---
author: TED
date: '2026-01-01'
guest: ''
layout: post.njk
source: https://www.youtube.com/watch?v=0R_CJjGRX7o
speaker: TED
tags:
- economy-ecology
- regenerative-agriculture
- resource-recycling
- ecological-restoration
- ai-robotics
title: 重塑经济与生态关系：AI驱动的自然修复革命
summary: 演讲者Tom Chi揭示了一个深刻的悖论：人们普遍热爱自然，但集体行为却导致了大规模的生态破坏。他认为，这种矛盾源于将经济与生态视为对立面的“破碎心智模型”。通过物理学和AI的视角，他论证了经济实为生态系统的子集，一切经济活动均源于自然界的“开采或生长”。为解决此问题，演讲者提出三大技术驱动的解决方案：一是通过先进回收技术实现资源闭环；二是发展AI辅助的再生农业，提升土壤健康与作物适应性；三是利用机器人技术进行大规模生态修复，从监测到主动种植。最终目标是利用AI和机器人技术，将经济活动从对自然的破坏转变为有意识的修复，构建一个健康星球与健康经济的共生未来。
insight: ''
draft: true
series: ''
category: ai-application
area: tech-engineering
project: []
people: []
companies_orgs:
- Chloris Geospatial
- Reefgen
products_models: []
media_books: []
status: evergreen
---
### 自然悖论与经济生态的误区
在过去十年里，我一直在致力于解决一个深刻的悖论。简单来说，如果你随机与一百个人交谈，问他们对大自然的感受，你会得到极其积极的回答，比如“大自然令人振奋”、“大自然是最美好的事物”。这种喜爱体现在方方面面，人们将自然景象设为桌面或手机壁纸。然而，尽管每个人都对自然怀有积极情感，但作为一个集体文明，我们却在以行星规模摧毁自然。这就是悖论所在：一个热爱自然的个体群体，如何创造出一个对自然进行行星级别攻击的工业化经济体？我认为，这源于我们无意识中采纳了一个“破碎的心智模型”。
这个被我们意外采纳的心智模型，将经济的增长与生态的健康对立起来。它认为，要想获得经济上的成功，就必须牺牲生态环境；反之，如果想保护生态，就必须牺牲经济利益。这就像一个跷跷板，根据我们对经济或生态的重视程度，天平会向一端倾斜。虽然这是一种可以采取的心理立场，但它在物理现实中并不完全准确。我是一名受过严格物理学训练的人，我思考的是物理上的真实。
Original English
So over the last decade,
I've been working on resolving
a profound paradox.
And, simply stated,
it's that if you go and talk
with any person, actually --
you could go grab 100 random
people off the street --
and you were to go ask them
how do they feel about nature,
you're going to end up
with extremely positive answers,
ranging from, you know,
"nature's inspiring,"
"nature is the most beautiful
thing that exists."
And you'll see it everywhere.
They'll put it as the backdrop
of their desktop and their phones, like,
every little thing, absolutely.
Every single person you ask,
you're going to get, like,
a pretty positive response.
And even though, if you ask all these
individuals how they feel about nature,
they're really positive about it,
somehow, as a collective civilization,
we've come together and we are destroying
nature at a planetary scale.
And therein lies the paradox.
How did this happen?
How did a group of individuals
that all love nature
somehow create a civilization,
create an industrial
economy that is out there,
effectively planetary
level assault on nature?
Well, I think it actually stems
from a broken mental model
that we've kind of unconsciously adopted.
And I'd like to start
dissolving that right now.
Now, you can see the mental model
with the X below it,
and that is the one
that I would like to say
that we've accidentally adopted.
And that's a mental model
that whenever you get economic wins,
they are traded off against the ecology.
And if you really care about the ecology
and want to make that healthier,
unfortunately, you're going to have
to trade off economic wins, right?
It's one or the other,
kind of a balancing act.
And depending on how much
we care about the economy,
or how much we care
about the ecology at any given time,
the pendulum swings one way,
the scale swings one way,
or it swings the other.
And I'm going to tell you that, sure,
that is a psychological position
that you can take,
but it's actually not one
that is particularly physically true.
Original English
I'm formally trained as a physicist.
I do think about what is physically true,
and what is actually much more true
about the economy
is the economy is not versus ecology.
The economy is a subset of the ecology.
And maybe this is a new idea to you guys,
but I can prove it to you very quickly.
