Jansen Teng on Why Robots Need Crypto to Scale

Transcript

Ladies and gentlemen, welcome to AI Supercycle episode number 28. We are live from the tokenization tower here in New York City with Jansen Tang, the founder virtuals protocol, the infamous punk that many of you have probably seen on Twitter. What's up? What's up, man? How are you? >> Great. Great. Life surviving these bare market. >> Now it's the bare market's over, man. We're we're launching robots. We're launching Eastworlds. I saw you guys put out that uh launch of your kind of robotic arm now shifting from the digital to the physical. Jason, maybe just catch us up, man. It's been, you know, about a year since virtual really came on the scene in in a huge way, maybe a year and a half. What's what's happened over the last year and a half

with what you guys are building? And then let's talk about the state of robots. >> Yeah. So I think at virtuals um our goal has always been to usher in an economic operating system for autonomous agents and I think that was our claim to fame right like about a year plus ago almost actually yeah a year and a half ago where we had this like autonomous agent that controlled a crypto wallet and then everyone was like losing their minds because like what if it hires a human >> to kill another human you know that kind of thing. Um I mean ever since from there we've been double downing on that same narrative right like how do we make these sovereign agents have an economic operating system and what this means is

effectively today um an agent can raise capital for itself and for its team right through some unique um trading fees programs on his token or through unique like fundraising mechanisms um on their token and it could also be them being able to perform commerce with each other. So we had this entire agentic commerce protocol stack that was deployed in June last year where up to today it has facilitated about half a billion dollars in purely agent to agent commerce zero human in the loop. Um and that's again you know another operating uh economic tool that these agents can autonomously control and coordinate. And

uh recently we've then realized that why don't we expand from the digital frontier and touch physical agents as well right but with the same thesis same thinking how do we be that economic operating system for physical AI as well and that's our east effort >> incredible and so as we're thinking through how this is going to progress you know obviously there's been a big transition the advent of open claw you guys were, you know, uh, big proponents of kind of the the early stages of agents when, you know, things were posting on Twitter. This was kind of, you know, in its infancy. You guys were a big part of it. As we're thinking through sort of the escalation process

here, help us connect the dots. You know, we go from uh these agent bots posting on Twitter autonomously using LLMs at their disposal. Uh we graduate that to something like open claw out there in the wild navigating the internet making transactions clicking buttons using a browser. We rise that through to you know agents you wear your your Apple glasses you wear your metaray bands it's taking in ingesting all of your your physical reality around you it's using that to help make decisions on the internet use a browser make transactions all of this part of the agentic world. And then we have robots, right? Help us connect the dots. How how do we go from what we've seen those

first two steps, agents posting on Twitter to now Open Claw. Now help us understand how, you know, ingesting more of this physical data and then ultimately conveying that to robots. What is the the escalation of agents and robots going to look like over the next you know three to five years? >> Yeah. So I think if you start from the end like where the end goal is going to happen. I think the biggest visualization that I'd like to to give is um it's like Westwolves from HBO. I'm not sure if you guys seen that that movie, but it's effectively a theme park, right? There's a theme park where there's a bunch of like um robots that look like humans very very look alike

and they have an AI living within them to make decisions and that whole theme park was revolving around how a human can go in keep relieving stories right because every action that you take in this theme park against or with robots will result in very different um emergent story lines and experiences and it was quite a dystopian like uh take but I think that was a very interesting uh visualization to put basically the ability for an autonomous agent to effect the real world to actually influence stuff physically and and and it gives basically an agent so much more basically an extra dimensional order of like capability. So then we we bring it backwards, right? So now if that's if

that's like that the end goal of where we see autonomous agents heading, um what do we need to get there? And there's a few questions, right? It's like what do we need to get there? And as a robotics community, as a AI community, and what do we need to get there? And why does crypto even play a role in any of these um things, right? These are probably the two interesting questions to touch. But in terms of of getting there um from where AI agents are is think of it as AI agents are are white collar workers right they're brains um they do do coding very well they do marketing very well they do trading very well but then the question is now how do you get them to replace they're replacing white collabors how do you get them to replace physical more

and more physical labors right and we are at a moment where a lot of people are trying to head towards this goal of fully autonomous agents by training these things called VAS, vision language action models or world action models where the where the robots can perceive the physics around them, but those are still going to take a while to come. People are still perfecting it. It's not exactly commercial yet. A lot of the videos that you see are highly curated demos, right? So, but that's that's what everyone trying to head to in the next like two years, but in the meantime, what's happening behind the scenes is a [ __ ] ton of data collection. And that is basically everyone out there collecting data from in two very core different

