Nils Pihl (CEO, Auki Labs) on Making the Physical World Accessible to AI

And we're back. Well, actually, we're not back. We're here at the inaugural Intercognitive Foundation podcast. My name is Rich Robinson. I am founding chairman, proud founding chairman of the Intercognitive Foundation, which was established with the mission and vision to make the physical world accessible to AI. And ladies and germs, boys and girls, I was silver-pilled by the gentleman across from me, the co-founder and CEO of Aukey Labs, Nils Pihl. Everybody put your hands together. I don't care if you're driving or you're in a bathroom stall and it's kind of weird. I just want a round of applause for, for, for this guy. Thank you very much. Excellent. Thank you, Jamie. Um, oh, are we on the Rogan Show? Uh, this is a, a 7-hour podcast with Nils. And let's go. We're going into it. Thank you so much. We're here in my hometown, the warm embrace of Mother Ubud on the island of the gods. Nils is visiting with his family after being all over the multiverse, getting a mini shampoo collection, shouting from the rooftops about Aoki and his mission and vision. I would love for our listeners to hear a little bit of a background, some stories, anecdotes from you growing up in the Nordics, and then your path to China where we first met. Now you're based in Hong Kong, and then we'll lead up to what you're doing right now and Intercognitive Foundation and the future. But let's rewind back to the origin story when you were bitten by a radioactive robot back in the day.

Sure. So, I was born in '86 to software entrepreneur parents, which was a lucky hand of cards to get dealt, for sure.

Sure.

As a result, I grew up on the internet, and the internet was a very interesting place back then because people didn't know how to use it yet. Your typical parents didn't know how to teach you how to get on. So the, the internet was a very cool place full of very pioneering people that had figured it out on their own. You know, this is before my grandma got a Facebook account, right?

Mm-hmm.

And, um, I spent so much time online that my, uh, by, by my mid-teens or so, my, my English was better than my Swedish, which my mom thought was silly. We had a lot of arguments about being social, and I kept telling her, like, you don't understand just how social I'm being. You know, I'm having multiple conversations at the same time with people all over the world. I remember, actually, I remember 9/11. As soon as it happens, I can go online and I talk to my friends who are in New York and are seeing it out out their windows and feeling that this is the ground floor of the metaverse, right? Like, there's people here whose existence is online. Their emotions are online.

Mm-hmm.

I know them by their usernames. That's how they identify. And information is propagating in, in real time, and emotions are propagating in in real time and just realized I really wanna be a part of building the metaverse, if you will. And I came to the conclusion what the metaverse is, is if anything, a state of mind, right? The metaverse is when you embrace your digital identity. It's a suspension of disbelief.

Mm-hmm.

It's when you, You are in the metaverse when you are that online persona and you fully accept your online friends as real, et cetera. Maybe it seems mundane now, but it shouldn't be like, this is a still a very new thing for humanity. I gave a talk a couple of years ago.

Yeah.

about kind of the shamanic progression of technology where, you know, if you go far enough back, we lived in an oral tradition, right? And what's interesting with oral traditions is any piece of information that we want to retain, we have to keep it alive in conversation. If you are the last person that knows how to make a spear and you die, then spear making is lost. So oral traditions require a lot of attention and care to keep information.

Storytelling.

Yes. Right. And you have to be intentional with your storytelling to be able to retain a knowledge base for your tribe. Right. And then writing was invented, which allowed us to persist memes over time. Super big, super, super big. And allowed us a new mode of consciousness, reading, right?

Mm-hmm.

It's like a new human being. And after that, we get the printing press, and the printing press allows this like mass propagation of memes like never before. And now you can build up group consciousness in a way that you couldn't. Uh, and then, you know, we, we get the, the telephone. And the telephone, I think, is underappreciated for its cyberdelic impact because the telephone was the first time that you could have a moment of togetherness at a distance.

Hmm.

Right? So what do I mean by togetherness? You know, in talking about the metaverse, I think a lot of people intuitively think that, oh, being together with someone is like this physical thing, but everyone has felt alone in a classroom or at a party, right? So it can't be that togetherness is a function of proximity to people. And growing up online, I realized that you can feel togetherness with people at a distance.

And at 9/11, you felt like you were right there talking to established contacts and seeing 9/11 through their eyes in New York. It's fascinating.

It was interesting to have an emotional— I wouldn't say I was right there, obviously, stolen valor, but, uh, it was interesting to, to, to get to feel with them.

