Subscribe
Share
Share
Embed
The World is Moving Fast. Futureproof Yourself Weekly.
Your front-row seat to the Future of Business, Tech, Life, and Human Potential.
On Futurecaster with Kimberly Bates, learn through thought-provoking conversations with world-leading founders, AI experts, scientists, engineers, doctors, business leaders, technologists, investors, and everyday builders. Each episode unpacks the real opportunities shaping what’s to come and the blind spots along the way.
Kimberly Bates is a world-renowned futurist and executive brand leader. For over 20 years, she has advised some of the world’s most valuable global companies and iconic brands, from Fortune 100s to fast-moving startups, helping them anticipate change and shape the direction of their industries, customers, and culture. Her work has driven future-ready brand and business transformations, breakthrough products and services, and entirely new business models and revenue streams.
The Futurecaster® Podcast is on a mission to make future thinking available to everyone. If people can see these opportunities for themselves, they can help create them.
©2026 Futurecaster, LLC | Futurecaster© is a registered trademark
best robot up until that moment was trained on 1.3 billion training instances and it
scored, say, a level 50.
And so we brought our system in to do the test with no pre-training, just the very basic
things of what a house is or what a table is or different kinds of basic...
concepts and then gave it the goals and then it solved the puzzle at a level 70 with no
pre-training.
And so that means there's no cost and that means it can learn and evolve and grow.
So it could look at the problem and realize it couldn't solve it the obvious way.
It had to go around some obstacles and then solve it a different way.
And it figured it out itself.
And so that's really what we want from AI.
So this world of active inference AI is a world of
of AIs that can reason, they can learn, they can adapt, they can plan, they have agency.
So the problem with LLMs is
Welcome to the FutureCaster Podcast where we give you a front row seat into the future of
business, life, and human potential.
While the world obsesses over chatbots and text generators, a quiet revolution is
unfolding.
And it may be the real path to agentic AI, Super intelligence, and AGI.
When most people imagine the future of AI, they picture smarter assistants that book
vacations, automate tasks at the office, or shop for you.
But the real breakthrough won't come from scaling models.
It'll come from embodied AI, machines that learn like babies do, by seeing, touching,
hearing, moving, and constantly updating their internal model of how the world works.
We call these WORLD MODELS.
Just as infants spend hours wiggling their fingers and watching gravity in action, future
systems will learn through physical experience, building a far deeper understanding of our
messy, unpredictable real world.
This shift is huge because language models can only take what's been given to them and
guess at reality.
They don't truly understand the consequences of it.
Embodied AI is what will unlock reliable robots, real autonomy, and systems that can
reason, plan, and act with common sense.
Chatbots were only just the opening act.
The real future of AI isn't just on your screen.
It's out here with us learning, moving, and shaping the physical world.
Today, I'm speaking with DAN MAPES, Founder of the Spatial Web Foundation and Verses AI
(World Model Company), whose Chief Scientist is Carl Friston, one of the most published
and respected neuroscientists in the world.
Mapes argues that the true breakthrough will come from the FREE ENERGY PRINCIPLE and
ACTIVE INFERENCE a Neuroscience-inspired framework that could give machines the ability to
understand, adapt, act,
and self-regulate in the world like human beings do
If you've heard about the future of AI, but haven't heard about active inference and the
spatial web, you haven't seen what's coming NEXT!
Hi, Dan.
Welcome to Futurecaster.
I have been incredibly excited to speak with you.
I couldn't wait to get you on the podcast.
Thanks for joining us today.
Kimberly, it's my pleasure.
Real exciting to be here.
I love your futurist orientation and your understanding of AI, so it should be a good
conversation.
I think it's interesting for the audience to understand who you are and where you've come
from into this conversation today.
let's start off with your personal journey.
What first drew you into the spatial web and AI and all the work you're doing, working
with Karl Friston, meeting him and just your entire history?
Let's hear about it.
I started off as a young age, my grandfather and my father are both uh engineers, car
engineers at General Motors.
So I grew up in an engineering family, but obviously in my generation, it was obvious that
uh the computer was the thing to focus on, not the car.
And so I looked into uh computer engineering and really studied that in undergraduate and
graduate.
degrees.
then of course, you know, you're growing up with science fiction movies and, you know,
Star Wars and Star Trek, and you've got these artificial intelligence programs like the
HAL 9000 and data and now Jarvis and all these things.
So obviously, the pinnacle of computer technology would be AI.
And so that led me to getting a fellowship to study
at Berkeley in advanced systems theory with West Churchman, one of the top people in AI at
the time.
And that was a natural evolution from undergraduate to graduate to super graduate school
and looking into the depths of how we might uh make an artificial intelligence program.
And then uh the other area that interested me a lot was virtual reality.
And so uh
it seemed like the two would go together.
So you would have virtual worlds that are kind of like the holodeck on Star Trek, then you
could have AI in the world.
So then we would have a kind of a spatial AI internet.
And so that kind of was what I had in the background of my mind.
And so that led us to found a company back in 2018 called Verses to
to really bring these technologies to the market.
Talk a little bit about the spatial web and why it's so game changing for the world and
how it ties into Verses.
when we look back from the far future to this period of time, we'll realize that the
computer was absolutely a fundamental breakthrough in 1945 that helped us crack the German
code.
And uh computing was really massive.
But it was really when we discovered how to create a network of networks, an open network
that allowed anybody in the world to plug into it without having to call Microsoft or IBM
to see if they could join.
really an open standard internet.
And that began in the 1970s and that enabled email.
And again, for 20 some years, we thought, well, what more would you ever want?
You've got a communication network now, got a global web of communications, and you've got
a personal computer on your desktop.
What more could you want?
And then Tim Berners-Lee said, I've got an idea.
We've got more bandwidth now.
Instead of just sending little text files to each other called email.
Why don't we format full pages and we can do scientific research and send our white papers
around to each other and corporations could tell their story better and other things like
that.