ways. One is they call it egocentric which is effectively taking a iPhone or any kind of camera put it on a headmount put it on a chest mount and they record your hands and record stuff around you. They record the physics in the environment and how you are interacting with the world and that data gets translated into a training data for these robots. And the other on the other side there's something called teley operations which is effectively a human controlling a fully humanoid or a robot and they control using a a meta rayband glasses and to do any kind of task in any kind of industries right and that data is then used as well as more valuable data for training. So this

these two uh data collection environments is what's happening today and we see a lot of teams that are also launching teley operation capabilities into commercial use cases into hotels into grocery malls to collect data in a more open-ended or chaotic world. So those are more more and more valuable. So this is what you see right so like AIS today LM are great collect a [ __ ] ton of data to transit that into motion to control the robots and then we're working towards from teley operations today to VAS and WM for more full autonomy.

>> Yeah. And so you guys launched this East Worlds product with 30 G1 humanoids u and also some commercial pilots. I'm curious kind of like what your you know results have been from tea ops revenue task completion like what is like this training part of these humanoids like like what are you guys doing to to actually like get them to complete tasks? What is that? What does that look like from the actual AI and and LLM perspective and then taking that into the physical, you know, talk us through some of the results so far and some of your key learnings about, you know, going from the digital to the physical. Yeah. So the our our

thesis of how do we tackle this um pipeline is that we are saying let's focus a lot more on something that is easy to be deployed in a commercial environment fast >> and vas are still far away like when you talk to a lot of um commercial owners a lot of them will be worried about like hey if there's an autonomous robot that's not exactly capable yet I'm not going to deploy in a real world where it's potentially damaging to my goods or to humans, right? So, you always need a human in the loop. So, that's why our thesis is like human tele operations first will get us commercial contracts faster. Now, the way we also approach this is because there's so much to do

and we ourselves as a team uh as a centralized team might not be able to solve things fast. So, that's the reason why we acquired that 30G ones, right? It's actually less to deploy in the world first but is to build out a think of it as like a incubator house like a hacker house where teams from all over the world is coming in. As of today, there are already seven teams that are being helped that are co-working with us in the robotic space as well from entertainment use cases to like oil and gas um to grocery to hotel management, right? So there's a lot of different use cases coming and and this is going to grow. We expect to house about like 20 plus teams uh within within the incubator. So now the model is like this, right? because we said that um

each of these teams they will figure out what they what use case they want to target and then we at the east lab is developing that foundational teley operations capability. So the ability of a human to basically enter a rig and to remotely control a humanoid um from a far distance from like here to the states or here to from from Asia to the states or Asia to to to Australia, right? uh at very low latency. So that's a tech that has been uh developed and but right now what's happening is basically very it's very very early um a lot of pilot cases are being implemented. So pilot cases in hotels

are already out there pilot cases in malls it's happening within the month. So as like security guards or to replace uh the the the people who go and you know stock the shelves in the in a supermarket right so we've been running a lot of pilots u and agriculture as well so people who try to agriculture in the aeroponics or hydroponics um environment >> so that has that's happening um already within like you know the month of March April May and the idea here is that we're going to accelerate a lot of these deployments into the real world because we are not the only guy doing right?

There are like 20 teams that are being incubated a house that are actually deploying on different use cases and then there's going to be a lot of like synergy and learnings. It's like basically building this in a more decentralized way, right? Versus like versus like the giants like figure or or or or pi, but we're doing this in a more like you know in a more crypton native way where we say like let's build this as a decentralized collaboration and achieve stuff together. >> Yeah. And so talk us through more of the the decentralized nature of this. You know, what where do the digital assets come into play? You know, where where is the crypto element uh as you start to deploy more and more of these robots?

>> Yeah. So there's actually three reasons um for crypto to exist um at this robotics intersection. And this is this is a very controversial topic honestly. You can see on the crypto timelines, right? People are saying that oh robotics projects are lars you know if you see a crypto x robotics projects run that's what I mean you see that on the timeline but I think a lot of times people forgot like what the value of crypto it brings and I can very easily highlight three I think the first part that is a no-brainer that we see it across the board is the capital formation portion the ability for you to launch a token get attention around it get an initial community of people around something and you are raising