Yeah, sure, sure. And that kind of— that took you there in a way that reading about it on the news or just, you know, hearing about it secondhand, you know, there really is a— there really is a connection there that, uh, I think I'd like to pull on those, those, those threads a little more. And by the way, Uh, tell your mom that whatever kind of social stuff you were doing, it's, it, it worked because your, your storytelling ability and your, um, ability to plant a meme inside somebody else's head really works. And I know because you unscrewed the top of my skull and then took a tiller to my brain and then planted a whole bunch of seeds and now they're coming outta my ears.

So, well, I, I think, um, uh, Beijing, back when we first met, was a lot like the internet that I grew up on.

Hmm.

In that, um, you know, growing up online in the late '90s, early 2000s, people were always being like, hey, don't you want to go outside? Don't you miss the blue skies? Shouldn't you be doing something else with your time? But like, no, I, I want to be here with all of these other pioneering people that have found their way here somehow. And that's what I fell in love with, with Beijing. You know, like Beijing is full of people that are not Beijingers, right? They're from all over China, they're from all over the world, and everyone's made it there for some reason and want to be part of something. And you will hear the same kind of haters, you know, don't you miss the blue skies? It's like, this is— Beijing felt like the early internet before my grandma got her Facebook account.

That's great. And you were in Beijing from when to when?

From 2009 to 2017.

Yeah, so that was a really golden, golden, golden era that I think in the expanse of Chinese history of 5,000 years, that little window from like, I would say '99 to 2019, 2020, like that was a very magical era. It was a catch basin for people who were self-selecting. And it wasn't just China 'cause Beijing was tougher. Beijing was barely fit for human habitation in some ways with the pollution and the weather and the sandstorms and the traffic. And I love Beijing, but I don't like it sometimes because of those challenges. But I wasn't leaving and I stayed there 20 years myself. So I'd never thought about that, that there really was this parallel between the internet and very pioneering. And what took you there and what kept you there?

So in my early 20s, I took a sabbatical. I thought I was writing a book. It turns out I was not. But I took a sabbatical to write a book. And my girlfriend at the time was a long-distance relationship, American-born Chinese, had never been to China. And she said, I want to go do a semester abroad. I was like, well, I'm not doing anything. I'll, I'll tag along. And I was in love with Beijing already on the cab ride from the airport.

Yes.

Right?

Yes.

Um, I was—

If you know, you know.

Yeah. I mean, I was so excited to see all the construction that was going on. It was fun trying to, you know, learn some Chinese just by reading the road signs. And it just felt very much like a great computer game and a great adventure that I wanted to be part of. And long story short, she did not stay in Beijing, but I did, right?

Mm-hmm.

I ended up staying 7 years.

Um, it's not for the faint-hearted, that's for sure.

Yeah, I could— so I have this pet peeve with the, with the Western psyche. I, I think my generation was brought up to believe that we live in post-history, that like history is done. Like after the fall of the Berlin Wall, we now live in the established world order. Um, Yeah, in post-history. And it was very clear to me when I, uh, got to Beijing that all of that was a lie. History's still on, like it's happening. You can still do great things. People are gonna do great things. Um, and as a, as an avid strategy gamer, you know, I could, I could tell that like, oh, the turn-based science output here is very, it's very great. If, if I stay here for 10 years, I will have seen China go from, you know, arguably a third world country to the world's leading superpower. And yeah, over the last 20 years, they've just gone through this amazing transformation. And I think just the last 5 years in particular have just, or since COVID just mind-blowing. Even living in Hong Kong, kind of feel a little bit like that. that frog getting boiled. You know, you don't, you don't notice how much it's developed until you leave. When you go back to Europe, when you go back to America and you think like, oh wow, the rest of the world has not moved at this pace. There's something really astonishing happening in China right now. And of course, part of what's coming out of that is now the robotics scene. This amazing plethora of robotics companies that are building very openly, making programmable robots. They're collaborating with each other and just accelerating the timeline of civilization.

There's, there's that word civilization. That's what you used. There's so few people that use the word civilization in Business or normal conversation outside of gaming. And it's something that I've heard you use a number, a number of times. And I remember hiking with you in Hong Kong and you talked to me about this robot revolution and where it's, where it's going, uh, in the, in the future on a civilizational scale. And I just couldn't help but take— you take this blue pill and go back to your bullshit regular life. Take the silver pill. And come with me, Rich, into the future.

I, I think there are a couple of civilization-scale things that are happening now. Um, almost like unstoppable winds that as long as we can, you know, hold on to civilization, we'll see some, yeah, some indeed, some pretty—

Civilization can change. Yes. Mm-hmm.