And so that led to a development called the World Wide Web.
And at the fundamental core of both of these breakthroughs in the original internet and
then the World Wide Web was a simple protocol that enabled any machine to be plugged into
the internet anywhere in the world and get an IP address.
And then they could send little messages to each other, that became email.
And then the next one was every page would have a URL and uh they could be all connected
through HTTP and HTML and we could have a worldwide web of pages.
And of course that's created trillions of dollars in market cap and has been amazing.
But that isn't the end of the story because that was a 2D web and yet we had 25 years of
3D computer gaming.
And we had 25 years of a 2D web.
But every year, the chips are getting faster, the web's getting faster.
And the reason we couldn't play 3D games on the web was we needed to play them on a
PlayStation or an Xbox or something that has special graphics processors to handle the
heavier information of uh a uh 3D game.
But it was obvious at some point we would have enough bandwidth and enough processing
power to have a 3D web.
Well, once you got a 3D web, wow.
Now you can do a digital twin of a building, hospital, uh thousands of buildings, an
entire city, maybe even ultimately the entire surface of the Earth, a complete perfect
digital copy of the entire surface of the Earth.
And then fantasy objects too, like Hogwarts School or the world of Star Wars or whatever,
all inside virtual worlds and all networked and connected with each other, all with AI
inside.
That's the exciting thing about the spatial web.
And so we had the lucky happenstance to uh just take on the uh same job that Tim
Berners-Lee did for the 2D web with HTTP and HTML.
And we developed the hyperspace modeling language and the hyperspace transaction protocol.
So HSTP instead of HTTP, HTTP meaning hypertext, and now this is hyperspace.
And then the modeling language instead of a markup language.
how do things work in three dimensions?
So now we can throw on a pair of glasses or headsets or whatever and we'll be able to
navigate uh inside learning environments, medical environments, your MRI scan, all kinds
of new entertainment, new kinds of shopping, other kinds of things, all will happen in
more naturally.
We have these two eyes, so we have binocular vision.
So we like to be able to look around the world in 3D.
And so now we'll be able to...
do that digitally and have it all networked just like we have a World Wide Web of
websites.
Now we're going to have a World Wide Web of web spaces.
And so that was uh a critical foundation layer.
And then you put the AI in on top of that.
And now you've got a web of spatial intelligence, if you will, all over the world.
So that's kind of the dream that we've everybody in computing has had, but we didn't know
how we were going to get there.
And but Tim kind of pointed the way with the World Wide Web.
And so then we built something called
the Spatial Web and we wrote a book about it.
It's called The Spatial Web.
It's on Amazon.
And now we just got IEEE approval last summer So now it's a new standard and it'll be
rolling out here in 2026.
how did Karl Friston's history and path and yours collide to build Verses AI?
if you look at like the current first generation of AI that we're seeing today that people
are using and OpenAI and Grok and Meta and Gemini and DeepSeek and Mistral and all the
current generative AI models, They're very centralized.
You build a giant large language model, you put it in a big data center, and then you get
to ask the
big think questions and it gives you answers, more or less accurately, a probability
level.
But it's looking into its memory to answer the questions.
what we do as humans, we are looking at the world most of the time to answer questions.
So we have smaller data models.
So we know what a house is.
We don't have to have every house in our mind just to know what a house is.
A house has a kitchen.
It's got doors and windows and things like that.
So then you can interact with almost any house.
in real time because you've got eyes that are navigating and building world models for you
as you're moving through the world.
And so we're seeing a big movement now from top scientists moving away from word models
over to world models.
so uh Yann LeCun at Meta just quit Meta and he's starting a new world model company.
uh Ilya Sutskever left to OpenAI and he's building a safe AI company.
very linked to world models and WorldLab is building world models and we're a world model
company.
so with a world model company, then you don't want a big AI sitting over there somewhere.
You want AI in all of these separate worlds, all talking to each other, just the way the
World Wide Web does.
So I can build a website and it'll link to every other website.
I can jump from my website over to Google, Amazon, Facebook, eBay, whatever.
And so...
uh
So we wanted a decentralized AI, not a centralized AI.
Because really, the internet is all about decentralization.
Because once you've got an open global standard, then anybody can build a website, or
anybody can build a spatial website.
Wouldn't it be nice if anybody could build an AI instead of having to use these big AIs
that are built by these big companies?
Then it would be more like the World Wide Web.
So Karl Friston was doing some great research at University College London, which is, the
way, where Google DeepMind came from.
But he's a neuroscientist.
So he was looking at it a little bit differently than the computer science people were
looking at.
Computer science people were looking at it, developing neural nets, which were basically
pattern recognition machines from 40 years ago, from 1986, when Gregory Hinton came up
with back propagation.
And so we've really got the ultimate end of
which you can do with a word model uh based on a neural net.
And that's what we're seeing today with OpenAI and Grok and all these things.
But as you can see, they're really expensive to build.
And one company has to build them.
And then everybody uses them.
Well, that's a little bit like aol.com versus the World Wide Web.
Once the World Wide Web came in, everybody quit using aol.com because why am I reading the
New York Times on aol.com?
I can just go to the New York Times website.
And so
So what Karl did is he was building uh a new kind of AI called active inference AI.
And you could build a specific, if I'm a cardiologist, I could take this and just build a
cardiology AI myself and then mount it and then let everybody use it the way you might use
a website.
And then somebody else could be doing a farming AI in Kenya for farming in East Africa,
which would be very different than farming in uh Japan.
So you want hyper local development of AI all over the world, the way the World Wide Web
did.
And so uh Karl's AI had that capability of building domain specific small AI's that then
because we had the protocol, then they could all talk to each other and share data just
the way a group of scientists all over the world share data back and forth, but writing
white papers and sharing information.
Now you've got a collective intelligence.
that spread all over the world.
It's not built by one company.
It's built by millions of people just the way the World Wide Web.
