In this episode I speak with Dr. Steve Brumby from Impact Observatory about using data and AI to help determine climate change risk and what is "provably green".
Topics:
Links:
https://www.linkedin.com/in/stevenbrumby/
https://twitter.com/stevenpbrumby
https://www.impactobservatory.com/
Welcome everyone to another episode of Prompt Pod, an open-ended
exploration into world-changing technology and my quest to document
conversations with bright minds in the space.
I'm your host, Danny Kirk, and today I'm joined by Dr. Steve Brumby. Steve is the
co-founder and CEO of Impact Observatory, a mission driven technology
company bringing AI powered geospatial monitoring to the environment,
climate and sustainability risk analysis. Impact Observatory produced the
world's first fully automated high resolution land use and land cover annual
maps using deep learning at global scale and commercial cloud,
released as a digital public good via their partners [00:01:00] at Esri Living
Atlas and Microsoft Azure Planetary Computer. Steve has built and led applied
science teams at the National Geographic Society and at the US Department of
Energy's, Los Alamos National Laboratory. He is also the technical founder of
Venture backed AI plus Space Startup Descartes Labs.
His PhD is in theoretical particle physicists, and he is the author of over 150
science publications. Steve, welcome to the show.
Steve: Thank you. It's, uh, great to.
Danny: I think we would probably need a two hour interview to cover your
entire background, all the way from degrees in particle physics, um, all the way
to Impact Observatory. But could you at least give us a bit of an overview as to
how you got where you are today considering that your experience and
knowledge is quite diverse?
Steve: Thanks. Yes. So, essentially as a kid, I, I was fascinated with space. I,
was born in Australia, [00:02:00] Melbourne, Australia, and at the end of my
PhD. In, particle physics and astrophysics and cosmology. I was tremendously
fortunate to have the opportunity to move to the United States to take up a
postdoctoral position at, the United States' leading, national Science Lab.
Which is Lamos National Lab in Los Lamos, New Mexico. once IMOs, I gotta
work on some, you know, gotta fulfill childhood planetary exploration mission
missions to Mars missions. but my advisor Atmos, strongly recommended that
there was a lot more, long term career stability if. Looking at other planets.
So, um, I made the transition to looking at, uh, to, to helping with earth
observation. I think the thing with space is that the rockets, the satellites, they
get all the attention and the glory, and it's ishly difficult to design and build and
launch and operate a satellite, but in a lot of [00:03:00] ways, getting a, for
example, getting a camera into orbit and starting take pictures of.
Planet or the earth is really only the beginning of the problem that you have if
you wanna actually do something useful with the data. so my, my career has
been focused on augmenting human capability to understand entire planets by
basically bringing artificial intelligence to bear on helping to.
Understand observations at planetary scale. and uh, the work I did for the US
government over 15 years, I worked my way up from postdoc to senior research
scientist. developed some technology that, we were fortunate to win some
awards, and was the best in the world at doing map making from space.
that technology led to me having my own research team. And about 10 years
ago, I. The chance to spin my team outta Los Somos, start my first venture
backed company, Decar Labs, which was based in Santa Fe, New Mexico, and
in San Francisco. that was a [00:04:00] tremendous experience. I was the
founding cto how to do startups.
I was, after a few years at Dekar, I was fortunate enough to take a. Uh, I was
able to move on, get back into Mission Space, was recruited by National
Geographic to set up, a global sort of environmental monitoring and, and
conservation monitoring lab in Washington dc where I met my current co-
founder, Sam Hyde, who's, uh, who used to.
Work at White Science and Technology Policy Group. and the two of several
years ago, now three, almost three years ago now, we, uh, geographic, went off
and started Impact.
Danny: As far as mapping and things like that go, could you give us a brief
history of how maps have been made over the course of, let's just say the past
50 years? So there's no need to go back to Christopher Columbus here, but, um,
the past kind of 50 years as far as how map making has, uh, been happened in
the past.
Steve: Ah, interesting that you [00:05:00] used Christopher Columbus as the
example because I think a lot of people may remember from their school history
lessons that Christopher Columbus was notoriously bad at maps. he thought he
was off the China and discovered this whole other continent in the way. Cause
he sort of used maps that were of poor and ignored information that.
Didn't wanna think about, which was that there was an awful lot of ocean
between Europe and China. If you used some of the best, um, existing estimates
of the size of the world. So, so the more recent history of, is that. It has. So, so
for example, since the, um, since the end of World War ii, which is sort of the
modern period, since the end of World War ii, there's been, tremendous
capability first with aerial surveying technology.
