valley, so right. There's this huge gap in being able to scale the decentralized way with blockchain is what's ultimately going to give us the, give us the diversity needed to achieve general models.

You can think of it as they, they, they call blockchain, right? Web, Web3. And one way to think about it is that web, web one, web one was completely decentralized.

It was a bunch of people posting on forums and making their own blogs and talking about what they're passionate about. And from that era of the internet, we got the incredibly diverse coverage necessary to train these LLMs. And then like web two, it became more centralized.

There were content creation platforms. People have figured out how to monetize it. So Web3 in a sense, is sort of bringing this decentralized diversity back, back into the, back into these projects in order to really scale them.

So I get that blockchain provides a distributed immutable store of information, but how does that get plugged into a workflow for the robots? It's, it's more about the, it's more about coordinating the people behind the robots and less about the robots themselves For us, basically, uh, right. The problem with Web one was that people were went and did things they were passionate about, but there was, there was no way to incentivize these people.

Uh, like people would blog and write and post for free and Web two, what Web two added payment mechanisms, but in a centralized way. And in doing so, you, we, it sort of collapsed the data distribution. If you go look at any one of these big content platforms, you can clearly see that the content is sort of somewhat repetitive.

And, uh, the idea behind using blockchain is that it creates incentive mechanisms which drive the correct, uh, activities out of the people. So take, take, uh, teleoperated data collection, which is something that we do, right, where people will come in and remotely operate a robot in order to collect data. If you hire people, uh, in a centralized way and then pay them, pay them a fixed amount to operate the robots, there really is no incentive for these people to practice or become better and better than operate at operating the robots other than if you went and like very rapidly fired people, which is just not great and regulatorily questionable.

You can't like fire people on an hourly basis until you find the good ones. But by setting everything up in a decentralized way with leaderboards, with scoring, with incentives, you let the, you let the sort of swarm of people self-discover what's the data that's valuable, the people who are skilled at operating the robots. And, and that helps create much better data sets and much faster scaling than if you did everything but just with centralized payments.

Mm-hmm. And is that for training purposes or will we also have that model for use cases in the field and production environments? I mean, how far do we go?

I, I think, I think we can go very, very far. Right now we're focused on training data because that's what the industry needs right now. But I think once robots scale into the field and we have more robots in the field, uh, teleoperation and discovering the best tele operators will also be critical for maintaining the robots in the field and ensuring their reliability, right?

It's sort of like autonomous driving where, uh, the cars are autonomous 90% of the time, 95% of the time, but once in a while someone has to step in and drive the Waymo out of a sticky situation. The other, the sort of other flip side of the industry is coordinating, coordinating transactions between robots once robots are in the field. And I think that's a really interesting use of blockchain as well, because mainstream payment processing networks, visa, MasterCard, discover a CH Swift, are unlikely to be willing to process automated machine to machine transactions.

And having a decentralized transaction medium would allow robots to essentially hire each other to do work, which is pretty cool, Right? Um, will we have, uh, you know, a a bunch of robots that are narrowly trained to perform a specific set of tasks, or as this evolve, will the robots be trained to perform a wide range of tasks that may be related to each other, but they're gonna be, um, maybe semi or fully autonomous? I think the industry is going to follow the same path we saw with Computer Vision and NLP, where if, if you remember back in like 20 16, 20 17 when we first got nns and CNNs and Bert NLP and Computer Vision snuck in, not, not as like a big fancy chat bot or an autonomous, or like an autonomous beast, but as like better Google translate and reverse image search and feature extraction, all of these things, I think we're gonna see the same thing for robots where the first autonomous robots are not years away.

They're like months away, and they're gonna sneak into some, some sector of industry or some part of our lives, or like restaurants or hotels or something and just sit behind the scenes and operate. And there'll be AI in these machines. And the act actual fully autonomous robots are sort of the vision of the whole industry here, right?

The idea is robotics is already a, it's, it's an excellent industry. It's worth tens of billions of dollars a year, but it's unable to scale because they've sort of capped out the singular use cases that are worth a lot of money by building robots, by building robots, which are, uh, more general, you can build a sort of one, one tool fits all and then your, your one tool can capture multiple valuable use cases, none of which individually are worth servicing, but the sum of which are worth servicing, right? I think, uh, to, to, for, if not, not to be like, uh, uh, trite or anything, but domestic is a good example where you probably wouldn't pay thousands of dollars for a robot which went and like put the dishes in the dishwasher or folded the laundry or, or any one of these things.