People have estimated that over 100 million people worked on World Wide Web websites.
There's a billion websites.
So we want to do the same thing for AI.
so layer one would be the web of communication.
That's the email layer.
The second layer is a web of information.
That's the World Wide Web layer.
Then the third layer is a web of intelligence.
Well, of course, that's the right way to do it.
Even the biggest companies only have maybe 100,000 employees and of them, maybe only
20,000 of them could work on building something like an AI for the world.
But when you give it to the whole world and invite them in, you've got 100 million people
building the AI.
Well, of course, you're going to get a way better result.
so Karl had it working in his lab and uh we talked with him and showed him what we were
doing and we all got together.
And then he became chief scientist of our company.
And uh then we hired all of his top PhDs.
And then we've spent the last five years and uh millions and millions and millions of
dollars commercializing this.
And now it's ready to go.
We went into beta testing last summer.
And uh we'll be coming out of beta here in next month or two and uh announcing our first
customers with it.
I also see it as a bigger mission in the world as what you're building is democratizing
intelligence.
And if we think about Shopify If you sell on Amazon, you don't own your customer data,
you have a Shopify site, you own that customer.
So I can imagine what you're building you could allow brands today to build not just these
2D e-commerce experiences, but spatial commerce experiences.
And then even on top of that, their own AI that they own and control and that
automatically learns as it goes.
And they don't have to sell that through the app store anymore.
this democratizes it, they could sell it directly, brands to customers.
You put your finger right on it.
You'll have the big brands, uh instead of going to gucci.com, you'll go to Gucci Space,
you'll go to Gucci World or whatever.
you could, Charlize Theron or some famous actress might come out and go, Kimberly, welcome
back.
And uh let us show you our new collection.
And then you can design the dress yourself.
Like, I want the hemlines here.
I want to use this particular fabric.
And then the robotic cutter does mass couture.
with four or five pictures of yourself sent into an AI, it'll know exactly your
measurements and the robot can cut the dress perfectly for you.
So it's mass Couture at that point, you know.
And so you can kind of see, you're gonna see what's coming.
And it's just so natural, but it's been in science fiction up until now, but now it's
about to become a reality.
So.
uh
So the potential of what it unlocks is just massive for like medicine.
We're working with City of Hope on different kinds of ways of using this for cancer
diagnostics and cancer treatments and other kinds of things like that.
But it's an endless new set of opportunities.
We can't even imagine.
There's obvious things, like you just said, immersive shopping experiences, immersive
learning experiences.
mean, you're studying physics.
You can get right down to the angstrom level and look at how the electrons are interacting
with the nucleus and things like that.
Even an eight-year-old girl can do that now.
It's like that's going to be PhD level work by a 10-year-old.
But then there's things we can't even imagine is going to happen, how it interacts with
robotics and how we get an autonomous nervous system, if you will, for our cities.
So they really start to run more effectively and deliver services more effectively at a
lower cost and things like that.
So I think we're getting a new set of tools that
humanity can apply to things like climate management, ocean health, smart cities, supply
chains, global banking, uh other kinds of things that we just have never been able to
crack the code on.
And that's why we call them world models, because they're modeling our world.
And so then the AI can be loaded in with the scientific understanding of
of how this thing should really operate at most effective levels.
And then the AI can take over and deliver that understanding autonomously at a speed that
we never could.
So I think it's a very, very hopeful and very positive potential for the next level of the
internet, if you will.
I call them intelligent cities or knowledge cities in the future.
because they really understand the people and the biosphere within it, not just connecting
things and making everything autonomous.
They're truly knowledge cities in the future.
What I think is also game changing is Gucci can sell Gucci robots now directly to
consumers.
And it would be maybe it's the designer and it's, you know, your personal
fashion assistant at home.
This isn't something that's happening right now.
this is bigger than just 3D worlds to me.
This is we're selling intelligence directly to consumers.
I can see that becoming very real.
in the next five years with what you're building.
But getting back to what you're building that's so different is, you use the term active
inference.
And I'd love to educate the audience on what that is and why it's such a radical shift in
how we think about intelligence, what makes it completely different from learning
architectures that dominate today agency and causality adaptability, reasoning, and being
able to really self-regulate itself.
This is something that
just hasn't existed before.
And we need this to get to true robotics, AGI, and super intelligence.
Am I right or wrong?
You tell me.
100 % correct, Kim.
I mean, if you interviewed anybody in 2007 and said, what's the best smartphone in the
world right now?
They'd go, well, BlackBerry, of course.
And then a little later in the year, Steve held up the iPhone and went, uh we call this
the iPhone.
And uh it's not smart, it's not a phone, it's a computer.
And so the BlackBerry was based on a
phone technology and they tried to add the internet to it.
And Steve said, no, no, no, no, no, make a laptop in the shape of a phone.
So it's got the form factor of a phone, but it's a fully powered laptop computer in there
with a different interface with a touchscreen.
And now you can have applications called apps.
And suddenly now the world blew up and it's trillions of dollars in smartphones all over
the world.
So you could argue that the first generation of AI has been these neural nets.
They've been around since, oh in pretty active form since the 1980s.
And they're probability guessing machines.
And you load them up with a lot of data.
The more data, the better.
uh And the more chip power, the better.
they look at all the words that have ever been written about, let's say, the city of
Paris.
you ask it, tell me the story of the history of Paris.
then it'll construct a uh set of tokens in order that kind of makes sense.
When you read it, it looks like it's the history of Paris.
It's just mathematics are going like the vector relationship between this word and this
word is pretty strong.
Paris and Eiffel, Paris and art, Paris and love, and things like that.
And it can construct a really cool story for you.
it doesn't see, it's not looking at Paris in real time right now.
if I ask an AI right now,
Where is my coffee cup in the house?
It has no idea, but it'll go like, well, coffee cups are usually 85 % within 20 feet of
the refrigerator or something like that.