Things are gonna change pretty, pretty quickly.

End of, end of history, maybe not. End of civilization, hopefully not.

Yes. I think, One of the first civilization-scale things that I got hooked on was this concept of decentralized infrastructure, which is really a business model innovation. Our good friend Mike—

Mike Horton, GeoNet.

Mike Horton, GeoNet, is, I think, one of the best examples of what is decentralized infrastructure and why is it so powerful? So a lot of things that we need in the modern world requires infrastructure. Wi-Fi needs Wi-Fi infrastructure, cellular needs cellular infrastructure, running water, all of these things. There's a lot of infrastructure that we need. And typically, historically, it has either been the state that puts down the infrastructure or some corporation that puts down the infrastructure.

Right.

And they have to own the capital expenditure. They have to own the strategy and the planning. But people like Mike have demonstrated that you can leave infrastructure buildout to the original artificial intelligence, the free market.

Mm-hmm.

So the free market has amazing emergent intelligence capabilities.

Wow. I never thought of the free hand of Adam Smith as artificial intelligence, but it certainly is. It is.

Yeah. That's our first AI, or maybe our second, depending on how you view the big memeplexes like monotheistic religions. But it's a very powerful AI, the free market. And the idea was that, okay, so In Geodnet's case, they put out something called RTK base stations. RTK is a piece of hardware, stands for real-time kinematics, that allows you to get better accuracy from the GPS. So typically, the GPS is a line-of-sight technology. It typically only gives you a few meters of accuracy. It's not good enough for something like an autonomous combine harvester or any kind of agricultural machinery where you want to be able to know that you're going in a straight line. You want to know where you've planted the plants. You want to get that down to centimeter accuracy. And you can't get that from the GPS in the sky. But with RTK, you can. You put base stations out in the world.

I have one on my roof here in Ubud. Yes.

There you go. Right. And you trilaterate against those and get better.

And instead of plus or minus 20, 30 meters, it's plus or minus 2, 3 centimeters.

Amazing.

Which is extraordinary. Right. Orders of magnitude. Right.

So And what, uh, what Mike did is he told the free market, look, I'm gonna put out a map and show you where we have coverage. I am gonna let you buy this infrastructure and you can place it. And if you place it somewhere where people end up using it, I'll share the revenue with you. So that means that he shares the CapEx with the free market. He shares the analysis of where the nodes need to go with the free market. And he also shares the upside with the free market. And long story short, he has built the world's largest RTK network. And this business model is often called DePIN, decentralized physical infrastructure networks. There's a couple of really amazing DePINS, not just Geonet, but also things like Helium is a very interesting a DePIN that helps give cell phone coverage indoors. So I have friends in San Francisco that just as a hobby go visit different restaurants and put a Helium base station there so that there's proper coverage inside and make good money on making sure that there's coverage in all of these indoor spaces.

Mm-hmm.

What we're building at Auki is also a decentralized infrastructure. We're related to what Mike is doing with GeoNet. We're working on how machines understand the physical world, how they do positioning, and how they interpret their sensor data to make sense of the world around them collaboratively. And what the reason I fell into that particular rabbit hole was in, in 2019, uh, I just, uh, lost my job and I was gonna do something fun. I thought, uh, I got into Warhammer, the tabletop game. Um, and slippery slope.

That's a, that's all consuming. My boys are into that.

Yeah. I got super fascinated with the company that makes Warhammer Games Workshop, and I thought, I, I'd like to go work at Games Workshop. But I'm not gonna go apply like a schmuck. I'm gonna build something and sell them so I can go, so I can go work there.

I don't know this story. Oh, excellent.

So, uh, I spent the summer thinking about what can I build and sell to Games Workshop? And I decided I am gonna make augmented reality overlays, uh, so that people can do measurements through terrain. They can see, you know, which things are in inside the range of their aura, but also make it so that you can have a proper esports viewing experience for the 2 million people or so that were watching Warhammer games online during COVID right? It's like, I'll do that. I'll sell that to them. Didn't know anything about augmented reality, right? So I immediately had the cold shower that I think people are increasingly having when they get into things like robotics. that the GPS is just not gonna cut it, right? You, I needed to build shared augmented reality experiences where we would see the same thing in the same place at the same time, which means technically you need to make sure that your device and my device are in agreement on a coordinate system down to centimeter accuracy. And they don't have that out of the box, right?

Mm-hmm.