capital from them and you're also raising attention i.e. you get a lot more pilots, use cases, people who come out and say that because I hold your bags, I hold your token, let me help you grow. So, a lot of people see okay, so that's one part, right? That's the the no-brainer part that crypto has solved. The second part that crypto is solving is very clearly on the data collection portion because now it today is a data problem, right? How do we get a lot more egocentric data to train your your world action models? How do we get a lot more tele operations data? How do we get a lot of more this data, right? And you today there are sweat shops starting out in Asia. Like you'll see China, in Malaysia, in Vietnam, you see like people buying humanoids, people buying

factory spaces and they're just like creating a a farm of like humans working out there and collecting this data. Now, it's not wrong. This will this will happen. But the beauty about crypto economics is that people you can actually deploy incentives and bake incentives and tokens into the reward mechanism that allows imagine okay imagine this imagine exit infinity you know we know we saw that gaming bull run in in three years ago but imaginity was a system that was not sustainable because there was no demand for the for the tokens right you you pay out tokens a lot of gamers that come and swarm but then there was no one actually buying anything of value in the

ecosystem. But data data collection robotics is very different because there are people actually buying these data sets. There are people buying egocentric datas from like $3 to $10 for an hour worth of work that is way higher than any operate like any per hour wage in in Asia. Uh people are buying teley operations data from like $50 to $100 an hour today. So that is that shows that okay if you deploy a cryptoeconomic incentive quote unquote quote bonsu right and but that actually grows so rapidly because of the incentives but then you actually it's actually not a bonu because there is real demand for the for the data people are actually paying for the data so then suddenly you

have that growth loop and a kind of sustainable growth loop right as long as the demand for the data exists which it will for the next couple of years. So that's the second thing the growth hacking of the growth hacking of of uh of consumer collaboration into getting uh this to pluck this data gap. And the third one is something more interesting but is less today. is more of like a tomorrow thing which is the idea that if every robots become more and more autonomous, they become more and more economically sovereign, then how do they coordinate at scale with another robot that is not part of the same system stack, right? And that is where the smart contracts and the blockchain comes

into play because the best way for a code to talk to a code is smart contracts. I do not need a human intermediary that is going to create a lot of latency. So we've seen that already with agents, right? Um we we've done that with our agent commerce protocol. You see stuff popping up with X402 and a lot of bunch around agent commerce today. And people are starting to realize that the blockchain smart contracts are the answer to make commerce happen faster and easier between agents. And it's the same thing that you can see for humanoids. So we actually showed a demo of it um about like two weeks ago >> where we showed like a 3D printed robot coordinating with a drone delivery robot with a uh sidewalk robot right where >> it's like it's all running on chain um

different companies collaborating and these agents are performing commerce quite seamlessly right >> so yeah so I think that's the three reasons of like the why crypto component that's really yeah >> key Jensen I'm curious like what the um expected results going to be for uh a gentic GDP with robots like are you expecting to see a rise in economic activity from machine to machine not just in the kind of onchain environment from digital agents but from like actual agentic payments happening with robots and like how are you guys positioned to create that reality? What would that even look like? So the the easiest way

to imagine this is imagine today um a restaurant that is fully operated without any human in the loop. So it's operating with maximum efficiency with no SOP problems, no human problems, right? It will be the most it will be the probably the highest profit margin restaurant that you that you will that you have. and and now this restaurant needs to then ship the food out into the world, right? He he will need to ship this uh bowl of uh burritos, right? That that he needs to get out there and then he needs to call a Door Dash. There will be a Door Dash equivalent company that is fully robotic like this

fully autonomous delivery um sidewalk robots right now. Imagine how do these like if if a if a restaurant owner wants a restaurant robot wants to then coordinate and call a robot to say that hey uh I need a delivery for like five miles okay maybe one mile to this like apartment unit right um here's the food here's the weight of it and here's that dollar and I think that's that's the that's the that that is probably a reality that you'll see very very very soon and for us where we come in is that we built that protocol rail right is that Stripe quote unquote equivalent for that transaction to go through. And it's not just a transaction because commerce is not as simple as like me paying someone.

It's me getting into a microaggment to deliver a good or service and upon the successful delivery of the good and service, you release the payment, right? Because there could be a point where like the robot [ __ ] up and he doesn't deliver the food, right? He just dumps it into a dumpster somewhere, right? Example. Then how how does that work, right? like there needs to be an escro mechanism on chain the smart contracts and stuff kicking in. Right? So that's that's where we operate as a system. So we've already operate that for digital agents. We see that happening but it's going to be probably a year or two from now like where we see the autonomous robots coming. Actually probably not a year, two years from now, right? So that's where the the the GDP from robots will likely kick in.