But if you could buy a machine which did the sum of all of these tasks in your home and could do it repeatedly, it could do it accurately, that that would be worth money. So I think ultimately that's where the industry wants to go, but it's going to take a little while for the industry to scale to that level of generalization. Yeah.

So it almost seems like there will be robots that are specialists and train at a particular task, and then there'll be others that are more general purpose platforms, for lack of a better phrase, that will perform a range of tasks that might be related to each other, or at least in the same general discipline. Yeah. Yeah.

I, I think, I think eventually the specialist robots will slowly phase out and be replaced with the generalist ones as they get better one once again. Right? The NLP analogy is a good one here.

We had a lot of con ai, like RNN small transformer based contraptions back in the late, late 2010s, like 20, 20 80, 20 19, 20 20, and they've all been single handedly replaced by foundation models. Yeah. Um, as you kind of put all this together for a minute, um, am I getting the AI agents or LLMs that are put in the robotics systems?

Are we going to, um, use natural language to, I don't know, typing commands? Or is that gonna be more of a voice interaction eventually? I think that that's just a, that's a really subtle level of the user interface, which I think is going to depend on what the, what the use cases are.

Voice, voice input is harder than you think it is in noisy environments. Like I'm sure you, you've used voice transcription tools in like a quiet setting, and it's been, they've been reliable, but as soon as there's background noise and interruptions it, it becomes kind of flaky. So I, I think, uh, I, I don't, I don't really have a strong answer on this one.

I think that now I think that typing and text are a very reliable way to interface with machines and with a machine that interacts with the real world, that kind of reliability might be something that you find desirable, but vo voice is useful in some applications, especially when you, when when you're, when you're, when you need to be hands free. Mm-hmm. Where will the robots come from?

And I'm asking this question because as I look around, there are all these kits, and I'm kind of reminded of the early days of PCs when you went to some little club or fair and everybody built their own long before, you know, the big manufacturers came along, but so will people build their own little robots, or is there gonna be kind of, you know, the Dells and the Apples of the world, or Samsung, or wherever it may be, are gonna just start supplying robots like they do laptops and PCs? Robotics is an interesting industry because unlike basically any other technology and the history of technology, we're in a state where the hardware has outstripped the software. It's like a weird overfitting from the, the like hardware people where, uh, I, I guess the sort of reasoning behind it is that before we had modern advanced robots with legs and arms and fingers, we had like Roombas and little hot pot robots and people who built serbo motors for robots.

And if you look at the layout of, say, a modern humanoid robot, it's, it's precious few parts to go from a box of motors to like the whole robot. So what happened was all of these people who went and built the motors, what went and just straight up, they took their motors and put them in robot, in human shape frames and built humanoid robots out of them. Then went and like hired a grad student who knew the algorithms to make the thing walk.

So we're in a state where I think as of last count, there are over 50 humanoid robot manufacturers. Uh, so I think once the software catches up, there will be plenty of hardware supply, right? There's, it's, it's, it's not just the Asian manufacturers.

There's like a bunch of guys in the US trying to do it. There's a few guys in Europe trying to do it. There's, there's a literal explosion of robotics, robotics hardware companies right now.

Uh, not to mention like tabletop robot arms and stuff. There's, there's just more of those than you can count. If you go on Amazon Prime and search robot arms, there's probably a hundred different robot arms you can buy right now.

Are these things gonna be ready for public use? I can understand how I might use that within a controlled environment, whether it's a factory or even my home. But, um, we've seen some experiments.

Um, I live in New York City, they had a robot running around the subway, and then people eventually just picked it up and moved it into a trash can and it couldn't get out. And, you know, people are people, they're gonna do crazy things with these things. So, you know, are the robots kinda ready for, um, the general public?

Uh, I think you're right, controlled environments first for sure. There's a lot of teething pains and a lot of value to be had. It's not going to be just like big stuffy industrial environments.

I think the vision there is to bring robotics to use cases like local restaurants and small businesses, convenience stores, swamped small hotels where you can, you can generate a lot of value, right? The, the, the sum of all of these things is a trillion dollar industry, but no, no, one aspect of them is a trillion dollar industry. So he hence the need for generalization.

Uh, I, I think home robotics is where everyone wants to go because the, the dream is a robot in every home, and if you multiply the number of homes by the number of dollars a month, you can charge per robot. It's a lot of money and it's really useful. But I think you're right, that's gonna have to, we're gonna have to wait a while.