So outdoors you can use things like RTK. RTK is also line of sight. actually. But indoors, we couldn't use RTK in here for shared augmented reality. So you need different kinds of positioning for different kinds of scenarios. And it seemed clear that what was going to win indoor positioning is visual positioning, meaning that you compare what the camera is seeing to a previous model of the world and compute, well, if this is what you're seeing, this must be your precise location.

Mm-hmm.

That works comparatively well. But the issue is whenever you do a comparison like that, both datasets have to be in the same place. So either I'm going to be sending my model of the world down to your future AR glasses or down to your robot and These models of the world are big, so not super practical, right? Mm-hmm. You have to wait for long, long downloads to be able to figure out where you are, or you send what you're looking at to someone's cloud somewhere. And that is what happens in 99.99% of cases. And I realized that was a massively bad thing for our civil liberties, for our, for our cognitive freedom. It would be very bad for literally all of our future AR glasses to be streaming at all times to some central entity somewhere that knows what we're looking at.

Mm-hmm.

Because that's required for them to work, right? So I wanted to build a D-PIN for visual positioning where you could, instead of building one big map of the world, that everyone streams to. You allow any venue to self-host something that describes their venue, kind of like how the internet is not just one big cloud somewhere. It's a lot of people's different web servers, et cetera. And that's what makes the internet great and resilient and free. And the physical internet needs to work the same way.

Hmm.

In fact, I believe that the internet needs to grow into 3 new dimensions to be ready for the physical AI era. We need an internet of spaces, which is what we started building, right? Machines need to be able to browse and search and navigate physical spaces that they haven't been to before. So this is an internet of spaces where you can connect to a local venue and ask where the bathroom is and not have to go look for it.

Mm-hmm. Right?

You want your humanoid robot to be able to go to Walmart and find the ketchup without having to look through all of Walmart to find the ketchup.

Mm-hmm.

Right? The Walmart AI should be able to answer your humanoid, here's where the ketchup is. So that's an internet of spaces. But you probably also need, almost certainly, an internet of sensors because physical AI is not going to be very efficient if it's tied to just a single embodiment. They can do a lot more if they can borrow the sensor information from other devices around them. My favourite example of this is, A Beijing story for scale. For those of you that don't know, you know this of course, Beijing is a very big city. In fact, it's so big that there are more cars on the road in Beijing than there are people in Los Angeles. And the average commute by car is something like 2 hours. So if you do the quick back of the napkin calculation, you find that roughly the time it took to build the pyramids is lost in Beijing traffic every single week.

Boom.

Right. Um, now we can imagine a world where all the cars are self-driving and they are coordinating with each other. Um, and we, they don't have to stop, uh, for red lights. And, you know, we, we get some good fraction of a pyramid worth of human productivity back every single week. But that can only happen if the cars can coordinate with each other, if they can do things like ask, hey, you just went around this corner. Uh, may I ask you, is there any reason for me to, to slow down? Um, they need to be able to communicate with each other about where they are precisely in space, uh, how fast they're moving down to the centimeter, all of these things. Tricky, tricky stuff to build, but the economic incentive is just so huge, right? There are pyramids worth of human productivity to get back just, just in Beijing. So an internet of spaces, an internet of sensors. And then the third dimension is an internet of actuators, the ability for physical AI to borrow robots around them to solve different tasks. You can think of it as teleoperation, but for robots teleoperating other robots.

Wow.

If you will.

Wow.

Because why shouldn't your physical AI be able to sign into a robot in New York, go shopping for you for some Physical item, go drop that off in the mail, and then sign out of that robot. Why not?

Fascinating. I think right now with agentic AI, people are building agents to sell to other agents, not even to sell to other, other humans. So yes, of course, that generative AI that we're in now is, is going to agentic, and then that's gonna go to embodied, and then robots will be controlling or living through other robots. Wow. Mind, mind blown. Uh, once again, thank you.

There's a big conversation in robotics, uh, right now about something called cross-embodiment, which is, um, you know, an open challenge. And how do you get any kind of physical AI to be good at inhabiting different kinds of bodies? Uh, you know, can you have one AI that is just as comfortable living inside a dog robot as living inside a humanoid robot? This is the cross-embodiment problem. problem. But I think an underappreciated problem, um, uh, that I just wrote about recently is co-embodiment. The idea that you should be able to have many different AIs inhabit the same robot body and come and go, not as they please, but, you know, uh, be able to be, uh, tempor— temporary guests inside any kind of robot body.

Wow.