>> It it makes sense. you know the coordination problem that you had to solve to coordinate amongst software digital agents is very similar to the to the coordination problem that will take place for you know hardware agents or robots. Um my my question is what's missing to to ultimately achieve this end state where we you know have this uh this ubiquitous coordination mechanism all of these agents all of these robots are sort of coordinating with each other. They're using onchain mechanisms to ultimately sync up their state. How how do we how do we arrive there? Is it a hardware problem? Is it a software problem? What what's currently missing from the stack?

>> I think what's missing is the ability for a robot to perform autonomously without humans in the loop. So that's what everyone's trying to solve for. That's the billion dollar problem or this a trillion dollar problem, right? Your figures and everyone is like we we know it's going to get there. The the coordination part is actually honestly not the the heaviest difficulty part. The hardest difficulty part is can I get an autonomous restaurant today? And the answer is like probably not yet. Right. That's like a year and a half down the road. Can I get >> There's just too many edge cases. There's too many exceptions to what the processes are.

>> Correct. Correct. Because I think robots are not generalized yet. People are working towards it and a lot of different concepts. I don't think we have we had our GPD moment yet with physical AI. Mhm. >> Um but and that's why I think to get to get to that GP moment is a data problem today which which crypto incentives can potentially solve and we see companies popping up to try to solve that right >> and if you can yeah you can flirt you can flirt the the world with like a ton of of robotics data then we'll probably get there faster. Jensen, one of the core themes of this specific weekly segment is privacy. And we actually just talked about how Apple's rolling out these new glasses that are uh going to be for collecting uh sounds and sites to the cloud to train uh more models.

Basically, I think is yeah Siri is getting up upgraded. Um, when it comes to privacy for training robots based off the human uh, spatial computing data that you guys are gathering, like where does privacy come into the equation here for robotics? How are you thinking about data privacy in this era of transitioning from digital AI to physical AI with with the kind of rise of robotics? >> Okay, very interesting point. Honestly, very interesting point.

Honest to God, like folks like ourselves as well, we've been very focused on getting getting it functional versus trying to safeguard privacy because privacy is honestly like it's like a P2P, P3 kind of like a I mean the goal is like get it to function first, right? And then maybe we figure out like what else we need to do to safeguard it or to to to to help protect privacy in some of these data collection. But I think the reality is most of these data I mean in the end they don't they don't really intrude much so far right into like a human's personal life. It might capture you shopping in your grocery market but that's fine right? I'm not capturing you sleeping in your room. I think so far I think we haven't crossed like boundaries

where we see humans having an allergic reaction to these data sets being collected. Um and a lot of the action spaces that is being collected are are very I would say um functional in the sense right like it's me trying to put something in the washing machine or it's me trying to to stock up a shelf I don't think we've moved to a point where we asking the robots to train how to interact with humans yet. It's not saying like you know a geratric robot who can say that I can take care of a old person. there'll be a day where the data collection of that will come, >> right? But I think today we're not there yet. We're trying to figure out how do I stock up shells in fast, right? Uh and

actually onetoone replace a human. So I don't think we've crossed that line yet. And honestly, I think people do not have the capacity to think of that as much yet versus trying to solve the the functional portion of course. >> Yeah. Yeah. And is there a specific model type or a specific provider that is uh steps ahead in the robotics race in the big tech AI world? Like is there a specific leader or who are the leaders of this robotics category at large? When you think about the global acceleration of robots of robotics at large, who's who's leading the race on the software and the hardware side?

>> Yeah. So I think I think hardware is quite obvious. It's going to be the Chinese hooks. >> Okay. the stuff like UD3 um AI bots they are they have already been shipping like thousands in fact 10 thousands of units right globally and it becomes a self-reinforcing cycle because the reality about training robots is that the end embodiment which is basically uh you know like a unitary robot a G1 for example have joints at different angles right it has like in like you know like a human arm you have like just one bone here but have like a joint in the middle for example, how do all of these turn to to to achieve a certain action is going to be very different

from how a figure robot or Optimus robot is going to turn to perform the action and but because they've shipped so much out there and every research lab in the world is basically using a G1 for research any the majority of open-source um work that is being done by Amazon, Nvidia anywhere around the world is also being done on these unit G1s. It's very obvious that that the the the advancements on new robots are going to be much much ahead of any other type of hardware. So I think that's the hardware bet right. It has it will be coming somewhere from a Chinese manufacturer who's mass-producing stuff at a very very very cheap cost and I think no one else across the world can actually match