But a five-year-old child just uses something called their eyes, which are just a pair of
cameras, sensors, and they look around really quickly and they see the coffee cup on the
table.
They go, it's over on the table by the couch.
Do you want me to bring it to you?
So they fully integrated sensor data, the AI and robotics all into one thing.
They're not separate things.
And so they go and pick up the coffee cup and bring it to you.
oh current AIs can't do that.
And uh as you pointed out earlier they hallucinate, you know, because they don't see the
world as it is.
And so what we've got with active inference is something called embodied AI.
So humans have embodied AI.
small world models in our mind.
I don't have to get everything in here.
I don't have to cram in 10 trillion words into my brain to tell you the history of Paris.
I've gone to Paris and, you know, over the years, studied different things and I can look
stuff up if I need to and put together something.
But more importantly, if I'm dealing with something like a hospital or a city or a port or
a supply chain or global banking, the data is all moving in real time.
Well, you can't have any hallucinations in there.
You don't mind a hallucination so much if you're just getting a report because you can
read it and fix the hallucinations.
That's called human in the loop.
So you've got a human reviewing it before you publish it.
But if you're dealing with data that's moving in real time and it's messy, it's real world
stuff, then you've got to be looking at it with eyes, with cameras, with satellites so you
can see the traffic in the entire city and you can see what's going on.
And if there's an accident over here.
then you route the emergency ambulance services way around that problem so they don't get
stuck in traffic and things.
So you're right.
And so once you've got real time data and you're looking at the world, then you've got
just exactly what a human being is doing.
They're making adjustments in real time.
And so the key word here, the secret sauce to the whole thing is the word called
evolution.
So we don't go into the hospital every year and get a new chip.
But with neural nets, have to go GPT-2, then you have to wait, then they have to build
GPT-3, just like a new laptop or a new phone.
They're machines.
They don't have any agency.
They don't learn.
They don't grow.
They don't evolve.
You have to build the next one, then we use that one, then you build the next one, then we
use that.
We're up to GPT-5.2 now as an example.
And so, and what happens is they start to hit a wall.
So they're spending more and more money, but they're not getting that much more
breakthrough even with each one.
Whereas if you make a smaller lighter system, then it's learning and evolving all the time
and there's no upper limit.
So it goes, it goes past G, G, G, G, three, four, five, six, seven, eight, nine, 10.
And it goes to AGI by itself, uh just like we do.
I, you know, a girl or a boy who's seven or eight years old grows up into a man or a woman
who's 25 with no interventions from the outside.
We're just grow and grow and evolve and develop and our world models get
better and smarter and everything else.
so these world models are really key because that's what every child is doing is they
figure out that they come out of out of nursing and start to toddle around the house.
They first map the whole house.
And so they can be sitting in their in their room and they can hear the mother in the
kitchen doing the dishes and they can see in their mind how to get there.
They've got it all mapped.
So they crawl out to the kitchen and let her know that they're they want some milk or
they're thirsty or whatever.
Then they start toddling, they go out into the outside, into the yard.
When they get their bike, now they're mapping the whole neighborhood and things like that.
And not just physical things, but psychological things.
If you say this, people will react this way.
So the world model of a baby is growing a little bit all the time.
So that's what Karl Friston understood as a neuroscientist.
Oh, these world models are something that we're doing and we're running almost scientific
tests against our world model.
all the time against the real world and then updating the world model where it maybe isn't
correct.
So a young child might think, I know what a cup is.
You pick it up, it's got milk in it and you can hold it and you let it go, it'll crash to
the ground.
And one day they reach over and grab mom's cup and it's hot.
Well, that's a new piece of data.
Well, you don't have to wait for it to be reprogrammed and go like, well, some cups can be
hot.
They have coffee or tea in them.
No, they find out directly just by engaging with the world and that's a surprise.
So that's really what active inference AI works.
It loves surprises because then it can like figure out is my world model wrong or is the
world wrong?
know, like maybe some city planners decide we're to make every other street one way.
And then we do it actually, the traffic still gets clogged up.
And when you actually run a real test on it, should be every third street is one way.
And so you can either change the world to match the world model if you would know what you
want, or you change, generally you change your world model to give you better data on what
really is.
And so this is what active inference really is.
It's the first evolutionary
AI so it's more like a digital organism and it has access to sensor data just like we use
hearing and touch and sight and things like that to build our world models we can look
through satellite data and car data and phone data and things like that to update its
world model and we see it a little bit with with mapping for traffic you look on your
smartphone and it shows you it's red there well it's just tracking the cell towers and
noticing that they
They're moving really slow in that section of the road, so get to go around it, you know.
But now we've got something like that for the whole thing, you know.
And so now we've got an evolutionary model that we can load in our really understanding.
So ultimately, it's going beyond anything we might have ever discovered through our own
science after a while.
my own guess is we're probably, uh we're on a uh clear track to AGI in the 2030s, So
that's real exciting because that is a historic marker for humanity to be able to fully
replicate its intelligence.
digitally Once that happens, then wow, medicine, everything else has huge breakthroughs.
So it's an exciting moment.
Karl Friston certainly deserves tremendous credit for having discovered this active
inference AI which is based on something that he calls the free energy principle.
But now many scientists all over the world are starting to uh study this field.
There's an active inference institute in Belgium now, and there's active inference
gatherings.
And so it really is an exciting next generation of AI that's coming in.
And luckily, we've been working with Karl to commercialize it.
So we got it ready just as everybody's kind of realizing the limitations of the LLMs and
kind of going, what's next?
And we're going, well, here it is.
We've been over here in the corner building this thing for the last five, six years, and
now it's ready.
so it's like a baby that learns real time.
It's one of the first models in the world that actually self-regulates and continuously
learns in real time, which actually makes robotics real.
Because today, even with all the training, robots are pre-trained.
Even to kick and dance and do all of those things, But they don't have the ability yet to
continuously learn in real time without
having to spend more money training it.