Actually, you could prove it to yourself,
even just with the clothes on your body
or the things immediately around you.
Because if you think about and look
at everything the economy has produced,
everything in the economy
is either mined or grown,
which means it comes
directly from nature, no abstractions.
You guys might be wearing
some cotton, obviously grown.
You'll be sitting on a chair
that's got some metal underneath --
that was mined.
Everything in the economy
is mined or grown, full stop.
And some of you guys might be thinking,
"Aren't we moving to a digital economy,
a virtual economy?"
Well, not really.
Like, every line of code that is ever run
runs on a substrate
that was mined or grown.
Every single service you've ever used
is using server architectures
that are mined or grown.
When I say the entire economy
is mined or grown,
I mean it literally.
There's literally nothing
that doesn't come directly from nature.
And to the extent
that you damage the ecology,
you actually start to create
problems for the economy.
And this is what
we're experiencing right now.
And if you think about it
in this balancing-act-type way,
you're going to miss the right way
to actually fix these problems.
Original English
Now, let's talk about exactly how much
we are mining and growing.
This graph actually
kind of tracks 50 years
of how much extraction we've been taking
every single year from the planet.
At this point in history,
it's over 90 billion tonnes per year.
It comes out to about 11.5 tonnes
per person per year.
And if you guys feel
like you don't do that much,
well, I'll shock you to say
that 11.5 is the average in Asia,
but Europe is about two x that,
and America is three x that.
Yay.
There's two billion people
that live on less than five dollars a day
doing substantially less than that --
that's why it all balances out.
But this is exactly how much we're mining
and exactly how much we're growing.
Now, in the process of mining
and growing this much,
and using it to power
everything in the economy --
because, like I said, literally everything
in the economy is mined or grown --
we've been using really old
industrial ideas and industrial processes.
Most of how we're growing today
was invented about 50 years ago.
Most of the ways that we've been
mining, refining metals,
all that sort of thing, were invented
about 100 years ago, 150 years ago.
These are not technologies
that we've updated recently.
And with the arrival
of new robotic and AI tools,
I think it's the right time
to go ask new questions
about whether we could be mining
and growing differently,
in a way that starts to honor this idea
that the economy
is a subset of the ecology.
Now, this is where
it overlaps into my world,
because my entire career
has been built off of doing new inventions
and robotics, artificial intelligence,
advanced algorithms.
And I've shipped everything
from Microsoft Office --
sorry about that --
(Laughter)
you know, to web search --
I think that one was fine --
to self-driving cars.
So I've worked on relatively
sophisticated things,
and given that, I have
an interesting background, perhaps,
to be able to go look at these problems
and see if we can take
a different swing at them.
And I'm going to share
a number of examples with you today.
Now, these examples
fall into three major shifts --
remember, everything's
mining or growing --
and they're represented
by these three images here.
So on the left-hand side,
we have a bunch of mined materials,
and what we need to be doing
is we need to figure out
more and more ecological ways
to be able to go mine materials
and get the most of the ores
that we extract
so we do the least disturbance of earth
and watersheds in the process of mining.
In addition,
what is even better than mining
more ecologically and mining less
is to not mine at all.
And to the extent that we're able
to go close the loop
through really skillful
mechanical or chemical recycling,
we can have a larger and larger
proportion of the feedstock for industry
move over to closed-loop materials,
as opposed to virginally
extracted materials.
The second major shift has to do
with the way that we grow.
A lot of the way that we've been growing
currently is very unsustainable.
It basically is damaging soil function.
And little by little, we've been
kind of wearing down topsoil
in agricultural lands
all across the world.
And Gabe Brown is pictured here.
He's a friend who has
taught me a huge amount
about regenerative agriculture,
and I've really learned from him
that if you invest in soil function,
you can actually make it
easier to grow, cheaper to grow,
a higher margin to grow
every single year,
and do so in a way
that is regenerating soil function,
giving more services to biodiversity,
and even healing the hydrological
function of those soils.
And lastly, we need to be thinking
about large-scale repair,
because we've been
at the Industrial Revolution
for a couple hundred years now,
and there's a lot of landscapes
that we've heavily degraded.
And if we are serious about the task
of renewing the ecology
in order to support a vibrant
economy going forward,
then we're going to need better tools
for large-scale repair.
So let's jump into all three of these.
Original English
So what's pictured here is a company
I have the privilege to work with.
And they are a great example of moving
closer to that closed-loop world.