Um, because Well, I have some friends who make glasses. They make smart glasses. They just launched about a month ago. And like me, they are big believers that glasses are going to be just as popular as smartphones within a couple of years because it's the best way for you to work very closely with AI, right? With glasses, you can let AI see what you see, hear what you hear, just have the full context of of what it is you're doing and it can be proactive and just turn you into a superhuman. I fully believe that. I'm glasses pilled, right? But what these guys realized, they're called MENTRA, is that their big competitors like Meta, who's making the Ray-Bans, are all trying to build one AI for the glasses and you just have to deal with that one AI. But that's not how your phone works. That's not how your computer works. There's an app store that lets you choose between different capabilities. And that's good because different intelligence providers are good at different things.

Mm-hmm.

Like anyone that works a lot with AI realizes this, you know, Claude is good at this, ChatGPT is good at this, et cetera. So they embraced co-embodiment for the glasses and made an operating system for the glasses that allows multiple apps and multiple AIs to have shared access to the camera, shared access to the microphone, shared access to the context of the world.

Wow.

Because essentially, if you're wearing the glasses and you're in Shenzhen and you're negotiating, you want your note-taker to be able to listen to the conversation. You want your translation AI to be able to listen to the conversation. You want your personal assistant AI to be able to listen to the— like, all of these different subfunctions need shared access. And I think that they actually stumbled on something that's going to be very important for robotics, co-embodiment.

Wow. Tank, I need to learn how to fly a Black Hawk. Let's go. I know kung fu. Like, I can imagine what the different co-embodiment agents could be like, or AIs. Like, can you give an example of like one or two?

Yeah. So at Aukey this year, we're gonna be deploying wheeled humanoids into retail environments, but not to actually manipulate anything. Actually, if I can take 2 steps back, right? You could say that robots have roughly 3 capabilities if we simplify the world a little bit.

Let's simplify, yeah.

Yeah, so they have the ability to do locomotion, which is to traverse an environment. There's manipulation, which means to affect change on the environment. And then there's perception, to make observations about the environment. Modern robotics, of course, combines all of these capabilities to make increasingly more capable robots. But interestingly, you can say that the kind of jobs that a robot could do in the world roughly maps to these same 3 buckets, right? So there are locomotion tasks, like a Tesla self-driving car is a locomotion robot.

Mm-hmm.

Like, sure, it does perception, it has to, but it does it in order to solve a locomotion job.

Right.

Weave makes a laundry folding robot and it uses perception, of course, but it does so in order to perform a manipulation task. And I submit to you that there are also just full-on perception tasks where maybe you'll have to use some locomotion to perform it well, but the point is to go observe the world. And one of my recent favorite examples is my friend Jonathan Moon at Budbreak. They make robots for vineyards. And when I say that, most people will jump to the conclusion like, oh, this robot is going to either be harvesting or pruning or spraying, doing some manipulation task. But actually, no, they're doing a perception task.

Why?

Well, when you do a robot deployment, there are 3 things that you need to consider, right? Like, one is, what is the value of the labor that you're producing? You either have to be cheaper than the human doing the job or better than the human doing the job. And right now, with the state of the art, it would be very hard to make a grape harvesting robot or pruning robot or spraying robot that would either be cheaper than the minimum wage that, you know, is typical in these kind of agricultural scenarios.

Mm-hmm.

Or better than the human. Very, very difficult to even have a 10% increase. And then you have to think about cost, right? Like how much R&D is it to even get to something? How much is it gonna cost to maintain it? And, you know, between value and cost, we can calculate some kind of ROI. But the third thing, which is very important, is what is the risk, right? If this robot fucks up, how much shit am I in? Right? If you have a manipulation robot building luxury cars and the manipulation robot fails, you have either broken that car or you've stopped the entire production line. So you have a big reliability cliff. You need to be able to show the customer that we have 99.999 uptime because it's just very costly if the robot fails. And think of Tesla. Right. The self-driving car value comparable to a human driver. So sure, there's some value there. Cost doesn't need— I mean, there's a lot of R&D, but after the R&D, you know, once those billions are spent, a self-driving car does not have to be meaningfully more expensive than a normal car, which is good. But what's the risk?

Right.

When a self-driving car fails, People die. So very, very big reliability cliff. And what Jonathan at Budbreak realized is that there are very valuable, low-cost, low-risk perception tasks in agriculture, right? So for example, when it comes to vineyards, there's a single crop-destroying disease. I'm blanking on the name now, but it destroys up to $60,000 worth of grapes per hectare per year. And what Jonathan did is he worked with experts to recognize this disease and train computer vision models that recognized this disease very early. And he built a Mars rover for vineyards, like a very rugged vehicle that autonomously drives through the vineyard and does this very advanced computer vision, which is now better at detecting this disease than the people working the vineyard. Right. Which means that instead of doing the pruning or the harvesting better or cheaper than a human, it's instead increasing the yield of the whole vineyard because it can find these early signs of this blight so that the farmer can go—

The grape doctor is in, right? Yeah. Just like in the medical world, early detection. is really important. And then also I met the founder of Harrison.ai, Vietnamese founder moved to Boston, and they do AI detection of scans and they're already 40% better than most humans.