And this gives it a set of principles, and trains it on how to look at the world in its
own context.
you can embody it with
a technology purpose or a set of values that it then can go into the world and be more
responsible as a robot.
You got it exactly right.
There's a test for robots called the Habitat Test.
And as you pointed out, most robots, in fact, all robots up until active inference, are
pre-trained.
But the problem with pre-training is even with a billion and a half pre-training sets, the
world is way messier.
The world is just crazy.
mean, a pigeon can land on your robot.
It's never seen that before in its pre-training.
And what's it do?
mean, Waymos modes will stop in the middle of an intersection, let people off and weird
things because some light is reflecting off a puddle or something.
And so, so the problem with pre-training is you can't imagine everything that you're going
to encounter.
So it's better if you have less pre-training and more intelligence and reasoning.
And so then you can give it goal states.
and let the AI figure out the best way to solve the problem.
And so when we did the Habitat test, the best robot up until that moment was trained on
1.3 billion training instances and it scored, say, a level 50.
And so we brought our system in to do the test with no pre-training, just the very basic
things of what a house is or what a table is or different kinds of basic...
concepts and then gave it the goals and then it solved the puzzle at a level 70 with no
pre-training.
And so that means there's no cost and that means it can learn and evolve and grow.
So it could look at the problem and realize it couldn't solve it the obvious way.
It had to go around and go around some obstacles and then solve it a different way.
And it figured it out itself.
And so that's really what we want from AI.
So this world of active inference AI is a world of
of AIs that can reason, they can learn, they can adapt, they can plan, they have agency.
So the problem with LLMs is
if you don't have anybody in there, and there's nobody in there, it's just a bunch of
math, then you don't really have agency.
To have agency, you have to have an agent, and an agent has to be able to plan, adapt,
learn these kinds of things.
So uh Apple kind of wrote a big white paper a year and a half ago.
they said, well, one of the things we discovered is that LLMs can't reason.
Even though they use a thing called reasoning, it's really just reiterating steps.
And so doing task management isn't really agency.
If I tell you, do this, then do this, then do this, and do it in the next 45 minutes, and
you do those things exactly as I tell you, you really don't have any agency.
You're just carrying out tasks.
But if I tell you I've got to be in Abu Dhabi uh in two weeks, get me a hotel, and see if
you can find uh any cool events going on while I'm there, then you have a lot of agency as
my personal assistant.
they've been trying to convert LLMs, which are language models, into agents,
MIT pointed out that 95 % of all implementations in corporate environments for agency were
failing.
so it's because there's no ability to plan, adapt, learn, and these kinds of things.
so active inference, that's all it does.
It's constantly curious.
It's constantly probing, constantly trying, constantly learning and adapting.
So it's a path forward.
LLMs will always be around.
They'll be really great for content.
They'll make cool videos.
They'll help us write drafts of reports and things like that.
But they have their limitations when it comes to real world interaction with a messy world
that's changing and interacting with it in real time.
let's make this real for people Let's imagine that you work with Apple on Apple Watch.
So if someone falls today, Apple knows that someone falls it can send for an emergency
call.
If it's measuring your biometrics, it's based on algorithms and a model of uh broad data
across a huge population.
But with active inference, if you add that into the technology,
it actually delivers hyper personalization in the future because it understands, well you
fell, but you didn't really hurt yourself or, you didn't eat your breakfast this morning
It's not just looking at signals.
It really understands you and what I call in the future, the ambient home.
It connects you to all the surfaces in your home.
But it understands your habits, it understands how you think, it understands you as a
person, it learns along the way.
You don't have to train it, it will just learn automatically.
It'll reason and say, she fell because of this.
Apple today wouldn't know that, they just know you fell and they call.
Still beautiful, still amazing, but with this, it could be game changing.
No, you're right.
And you use the term ambient.
We call it context.
So absolutely active inference is constantly building context.
And that's how it can run a supply chain and things like that.
Because when the container leaves the port in China, it already knows the port in Long
Beach and knows exactly where the container is going to be placed when the ship arrives
and schedules the ship arrival exactly at the right time.
So it's not waiting around out in the thing for 10 days
and the same thing, as you point out, it would have the context for the fall.
But it could do then a whole lot more.
For instance, we do a lot with drones.
We've done a lot of work with drones, and particularly in Europe, the European Commission
on drone management uh in five cities over there on a big test we did called flying
forward.
it could know that you're actually having a heart attack.
It could actually send a drone right to you and call in anybody near you to come over.
And the drone lands and it's got the things that you put on your chest.
And it tells you exactly how to do it.
So you don't even have to be a doctor.
delivers it and it saves people's lives right there.
Or it could call an ambulance if it needed it.
and arrange a bed for you at the hospital while you're being picked up.
And particularly in the future, there'll be a drone ambulance, you know, it just picks you
up.
You fly right over all the traffic and right to the hospital.
But the hospital is already waiting for you and already has a bed ready because right now
it's a big problem with bed management and people coming in.
the whole thing is like a giant computer game.
And so one of the big areas that AI kind of comes out of is research in something called
game theory.
John von Neumann was kind of the
the big kind of main, he was kind of the Karl Friston of game theory.
Now we've applied game theory to evolution.
It's called evolutionary game theory.
Well, guess what?
uh The math of evolutionary game theory is active inference.
So it's going to be a very accurate, a positive game that we're all playing.
And
It's interacting with you and helping you design your life in a way that helps you achieve
your visions and goals.
And I think it's exciting to see it.
So the world is moving toward autonomous systems.
And Gartner just came out with a big report all autonomous systems
in 2035, which is only nine years away now, all autonomous systems are going to be running
on the spatial web.
And they'll all be something called spatial aware AI.
Well, so we won't be the only company doing that.
There'll be competitors by then, but we'll be one of the players.