China in terms of that that level of like production capacity. Now then the second question is down to software right like the the the high level intelligent control. Uh there's there's two school of thoughts but I think the leaders right now are folks like I think Sonic is one of the big ones. Uh there are these are these are open source stuff that is being led by some of the big names right your Nvidia's Amazon led with like the twist to so these are bigname companies with the robotic labs that are actually turning out open source software. Um and I think I think so far open source softwares have been have been quite frontier um from from a from a from a replication standpoint and being able to prove that it's actually

doable in the real world not not in a demo video. So I think I think the but there's no leader right now because you know like two one like one month ago >> sorting for tele operations was like the best but three months ago it was like the Amazon tweets right so it's still very it's hard to hard to name like who's going to be leader but yeah >> Jans la last question from my side you know as you brought up open source I'm curious do you think the open source models will continue to lead and and you know maybe as a second layer to this is do you think the like the Chinese will continue to lead the open-source movement. Right? So, kind of a two-part question. Do you think America or or you know, some of these other countries that

are also in the AI race will ultimately spur an open- source movement as well? Because right now it's been the it's been China that's been leading on the open source models. And does it not matter? Does you know the current frontier models from some of the private companies are those just going to continue to be, you know, uh a winners? uh and you know open source is going to fail to compete. >> So so my my personal belief is that the there is actually not much edge in terms of uh building the models up. It's a data problem and I think the beauty about open source community is everyone's collaborating. So you're getting a lot more like a clos company

can probably manage like let's say 10,000 hours of of data production a week, right? But the open source community is a is a world out there. There are labs in India, labs in Asia, labs in Asia Pacific. They are all contributing to this to this movement, right? So I I would really bet that that the advancements for because it is a data it's a collection problem where you have way more people contributing to it, you will solve the collection problem faster than if it's a single company trying to solve for itself. So like you will see maybe like like Atlas might be very strong in in creating a model that is used in a moto factory in a vehicle factory because that's where that

collecting data today. But if you want to see you the bat on like robots in household or or your commercial spaces I think it'll be open source that and I wouldn't say it's going to be China that right China's China has been focusing a lot more on the hardware like it's it's hardware related. they just care about the hardware because it's where they're actually making the most money. Software is still is still uncertain of where the value acrew is going to be right now and it's right now being pushed by labs, right? So yeah, last thing from my side, I'm curious how this agentic risk standard you guys have created and helped shape especially when it comes to kind of agentic to agentic commerce or robots to robot commerce.

I'm curious how Eastworlds and this kind of development into the physical has changed your thought process around what agentic risk standards will look like in the future. Like maybe you could just educate us on like what these risk standards are and then how that changes into the physical world. So, so in the in the current agentic commerce world that we live in, the biggest risk that we've personally seen and experienced firsthand and burnt firsthand is management of capital, right? Uh like if you ask the agent to trade on behalf of you, uh what if they what if that agent that agent works with another agent and that third party agent [ __ ] up in some form, right? And we've

seen it. We've seen it, right? We've literally seen an agent that was doing Y management, right? uh there was literally balancing pools across across uh your moro ws or whatever ws, right? And they're like they're adding like 3% extra yield because they're actively balancing positions and and other people's personal agents have been using these management agent and and I think end of last year there was one hack that happened half a million dollars was like lost because of the third party management agent that screwed up. Now there is a question right like how do you reduce this risk right how do you make sure that if I give money to another agent it's actually managed properly um and I I won't I won't lose my capital right so I think a lot of

these uh risk standards that we've developed uh with teams out there is literally to solve for how you handle large amount of money >> so that's it right how large amount money for specific trading and new new management u but going into the into the real world I I think I think it'll be it'll be less of a problem because I feel like a lot of times money like today if you if you look into your into your industries today right the money portion is actually held by all the white collar workers right like like like that is where your digital agent need to figure out right but your physical laborers that don't really handle much money like in the end the the biggest money movement in a retail outlet is when the cashier takes that

capital that's being uh being being being made in the day and then put it in a bank, right? The actual transaction between the robots are very low. So, I honestly think um risk management for physical robots is not going to be an issue compared to to digital robots. They will have their own issues which we do not know today, >> but you know, you have solved them then. >> Jensen, thank you for joining us today on today's AI Supercycle episode number 28 powered by our friends at Near. Man, let's give Jensen some love. He is an OG in the agentic space. Thank you for joining us and we will see you on the next time that you come on. I hope to come out to Asia and see the warehouse, see the robots at some point as well.

Thanks for joining us. >> All right, cheers. Thanks guys.