Amazing.

So why not bring that to other biological areas where there is some sort of disease or blight? Fascinating.

Exactly.

So that's where perception alone just creates a lot, so much value.

Allows you to capture some territory there, go out and do something. And what does this have to do with co-embodiment, right? Well, for one, you, there might be very many different perception stacks that you might want to run on the same robot, obviously. So one team that's not Budbreak might train some computer vision that does good yield estimation or something like this. And it would be really nice if they could run that same intelligence on the same robot. But also, Budbreak is putting a modular arm on the Mars rover thing, initially just to bring a camera closer to what they're looking at. But eventually, as manipulation gets better, that arm will be able to maybe do some pruning, maybe do some harvesting, right?

Maybe spray something for the blight, who knows, right?

That's right.

And all of the above.

The point is, Jonathan does not have to be the person to solve that necessarily.

Mm-hmm.

If you embrace co-embodiment, right? Some other lab can be like, well, okay, so there are all of these robots out there that have this hardware that could support my use case already. I'm gonna deploy onto those robots. So we're doing something similar in retail this year where we're taking wheeled humanoids, And we're deploying them in a retail environment to do perception tasks. Again, when we say we're going to put wheeled humanoids in retail, people assume it's restocking or order picking. No, the software is not quite ready for that. The hardware might be, probably is, but the software isn't. But what you can teach one of these robots to do good enough this year is task generation for the humans.

So that's fascinating. Let's double-click on that. So this is in your home country of Sweden that you're deploying these. You and I, in December, you came and gave a guest lecture at my class I teach at Peking University. I teach entrepreneurship. And you completely unscrewed the top of everybody's skulls and retooled, rewired their brains. And one of my students ended up getting into robotics and he's now working at Galbot, Which we visited together.

Yeah.

And we visited some other companies and you laid out your not just vision, but how you're practically doing this. Because of course, from the wrist up, the hand, I mean, that's as complex as the rest of the robot. So let's just wait until that catches up through AI and through advances and actuators and, you know, et cetera, and co-embodied and whatever. But in the meantime, so you just said the robots are helping the humans be more like, give tasks to the humans instead of the humans giving tasks to the robot. The robot is now giving tasks.

That's right.

To the human.

So it's a store manager robot.

So somebody shows up at this retail location And what are they— what kind of data are they being handed or what kind of tasks are they being handed? And how have you pulled that together?

So the first thing we built was an augmented reality task management system that allowed the human store manager to go assign tasks in physical space to different people. And now the robot is going to use the same augmented reality task manager to assign tasks to humans. And it can do things like find empty shelves that need to be restocked. It can find missing or erroneous price tags, things that need to be cleaned up, just the kind of stuff that the store manager would tell the associates to fix, the robot will fix. And what turned us on to this is we had one of Europe's largest retailers with over 10,000 associates out on the floor, which is small compared to like Walmart has 2 million associates out on the floor.

Wow.

But over 10,000 associates, and they flew to our lab in Hong Kong and they said, hey, here's a problem. We estimate that we lose 30 to 60 minutes per associate per day, predominantly in the morning, to them not being super sure what they should be working on.

Wow. I mean, that's a huge needle mover. Think about the literally hundreds of thousands of hours every month. That add up. And at Scandinavian wages, that's real.

So what we realize we can do is we can take hardware that is in theory ready for manipulation tasks like restocking, et cetera, and deploy them today to do perception tasks. And when the software is ready, either because we made it or someone else made it, it can co-embody in this robot. Like we can go capture that territory already and make sure that there is robots as infrastructure, that when someone has made an AI that can do restocking, there are already robots as infrastructure.

Robots as infrastructure.

That, that AI can inhabit, right? So you send your restocking agent to go live inside the GalBot that Aukey deployed. And the reason I think this is important is one, it's going to allow us to actually deploy the robots faster with something that gets demonstrable ROI now already. But last night I was having a long, painful conversation with a big retailer with 20,000 locations, one of the biggest in the world by location count. And they wanted to brainstorm about the robots. They're just so difficult to deal with because it's like, oh, we, you know, our dream is a robot that does absolutely everything.