So we're pointing the way toward this future
as the limitations of the LLMs are more fully understood.
companies are trying to build agentic AI systems, but it's really just automation and
workflows and things connecting to each other.
But with active inference, the AIs can actually, start learning from each other.
That's not happening today.
those multi-agent systems need to actually be communicating better and then learning from
each other on their own
Well, it's true.
So the other problem is the data is not curated.
They're straight scraping everything from the internet and throwing it in every Reddit
file and every cura file and every magazine article and everything.
Well, there's a lot of wrong data in there.
Well, if you find if I'm the Cleveland Clinic and I'm building a cardiology AI, I well
then I'm curating the highest quality data going in there.
And so now I can trust it.
You know, it's from the Cleveland Clinic is probably well reviewed
And it's encrypted and can't be hacked.
I can charge for it.
I'm a cardiologist in Kuala Lumpur and I've got a strange case.
I can query the Cleveland Clinic AI and it can help me diagnose the case and help me
prescribe the right medicine or therapies for the patient.
And the Cleveland Clinic can earn an income from that, just like a World Wide Web.
So you can see that's a much better model to have small curated models.
by specialists building them all over the world.
And then they're probably ranked and evaluated with some kind of a Yelp score by their
quality of their usage and how people like them.
just like Google right now kind of page ranks
there'll be multiple cardiology AIs, but maybe the Cleveland Clinic's cardiology AI will
be the best.
But maybe there's a specialist group that deals with infant cardiology, and they have an
AI that really specializes in cardiology for children from birth to two or something.
And well, for that, you might want to go to that AI.
And uh just like Google kind of helps you find the websites your personal assistant will
help you identify the AI that would be most helpful to you.
Or if you're a doctor, the AI that would be most helpful to the doctor
just like today, you're responsible for your website.
If you put out bad information, the world sees it really quickly and you kind of get
noted, hey, they're putting out bad data there or they have a big agenda or whatever.
it's also learning from other AI and it's also learning from its interactions with the
world.
And then it's fully auditable because it's a small
curated data model.
if it does make a bad analysis of a cardiology situation, we can analyze it.
How did you arrive at that?
we can see exactly the thought process it went through and the research that it
referenced.
And that research is incorrect.
It's been beaten by a new level of research that just came out and that hadn't yet been
loaded in, or maybe it hadn't found it.
So that auditability issue is really critical
where we've got mission critical real-time actions,
I think this is going to have huge implications for branding in the future as well,
because brands are going to be judged not just on their experience or their leadership
team or just their products in the future.
They're going to be judged on their level of intelligence.
they're going to be part of that web of intelligence.
So you won't just have brand equity scores.
You're going to have brand intelligence scores.
So you really have to think about the type of
AI that you build because like you said, Cleveland Clinic, has equity, it has credibility.
And so if it puts out its own AI, which then it owns, right?
It's not relying on a closed system.
And then if that closed system goes out of business someday or sells itself, then what are
they going to do?
it gets scored.
So now every brand has an intelligence score.
That's the future of brand equity and brand credibility.
It's going to be really interesting to see all of this.
you're right.
Intelligence is the ultimate currency.
It's even how we hire people and measure them in real life.
so uh intelligence isn't just a static number.
It's your ability to learn on an ongoing basis.
An intelligence system is
a system that's learning through experience.
so when you hire somebody, you don't just hire them for what they know.
You hire them to be able to solve problems.
know, a chess player doesn't just play past chess games.
They have to play this moment.
you know, so it's so the ability to learn in real time and evolve your understanding
That's going to be your right for every brand and every company and every city.
That's really the way forward.
The AIs that you're going to build in the future are going to want to belong to this
bigger network of things with belief systems and values built in.
So you can build it responsibly from the beginning.
It's the values that you train it with.
I think that's well said.
In fact, Karl, Karl Friston was on a panel with Yann LeCun a year and a half ago at Davos.
Well, we've got two huge stars of uh AI here tonight on the panel.
Why don't we start off with a simple question?
How would you define artificial intelligence?
And Karl.
Karl goes, actually, I don't know too much about artificial intelligence.
You need to ask Yann about that.
I specialize in natural intelligence.
I'm a neuroscientist, so I've been studying how brains function and how animals learn and
that kind of thing all my life.
So really what we've been coding up is just a digitization of what humans are currently
doing.
And so uh it was kind of a fun way to start the evening, natural intelligence versus
artificial intelligence.
I don't think LLMs are going anywhere.
I think they're incredibly efficient machines right now in terms of content and being a
thinking partner.
it's going to be good enough for a really long time to come.
Absolutely.
Yep.
But I do believe that world models is going to be the buzzword this year.
And so there's different companies with great intention.
You're one of them.
the way that these world models are going to be trained are going to be different from
Verses each one is kind of approaching it differently.
you sit in a really unique area.
I think about Google DeepMind Genie, which is really about the spatial web, it's kind of
like the baby with the imagination.
It's the gamer baby.
they use
large scale video data sets, simulated environments, and gameplay sequences used for
unsupervised prediction, and they build out these spatial environments.
But they're building world models and sometimes optimizing that with a reward system.
So different from what you're doing.
And then LeCun's new company, Ami, uses JEPA
it's a different type of world model trained like a student with curated data, video,
audio, all sorts of different inputs, but without rewards.
learning by predicting the structure of sensory input.
Integral AI is another interesting company to me out in Japan now.
this is a company that's training
for robotics and training models like philosophers by teaching them explicit concepts and
real deep reasoning systems.
again, it's not self-regulating but it's trying to get closer to true reasoning.
You're the baby that operates in real life and you're the only one right now that does
that.
So everyone has these great intentions of world models so that we can get to true agency
and super intelligence, but you're the baby that lives live
Yeah, we're doing the messy, the messy real world stuff.
And I tell people, I use a ChatGPT buti and Grok and everything daily, just like everybody
else.
They're awesome tools.