My, my guess is that domestic robots will, uh, materialize sometime in the next decade, but beyond that, it's really hard to say. And maybe, maybe I'm wrong, and maybe there's like a non-mobile form factor which will become commercially successful in like four or five years, or just something which bolts to your kitchen counter and helps you, helps you with the cooking and cost a few thousand dollars. And I, I, I think that's feasible.

But there's a lot of product, product fine tuning and PMF discovery to be done before you could build a product like that. Yeah. I don't know what I would pay for it, but it would be handy to have a robot that would just take the dishes outta the sink and load 'em and the dishwasher, right?

That would be, yeah. Yeah. Stuff like, stuff like that.

And just generally around, around the kitchen is like a big use case that people are going after. Everyone, everyone cooks and kitchens are, are pretty fixed spaces, pretty small spaces. So what ultimately is your best advice then to people who are looking at this whole area?

'cause I think as we've seen with ai, we tend to get a little irrational exuberance going, uh, yeah, what, what's real here? What's the curve of maturity? And, you know, map it out for us a little bit.

Yeah, yeah. I guess if you are a builder in the space, my suggestion is to actually build a product which people want. Uh, right.

As, as a, as a, as a startup founder, you have to build a product which can ship in a couple years. You don't get the luxury of going and waiting seven years to go finish your product. So try to build something which is feasible in a few years, and it's all product people actually want.

And I think the technology has matured to, to the point where there's lots of exciting use cases that can bridge you between, like right now and fully general robots. If you are looking at the, if you're, if you're sort of coming in from the investor side and trying to find signal in all of the noise, I would say learn the technology and learn what the actual challenges are, rather than just being excited by the flashy demos. Robotics is one of those spaces where the gap between what's, what you can do in a demo and what you can do in a product is, is immense.

All of the recent demo videos that have been rolling out, and I won't, I won't, I won't name, name names, uh, are all a little bit fishy if you stare at them. They're clearly going after use demo use cases, which are actually very easy to do. But as a sort of person not experienced in the technology, the, the demo looks very impressive.

Uh, and I, I, I, I think, I think that's, I think that's very important as well, uh, as a builder too, to just sort of sort of be, be more transparent with what the demos are and what they're doing. It's important for the industry to move forwards if there's no smoke and mirrors. True.

And finally, you hear a lot about, people are of course, of concerned about, you know, what the impact all this might have on jobs, but, um, taking the emotion out of it, just looking at it, it seems to me we just don't have enough people to do all these things. So maybe, you know, the only way forward is some sort of robot. Yeah.

Yeah. So I, I think one, one thing way to think to think about it is to remember that the robot is fundamentally a tool like any other tool, right? It's a, like, it's a six degree of freedom positioner on wheels that you use to move stuff around, and that, that's all that is.

And that's a helpful tool. It's no different from like a drill or a drill or a computer. It just happens to look a little bit like a human.

So I think it, it triggers a bit of, an bit of concern, anxiety among the general populace. And I, I think you're right. We don't have enough people to do these jobs.

And more importantly, I don't think anyone enjoys doing repetitive manual labor, like humans are built to sort of think and do very fine manipulation. We're not really built to like, move heavy objects around the sun for eight hours every day. So I think once we have more automation in this space, there will be sort of like a, a dip in jobs and then a curve, and people will be happier once, uh, once people find new things to do with all the time that's been freed up by the machines, right?

If you like, go back to every industrial revolution, we've had the same, same cycle every time we got the loom and people are wondering, oh, all of the people who are making clothes, where their jobs gonna be, then we got like mechanization and people are like, oh, where are all the people digging Mines gonna be? Then we got computers and people are like, where are all the people using typewriters gonna be? But in the end, the conclusion society has come to is we do not enjoy weaving clothes for eight hours a day.

We don't enjoy sitting at a typewriter for 10 hours a day, and we definitely don't enjoy Mighty coal no matter what for any number of hours a day. So being able to, being able to have machines help us on all of these things has ultimately, like, made our lives better. Change, change is, change is a tricky thing.

Like, and, and machines bring change, right? We're, we're in the midst of one now with LLMs where people are finding new things to do with their time, and it's not, it's not, and not, it's not just like customer support, like junior, junior, front-end developers are wondering what their, what their next jobs will be. But I think ultimately, ultimately people will be happier because of it.

They'll find new things to do that they love. Well, hopefully those hands won't be idle because you know what they say about idle hands, but the most important thing to remember about AI is who's the boss. Hey, Bailey, thanks for being on the show.

Yeah, yeah. Thanks. All right.

And thank you all for watching the latest episode, tech Strong AI Leadership Inside series. You can find this episode and others on our website. We invite you to check them all out.

Until then, we'll see you next time.