Mm-hmm.

Right? There, there has to be some initial ROI before we can do anything. Right? And we've looked at, you know, people working on maybe restocking robots, but if that's all it does, ah, we don't, we don't know. 'Cause we don't wanna have to get a second and a third and a fourth robot. Like our stores are not that big. It has to be one robot that eventually And maybe it's better for us to wait until there's a robot that does everything because we don't want to have to keep switching robots.

Which seems like an ignorant but kind of a fair proposition. It's a fair thing to say.

And I think the robot labs have not, because a lot of them are PhDs that never talk to customers, have not realized that this is going to be a big sales blocker. And what allows us to unblock these conversations with these big retailers is like, here's how you will get Very clear ROI on the perception tasks first, right? And this will justify you putting the robot in there. And that means that when these other tasks are ready, like restocking or order picking, that's a software update on the robot as infrastructure that you've already deployed.

Amazing. Just like in The Matrix, just call up Tank and he just, you know, Basically teaches you kung fu. Kung fu.

Yeah.

Wow. Wow.

So that's been a big push for me these last 100 days or so, talking to other robotics labs about, hey, we need to figure out some open standards for co-embodiment. Everyone's working on cross-embodiment and that's super important, but co-embodiment would be such a big unlock if we could figure out how to make something like a robot OS That allows us to share a robot body.

And you also mentioned, I thought was quite fascinating if you think about it, like I saw MrBeast, he launched Feastables, his chocolate brand, and he said that he had to dispatch a lot of people to the stores to make sure they were displayed properly and they didn't fall over. And you're basically in some ways instead of metal humanoids, it's meat robots that you're dispatching there so that you can perceive the world through them.

So, so, so funny that you say that.

So, so, so now, now there's your robot. It's, it's already in the store. Everybody's, everybody's productive. It's got a little free time. And, and then who, who could take over the robot during the day?

Yeah. So every Friday we have internal demo day. where we show each other what we've been working on at Aukey. And one of the demos from this Friday was sign in to do tele-inspection. So to give a little bit of background for scale, one of the world's most famous beverage brands told us that in the country—

Its name will remain Coca-Cola. Yes. Anyway, I'm guessing.

Told us that In Sweden alone, they have hundreds of field reps whose job is hop in a car, drive out to the retail location, make sure that things are set up properly and that contracts are being followed and blah, blah, blah. And back of the napkin calculation, they said it probably costs us on the order of $100 per inspection, right? Total cost of everything. Some inspections obviously cheaper, some of them obviously more more expensive, but, you know, on the order of $100 per inspection, they said. And what we demonstrated this Friday is that instead of hopping in a car, you can remotely hop into the robot. So you sign into the robot at the store that you want to visit and you say, hey, here are my Coca-Cola, Pepsi, whatever credentials. Please take me to my shelf so I can look at it.

In a cap or whatever. Yes.

And then the robot autonomously drives there. Turns off the camera so you can't be spying on everything else because the retailers are very happy with that part of the feature. And then when it arrives, it turns on the camera so that you can do a remote inspection.

And like full 360 camera video and captures and gives recommendations.

With a semi-humanoid, they'll be able to even, you know, reach in and look because there's a camera on the wrist. So they'll be able to look in and try to count the tops of like how many get a sense of how many there are. And this use case is interesting enough to some of the biggest brands out there that we're putting together a consortium of them now to actually sponsor the robots so that the retailers don't have to buy the robot.

Oh.

Just, they'll pay for the robot and put the robot out there because they save so much money on not doing the Not doing the inspections manually.

Fascinating. Love it. Thank you for triple-clicking on Alki Labs and your vision. And, you know, you and I knew each other from Beijing, and then we reconnected in late 2024. We went on a hike in Hong Kong, and I've been a serial entrepreneur, but I've also been a serial ecosystem guy. I love bringing people together and just creating a space for people to collaborate and connect. And you had this idea with Mike from GeoNet and the guys from, uh, Moari and, and, and Peak to, uh, found the logo on my hat, Intercognitive Foundation, which first of all, what an amazing, beautiful name, you know, knowledge in between, intercognitive.com. And you waxed futuristic about this vision of making the physical world accessible to AI. We've been talking about Humanoids at one end and then deep in and data at the other. And then of course there's training and orchestration and apps and payment and compute and maps and all these moving parts, literally and figuratively, to make the physical world accessible for AI. There are industry groups for robotics, for deep in, for payments and mapping and everything. But to have all these layers, these 9 layers that you've identified, to be able to sing and dance together. That's really a beautiful vision when humanoids will roam among us and be able to pass Steve Wozniak's coffee test of summoning a robot into your home to make a cup of coffee. And so please share with the listeners about the origin story and your vision of Intercognitive.