But we do the messy stuff, but that turns out to be 90 % of what we want AI to do in the
world,
So Gardner says there'll probably be five companies that are market leaders by 2035 in
this new area.
So we'll probably be one of them
But what we really the thing that we're excited about is uh really just having a naturally
evolving natural intelligence at the core of what we're doing
they're very lightweight.
They can run on a smartphone.
They can run on a laptop.
They can run on a bigger system, one of the cloud systems, Azure or whatever.
But AWS and all those things.
So those things already exist.
So we don't have to build new things.
So we don't have big capital risks.
We don't have to raise billions of dollars to build data centers for this stuff to go.
And then even in the spatial area, there's so many breakthroughs happening right now in
the geospatial area.
There's a thing called Gaussian splatting.
the robot is looking around the world and building the world in 3D in its mind in real
time.
And that's called a Gaussian splat.
So it's just taking the data in and turning it into a 3D world in real time while it's
walking around.
And that way it can solve problems and everything.
Doesn't count on something that you've already built a 3D model somewhere.
It's just building it as it goes, just like we do.
So those are all very lightweight things compared to the alternative.
in terms of the data centers that are being built, they're still going to be needed in the
future for other applications like graphics and 3D worlds, you know, and heavier building,
right?
Elon Musk has a really great vision for building them in space and using solar power.
as a futurist, I think it's a more sustainable vision than
building them in small towns and taking up energy and water and all of those things.
What other applications do you think these data centers will use in the future?
So I think you're right.
think data is the new oil.
I mean, we're going to have data.
Data is important, but we don't have to have a company that has to build a data center to
now deliver AI.
No, no, no.
The data centers of the future are just extensions of the existing cloud.
so yeah, we're going to need more of those there were 60 companies at CES this year, all
showing augmented reality glasses.
And so you can see what's coming in 27, 28, 29.
People are going to be wearing glasses and you're getting personal guides and you have a
digital nurse from the hospital coming to your house to remind you how to do your
exercises and take your medicines on the right time.
So we're going to have graphics delivered to the eye instead of looking at it on a phone.
You just wear glasses.
and have a voice thing and you're good to go.
And then that opens up huge numbers of services.
You could have a holodeck in your living room and go back in history and walk through
things or throw on a pair of AR glasses when you're going through a castle in England and
everything's
themed exactly the way it was in 1450 or something.
Or you put on a pair of glasses and you want to see everything kind of like Hogwarts
school.
so people are dressed, you've got the glasses on, people are dressed in Hogwarts school
style and all that kind of thing.
So you've got to see what's coming and there will be a lot of graphics.
I think you're right.
think the cloud will just continue to grow.
they may be used, as you point out, for other things in the future.
New kinds of education.
You're going to have lifelong education for everybody, and not just about
how to get a job, but how to become a better person.
If you look at Maslow's hierarchy, it's a pyramid.
And at the bottom of the pyramid is more survival skills.
And then in the middle is there are kind of the community skills, families and communities
and things like that.
In the upper level, it's really a self-actualization and arts and sciences and
philosophical concepts.
At the very top is enlightenment, self-
Self-realization, self-transcendence.
And so it may be that uh as we solve the abundance problem at the bottom and get rid of
the scarcity issues that more and more people move up into the area where the Beethoven's
and Mother Teresa's and Dalai Lamas are hanging out.
And that becomes a more normal thing.
Instead of getting a better job, you become a better person, you know?
But maybe in the future, as we dial down the focus on scarcity and move toward more of an
abundance economy, that more and more people start to look for ways to become happier and
more centered and more self-aware.
And once we have a civilization based on intelligence, it might be a civilization that
then moves toward personal development, something like that.
Well, it's funny that you say that.
A couple of things came to mind for me.
If you think about the eye care industry and lenses, we still have glasses, but a lot of
people moved to contact lenses.
I don't think that we're going to be walking around with goggles and glasses in the
future.
I think it's going to be lenses.
I truly believe that.
Secondly, you mentioned that we've moved into the fifth intelligent revolution, or you
were speaking to that.
I'm now calling it the fifth abundance revolution because I don't even know that it's just
intelligent or industrial anymore.
It needs to give back to society else it will collapse on itself.
yeah.
mean, take a look at the big five, uh you know, housing, food, uh medicine, uh education,
transportation, uh clothing, these kinds of things.
mean, gosh, AI and robotics are going to drive the cost to zero and all those things.
mean, there's 3D house printing companies right now that can print a house in 40 hours and
sell it for like $5,000.
um
So you can kind of see if we go out to 2035, 2045, mean, pretty much everything's going
almost to free because the costs of running robots right now are under $100,000 and some
of them are under $10,000.
So we go out five or 10 years, mean, robots are almost free.
And so the cost to make anything would just be 10 cents an hour, the cost of electricity.
if we're generally cutting the cost to 10 % of what it is in the next, say, 10 or 15, 20
years, well then, what a huge uplift that is for everybody.
And the web is available to everybody.
mean, even these smartphones, this thing that we call a smartphone, people don't realize
it.
The processing power of this thing is greater than a Cray 2 supercomputer was in 1995.
The Cray 2 supercomputer was $35 million.
There were only 120 of them in the whole world.
And now you're a futurist and you're interviewing me in 1995.
And you go, what's 2020 going to be like?
I go, see the Cray, six feet tall, four feet wide.
fastest computer in the world, supercomputer, $35 million.
We're going to give everybody a cray.
And you would look at me like, you're crazy.
that's not all.
We're going to give everybody a cray.
There'll be like 4 billion people will have a cray.
There'll be 8 billion crays, but some people will have two or three of them because
they're so cheap.
We're going to be able to make a cray for under $100 in 2020.
Oh, and then,
We're going to take every document that's ever been made and put it online.
And so the complete library of human knowledge available to everybody all over the world
through this device.
that's impossible.
That won't happen.
And then we're to take every product that's ever been made and put it online at wholesale
cost and deliver it to your house overnight.