So what you’re seeing here
is actually an image from inside
the largest lithium NMC battery
recycling plant in North America.
And this is an advanced form
of chemical recycling
that is able to go bring
all of these used battery materials,
because most lithium batteries
kind of have a 10-year life;
they don't go too many years beyond that.
And after that's the case,
you know, you can't use it in the car,
or you can't use it in the consumer
electronics device anymore.
You want to be able
to recover those materials.
The process that they do here
is about two times cheaper
than the next closest process,
and is able to return the material
to complete virgin quality.
It's better than the stuff
that you would have been able to mine
out of the ground in the first place.
And if we get really skillful
about closing these loops,
what's great is the car battery
doesn't just evaporate and disappear.
It's a relatively large object
that you can go handle,
and you can do a reverse
logistics supply chain
and pull these things together.
And whether it's robotics and AI
to do advanced mechanical recycling --
or, in this case,
advanced chemical recycling --
then there are really skillful ways,
with our new technologies,
to be able to go close the loop
and make it so that a higher
and higher fraction
comes from a post-consumer
or post-industrial waste stream,
as opposed to from the ground.
Now, moving over
into the regenerative growing side,
we're actually at a really
compelling point in history,
because there’s a mini renaissance
in regenerative agriculture
that's happening right now,
with different farmers around the world
discovering the benefits
of agroforestry, intercropping,
you know, no-till agriculture
and a lot of other practices
that really help to establish
healthy soil function
and a healthy soil microbiome.
And what you're actually seeing
on the left-hand side
is a new technology,
which also uses machine learning
to be able to go disambiguate.
It's a new form of Raman spectroscopy
that allows folks to be able to listen
to the soils in this really skillful way.
So effectively, they're able
to go and measure
all these compelling compounds
from the soil,
so they're able
to have the soil speak to them
in ways that the soil
can basically tell them,
“Hey, here’s the next couple of things
that you should do to make me healthier.”
Instead of it kind of being a black box
that needs to get interpreted,
now farmers can have
a direct relationship with their soils
and be really skillful in the management
toward greater and greater health,
fewer inputs and higher
margins every single year.
Moving on into other ways
that artificial intelligence
and machine learning
might be really useful for agriculture,
what you see pictured here
is the development of corn or maize.
And it was an Indigenous project
that happened over the course
of hundreds of years.
And they started with, basically,
an inedible bit of grass,
because corn is actually a type of grass.
And by selective breeding
over generations and generations,
they went through lots
of different varieties,
until we got to the "lots of calories
per grow cycle" version of corn
that is now feeding a huge percentage
of the calories around the world.
Now, this was an Indigenous activity
that happened over hundreds of years,
and I'm really thankful for it,
because most of the foods
that we eat today were selectively bred
to be as large and healthy
and nutritious as we experience them.
But using artificial intelligence
and machine learning,
we've been working with a company
that has been able
to rapidly speed up this process,
and not through genetic modification.
What they do is they're able
to take the sequence information
from all the existing commercial crops,
plus a bunch of native varietals
that are not in current circulation,
and work out what the different
gene functions do,
and then map out exactly
the crossbreeding pathway
in order to go get the desired traits.
So what we have here
is adaptive sugar cane,
which dramatically reduces
the amount of deforestation
required to get to the yield
level that you want.
You also have heat-resistant tomatoes
that are able to grow
in way hotter, way drier conditions,
which is really important,
because we're going to go
through at least a 50-year period
where we're going to be destabilizing
a lot of the farmlands of the Earth
as the climate destabilizes,
whether that's hotter
or colder, wetter, drier.
It's all going to happen,
and being able to have seed stock
that is ready for that challenge
is really powerful.
And lastly, a cotton that basically
is drought-tolerant as well,
requires a fraction of the water,
one-tenth the water,
and much less pesticide
and fertilizer input.
And all of these things
are fantastic for the planet,
but they're also fantastic
for the future of us
having viable food and materials
in a destabilized, growing environment.
Original English
So let's jump into all three of these.
So what's pictured here is a company
I have the privilege to work with.
And they are a great example of moving
closer to that closed-loop world.
So what you’re seeing here
is actually an image from inside
the largest lithium NMC battery
recycling plant in North America.
And this is an advanced form
of chemical recycling
that is able to go bring
all of these used battery materials,
because most lithium batteries
kind of have a 10-year life;
they don't go too many years beyond that.