The honest origin story. without fluff was, uh, that is— so, uh, Geoden and Aukey are both working on positioning for different kinds of environments. You know, they make really, really good positioning for, for the outdoors, very good for agriculture, and we make really good positioning for the indoors. Um, and Mike and I, we kept saying like, obviously at some point in the future you're going to be able to switch between these protocols and they're going to work seamlessly.

Mm-hmm.

We, we should work on that integration. But up until basically this year, it was never his priority or my priority to go do it at any given moment.

Of course.

Like, it would be a great tech demo, but who's the customer now? Like, the robots weren't there yet. The use cases weren't there yet. So eventually we said, okay, what if we pooled some money and put out like an open source bounty to go do it? Because our time is more valuable than our money. We could pool some money to get this done. And that idea just kind of spiraled into, actually, there's a whole laundry list of things that need to get done. Because if you take a step back, 70% of the world economy, both by GDP and by headcount, is tied to physical locations and physical labor. So the biggest impact that AI will ever have is in the physical world, right? Like when you go from—

Great framing.

From agentic AI to physical AI, you've 3 to 4x'd the total addressable market.

Wow.

Right off the bat, right? So, you know, Anthropic, OpenAI, they're gonna be significantly smaller. than whoever becomes the predominant physical AI company. Which means, yeah, probably the biggest opportunity in civilization now, apart from longevity, maybe. There's that civilization part that got me so excited.

The biggest industry in the history of mankind and civilization.

Yes. And, but we need, AI can only work on digital things, right? And the physical world is not digital. So an obvious thing that needed to be fixed is, okay, we need to start digitizing the physical world. We need to start mapping it. We need to solve positioning. But there are all kinds of other things that you don't realize is an issue when you're in the lab. Like, how does the robot have internet connection? How does the robot identify itself? Why does it need to identify itself? Well, because the robot needs to be able to, for example, call a cab or hail an elevator.

Enter a building.

Enter a building. It needs to identify itself.

Pick up something. Yeah.

Today, robots have a place of work, so it's not an issue. But clearly we want to get to a point where your humanoid robot can go to my apartment complex and pick up a cup of sugar for you and go back, right?

Mm-hmm.

But then, yeah, how, how does it identify itself? How does it pay for things? How do you insure a robot, right? How do you ship a robot across borders today? There are so many things that need to happen in the physical world before robots can start rolling out at scale. And we realized that there's no industry group that's taking a holistic view at how ready are we as a civilization to have robots. So let's put even more money in, not just put up one little bounty, but Let's try to fund an industry group who goes through the checklist of how do we get civilization ready for the robots.

And that's Intercognitive.

And, uh, that's Intercognitive.

I, uh, I learned a really good lesson in my life. I've met a few people, I like your, like your good self, who really can see around corners and then the next corner and then under and over and around. And Yat Siu, the founder of Animoca Brands, he asked me to be an advisor and I quickly said yes. And Balaji Srinivasan, the founder of Network State, I was part of the first cohort and I'm very involved there. And your good self where I feel a little bit challenged because who really does fully, fully understand the entirety of AI and robotics and making AI accessible for the physical world? Probably, probably very few people. But that's okay because I really love to create a space for people and to be able to bring people together. And then I bring together all the experts and then you guys do all the magic around research and standard settings. And then I just have to amplify that message. And then along the way, I get to selfishly build up an amazing network and see problems that I can then invest, advise, and, and start up. So typically I would do a startup and then I would start an industry group because it would be a great way to become a thought leader and create a network, not in a glad-handing transactional way, but in a, you know, I want to be able to have a philosophy of abundance and bring people together. But now you convinced me, and I'm very happy I took that silver pill to front-load the industry group and then do this for a couple of years and then figure out what I'm going to do in this space. And it's been an absolutely insane ride. And, you know, we've doubled membership. We want to double it again and, you know, double it yet again by the end of the year and bring in all the different stakeholders to be able to make the physical world accessible for AI. I thank you My friend, both for bringing me into— giving me this silver pill for Intercognitive Foundation, and also joining me while you're here on a well-deserved little family holiday on the Island of the Gods. Put your hands together once again for Neil's Pill, everybody. Thank you.

Thank you for the conversation and coffee.

Awesome. Thank you, my friend.

That was fun.