Yeah.
Okay.
Well, these are crazy ideas.
Well, clearly you'd have no idea what you're talking about.
You're some crazy futurist.
But that'll never happen.
Of course, we just did all those so easily we didn't even notice it.
It just naturally came with once you release the protocol, everybody builds on top of the
protocol and all those things happen.
Well, now we're releasing a new protocol, HSTP and HSML, the Spatial Web Protocols.
So get out of the way.
Young people all over the world go, wait, can just uh I can take World Labs and build a 3D
world and I can mount it on there and I can.
create a new education experience, I can create a new medical experience, I can do a new
kind of store, I can do a new kind of game, I can do whatever at almost no cost because
I'm using all these AI tools and vibe coding and building my game and everything and
suddenly I've got access to four billion people all over the world that can access it.
Well, get out of the way.
I mean, that's what's going to happen.
So we're probably going toward mass abundance They'll put the glasses on or put the
contact lenses in and suddenly, wow, you need a house?
Well, we're building.
houses out here in Joshua Tree.
We're building a new community and you're welcome to come.
The mortgage is $100 a month and your universal basic income is $2,000 a month.
That's the equivalent of $200,000 a year universal basic income in today's world.
But because the price of things are dropping, $2,000 a month would be like $20,000 a month
in today's market.
I think we are moving toward an abundance economy and uh you got robots just and AI
building things and of course AI and uh quantum computing uh will just solve all the
medical issues and we're get tremendous life extension easily right now where average
lifespan is in the 80 year kind of timeframe.
uh Some people are living as long as 120
I'm a techno optimist like you are.
And I'm hoping that people have some takeaways today like, there might be a new business
idea in here.
I can imagine in the next few years with the Spatial Web, you don't have a sub stack,
you're inviting people to your world.
one of the predictions that I made in 2020 in Forbes was that I was asked what the future
customer journey would be.
And I said, B to R to C, business to robot to consumer.
It's now 2026.
And we're seeing that.
People are shopping through their AIs.
You need to market to the machine.
a portion of the ads today that are being served are being served to robots.
We saw the first social media network of agents to agents, just for agents.
But the other prediction that I have and it's another model that I think is going to
become even more valuable in the future, the H2H, human to human.
So the companies that are real true human to human and bringing people back together, just
like we do when we meet
every month in our AI think tank meetings, those spaces are gonna become even more
valuable and important in the future.
No, you're right.
And I think as the the scarcity issue gets solved more, it frees up the human.
So as we move toward a more of an abundance economy.
those things lessen and the humanity comes out more in people.
When you're in a state of anxiety, you know, you're kind of looking at the world through
fearful lenses But I do appreciate the work you've been doing.
uh
as a futurist and helping people kind of see over the horizon.
And it's really been a kind of a pleasure to brainstorm with you today on some of the ways
this thing could play out because we get really close to the thing as we're building it.
Sometimes we don't even ourselves get a chance to really pull back and look at the
implications of what we're doing are.
So I really appreciate this conversation.
So this is a really big week for your company.
you're launching something exciting.
So I'd love for you to share that with our audience and let them get a sneak peek into
what's going on
Thanks for asking.
so what we're doing is we're creating a general intelligence that could be used for
anything, right?
It can run a port, it can run a hospital ICU, it could do banking, it can do whatever.
uh But you have to start somewhere.
So what's our first product?
So the first product that we've been working on has been uh an area of financial
management uh that uh AI really has a unique ability to solve problems in.
And that is uh
portfolio management.
So if I'm a large fund, let's say I'm big pension fund and I've got like 500 billion or a
trillion dollars under management and these are large funds in Japan, Canada, California,
New York.
that manage these pension funds or sovereign wealth funds or whatever.
I've got to be making intelligent decisions on investing the funds resources so that the
fund doesn't collapse but grows and provides this
pension service or return to investors that you went for it.
So we've got tools that have been studied at universities for the last 50 years on how you
do portfolio management.
And all these companies are applying those tools.
And so we were approached by one of the large funds.
we've been testing with them for the last nine months.
And using our AI, they're outperforming their best models by 40%.
so now we're signing our first customer, and it'll be a multimillion dollar contract.
But of course, that's just one customer of 1,000 potential customers.
And then that's just one product of 1,000 potential products.
So you can kind of see the potential of the business
But ultimately, we want to get out of the product making business.
We just want to give you the tools and let you do a cardiology AI, or you do a farming AI,
or you do an education AI.
Kind like the app store, if you will, or a website.
And then now the web is growing of uh these AIs all over the world.
And we kickstarted it.
Apple had to build the first 10 apps to show people what an app was.
So when you got the iPhone, it had 10 apps on it, messaging and email and maps and things
like that.
But then later, Uber and Airbnb and all these other people came up with these crazy wild
ideas.
And it all took off.
And now I there's 5 million apps.
And so same thing will happen with the AI.
We're going to build.
the first few to kind of get the ball rolling.
And then, you know, the people realize, there's this new thing, we can download the tools
from uh Verses and build our own thing.
So for the next year or so, we'll probably be primarily with our hands in the mix with
with our partners, but from 27 on, it probably just takes off and they don't really need
us to build anything.
They're just checking in with us and we're listening to them like, you guys could add this
feature, we could do this.
And then we're improving the fundamental active inference model.
That's really exciting.
Congratulations.
I've met some of your team members.
They're so great humans, so impressive.
And that's what I adore about you.
You really live your life through purpose and try to maintain flow and you have great
intentions for the world,
really appreciate everything that I've learned from you today It's always interesting when
we talk.
I feel the same.
Thanks a lot, Kim, for having me today and great conversation and more to come.
If you're enjoying the show and appreciate what we do, the simplest way you can support us
is by hitting the subscribe button and sharing this episode with others.
And by doing this, our commitment to you is we'll do our very best to make each and every
episode better than the last.
See you in the future.