And after that's the case,
you know, you can't use it in the car,
or you can't use it in the consumer
electronics device anymore.
You want to be able
to recover those materials.
The process that they do here
is about two times cheaper
than the next closest process,
and is able to return the material
to complete virgin quality.
It's better than the stuff
that you would have been able to mine
out of the ground in the first place.
And if we get really skillful
about closing these loops,
what's great is the car battery
doesn't just evaporate and disappear.
It's a relatively large object
that you can go handle,
and you can do a reverse
logistics supply chain
and pull these things together.
And whether it's robotics and AI
to do advanced mechanical recycling --
or, in this case,
advanced chemical recycling --
then there are really skillful ways,
with our new technologies,
to be able to go close the loop
and make it so that a higher
and higher fraction
comes from a post-consumer
or post-industrial waste stream,
as opposed to from the ground.
Now, moving over
into the regenerative growing side,
we're actually at a really
compelling point in history,
because there’s a mini renaissance
in regenerative agriculture
that's happening right now,
with different farmers around the world
discovering the benefits
of agroforestry, intercropping,
you know, no-till agriculture
and a lot of other practices
that really help to establish
healthy soil function
and a healthy soil microbiome.
And what you're actually seeing
on the left-hand side
is a new technology,
which also uses machine learning
to be able to go disambiguate.
It's a new form of Raman spectroscopy
that allows folks to be able to listen
to the soils in this really skillful way.
So effectively, they're able
to go and measure
all these compelling compounds
from the soil,
so they're able
to have the soil speak to them
in ways that the soil
can basically tell them,
“Hey, here’s the next couple of things
that you should do to make me healthier.”
Instead of it kind of being a black box
that needs to get interpreted,
now farmers can have
a direct relationship with their soils
and be really skillful in the management
toward greater and greater health,
fewer inputs and higher
margins every single year.
Moving on into other ways
that artificial intelligence
and machine learning
might be really useful for agriculture,
what you see pictured here
is the development of corn or maize.
And it was an Indigenous project
that happened over the course
of hundreds of years.
And they started with, basically,
an inedible bit of grass,
because corn is actually a type of grass.
And by selective breeding
over generations and generations,
they went through lots
of different varieties,
until we got to the "lots of calories
per grow cycle" version of corn
that is now feeding a huge percentage
of the calories around the world.
Now, this was an Indigenous activity
that happened over hundreds of years,
and I'm really thankful for it,
because most of the foods
that we eat today were selectively bred
to be as large and healthy
and nutritious as we experience them.
But using artificial intelligence
and machine learning,
we've been working with a company
that has been able
to rapidly speed up this process,
and not through genetic modification.
What they do is they're able
to take the sequence information
from all the existing commercial crops,
plus a bunch of native varietals
that are not in current circulation,
and work out what the different
gene functions do,
and then map out exactly
the crossbreeding pathway
in order to go get the desired traits.
So what we have here
is adaptive sugar cane,
which dramatically reduces
the amount of deforestation
required to get to the yield
level that you want.
You also have heat-resistant tomatoes
that are able to grow
in way hotter, way drier conditions,
which is really important,
because we're going to go
through at least a 50-year period
where we're going to be destabilizing
a lot of the farmlands of the Earth
as the climate destabilizes,
whether that's hotter
or colder, wetter, drier.
It's all going to happen,
and being able to have seed stock
that is ready for that challenge
is really powerful.
And lastly, a cotton that basically
is drought-tolerant as well,
requires a fraction of the water,
one-tenth the water,
and much less pesticide
and fertilizer input.
And all of these things
are fantastic for the planet,
but they're also fantastic
for the future of us
having viable food and materials
in a destabilized, growing environment.
Lastly, let's get on
to scalable restoration.
And what you're seeing here is an image
from the company Chloris Geospatial.
And it is, of course, of the Amazon basin.
But what's really compelling
about this company
is that they've done
really deep work on sensor fusion
across a bunch of satellite feeds,
and they also paired that
with over a decade in the jungles,
meter by meter,
doing ground-truthing data,
to be able to go really verify how much
terrestrial biomass is associated
with signals that can be detected
from satellites, via remote sensing.
And given this, they've been able
to make the most accurate,
both historical and current assessment
of aboveground biomass on planet Earth.
And the data stretches all the way back
to the beginning of the 21st century,
so over 20 years of data on that front.
And that really allows us to see
which landscapes we're hurting,
which landscapes are recovering,
and if people are developing
restoration projects or carbon projects,
this is a fantastic way to go
stay on top of how those are going.
Now, this is great technology
and also uses really advanced algorithms
and allows the things
that I've been talking about.
But in some ways,
it's a little bit passive.
This doesn't restore the forest itself.
This doesn't restore the grassland itself.
This just helps people
monitor the changing of that.
But what if -- going back
to this triangle for a moment --
we were to get more
ambitious and we were to say,
"Let's challenge this linkage
between the industrial machines
that run our economy and nature,
and instead of it having to be
an accidental relationship of damage,
what would it look like
if it was an intentional relationship
of active repair?"
And I'm going to show you that right now
with my last two examples.
This one's a short video.
You're going to hear these little ticks,
and every one of those ticks
is a mangrove seed being planted.
The pace of these ticks
is basically planting about 100 mangroves
per minute from one drone.
Video: (Frequent ticks over music)
Tom Chi: And that's what it looks like
when we start planting.
And then, two months later,
you see, actually, we have
over 90 percent that get to germination.
And 14 months later, you can see
the landscape is fully established,
over 85 percent full establishing
of the mangroves that were planted.
Now, the scale of this technology
and the scale that it's capable of
is incredible.
Just four people are able to go
plant over 80 hectares of land,
representing 120,000
mangroves being planted,
and over 100,000
being established in one day.
When you get to robotic scale on things,
all of a sudden, then human action,
human intention --
and if we have good intentions,
we can really multiply that
in ways that can completely
rewrite our landscapes.
And I've worked with this company
for about a decade at this point,
and I got really
inspired by them, you know,
because they have not just
restored mangroves,
but they've restored 20 different
terrestrial ecosystems
on four different continents --
dry land, inland, mountainous, you know,
nearshore, all these sorts of things.
And I got inspired, like,
"Could we also do this below the water?"
And I'm going to show you
something that I founded
and was the original
electrical engineer for.
And this is this is the Reefgen robot,
which is basically its own kind
of planting drone.
Oh, it's already going to do
some planting things, that's fun.
And this robot line
is the first in the world
to plant live corals
back into a coral reef.
It's the first in the world
to plant live seagrasses
back into seagrass meadows.
And it can also plant them
in seed form as well.
And this robot, you know,
has been able to plant 10,000
seagrass seeds in a single day,
which covers an entire underwater acre
with one robot in one day.
And the other thing that we did
is we wanted to make sure
that this robot was affordable enough
that we could make a bunch of them, right?
Because when I went around
and I talked to people
about the underwater robot
that I was going to go build
to restore these ecosystems,
they're like, "You should budget, like,
two million dollars for the robot.
If you spent less, it's probably
not going to do anything interesting."
And I remember thinking
more like 5,000 dollars,
and we’re not quite there,
but this is more like 10,000 dollars.
And in the grand scheme of things,
it's way, way less than two million.
And the whole point is you want this
to be an accessible technology
to all the communities that have
nearshore ecosystems to restore,
whether they be coral,
whether they be seagrasses.
You want something like this
to be also scalable
from the CapEx perspective, right?
Like, a single billionaire
could spend 50 million dollars
and have a fleet of 10,000 of these,
and that is actually meaningful scale
in terms of ocean restoration
of all different types.
I'll show you a little bit more here.
So right here is a stake,
because this is actually not
a seed-planting end effector.
That's a seedling-planting end effector,
because there's actually two ways
to plant seagrasses.
You can plant it from seed,
but then, there's other
types of seagrasses
that want to be planted as a sapling.
And they want to grow rhizomatically,
so they send out
these little rhizomes laterally,
and then the grasses grow up
from the lateral rhizomes
that are heading out.
And this is basically a stake
that we put the seagrass seedlings into,
and then that feeds in
through a tube into a hopper,
and basically, bit by bit,
this current layout
is able to go and plant about half an acre
of seedlings per day
with just one robot.
And our next version of it
is going to be able to do an acre
to an acre and a half in a day, per robot.
Original English
So we are really kind of moving
into this space
where, by really digging
into that mental model in a different way,
instead of economy versus ecology,
we start taking the best tools
that we're using in the current economy,
and robotics and AI,
and intentionally
using them to support ecology
so that we're able to go
build both a healthy planet
and a healthy economy for the future.
Thanks so much.
(Cheers and applause)
Thank you.