Um, so putting cameras in planes and being able to take pictures of the entire
world, At great expense and with a lot of human flight hours. and then since the
fifties, with the [00:06:00] ability to then move that observational capability to
orbit and do it from space, the people may remember that, um, you know, the, at
the, at the dawn of the space age, um, president Eisenhower commissioned the
Rand Corporation to what was worth investing. Essentially billions of dollars
for the United States to go to space. And the killer application that Rand
identified in their groundbreaking study was Earth Observation. That was the
original killer app that convinced Eisenhower to invest a fortune later on. There
also became the race element of, um, of having, of competition with the Soviet
Union.
But the original application that made the very conservative Eisenhower spend a
lot of money, um, was earth observation. So the way it is still mostly done, it
sounds crazy to say this, [00:07:00] like 70 years later since. It is still mostly
done by professionals, humans who are trained in the art of photo interpretation,
they look at aerial or satellite photos of the earth.
They look at stacks of these across time, um, and they are trained. intensively
trained, and then they're professionally, you know, they, they're just, it's a well
established profession of how you interpret the imagery, um, how you try and
bring in the science of what you can see, not just in the sort of, What we call
visible wavelength lights, uh, light, the, the sort of pictures you take with your
regular camera, but also the sorts of stuff that you can collect from space.
So infrared cameras, ultraviolet cameras, more exotic systems like synthetic
aperture radar and hyperspectral images. The, there's an extremely rich set of
physics-based instruments that can be [00:08:00] taken to space, can be used for
observation, um, and interpreting. All of that is, Has historically been a manual
task.
there has been lots of times where information has been slow to appear in front
of decision makers because it's dependent on humans noticing something. and
there's always been sort of concern. There's always been a concern that because
there's the risk of missing something important. going back to really to the, uh,
to the early days of the space program and the competition with the Soviets, um,
that, that we needed automation to help.
So, again, going back a long way now, there have been serious efforts to try and
automate the process, to bring early versions of artificial intelligence to help
assist humans. Look at all the, all the, all the every picture.
Catch up to where we, right [00:09:00] now's been 20 over 20 years. Front row
participant in, in, um, in, in this firsthand participant in this, we've developed a
set of artificial intelligence technologies and remote sensing science
technologies. that when combined with the compute that's now available in
commercial cloud and when combined with not just. A few fancy satellites, but
with hundreds of commercial satellites, we can now for the very first time,
essentially construct a living map of the earth.
and this is the sort of map making that people dreamed of. Going back to the
time of, you know, the great sort of cartographers of classical antiquity, um, and
then of the Middle Ages, they, they want, you know, they've always wanted this
sort of thing. And now we can actually do.
Danny: So that kind of brings us to Impact Observatory. Could you give us a
brief overview of what you all are doing there today and how you're using AI to
improve, uh, the [00:10:00] map making process?
Steve: Yes. So the, the key thing, so people may, may be wondering, people
listening in May, may be wondering at this point, well, you know, what's the
point of the ai? Cause you, you've known, we've known for now, 10 years, 12
years, um, since the sort of big breakthrough with Google Earth that, um, that
we can form satellite mosaics of the world, can see those things change through
time.
And we're all used to using Google Maps or Apple Maps or Bing Maps to go
plot, how to drive down to visit your cousin or go, you know, visit a, visit a
theme park or something. You know, visit the beach if it's summertime or just,
um, you know, just navigate around as you're driving around town, or plan your,
and that type of image derived. Is now super prevalent. The problem with it
though, is that a [00:11:00] photo is not a map. So the difference, the key
difference that I want to make sure people understand between photo and map is
that with a map, you have not just got an image, but you've actually labeled.
Every single small patch down to the, if it's a RAs image down to the pixel,
every pixel in the photo has a label, and the labels change across time.
So especially for this prompt pod audience and thinking about how you talk to
ai, the fundamental problem in map making. Is essentially a conversation
between the mapmaker and the planet and you kinda like the Dr. Seuss Lorax.
You're giving a voice to things that don't speak, and you're trying to provide
labels, which allows you, as you look at these maps, to both show and tell, to
understand the relationships between relationships. And to [00:12:00] be able to
automatically answer questions, like, okay, how much forest is in this area? Or
How much water is there here? And how much water did they used to be here?
Or how much forest used to be here? How, you know big is this? How big is
the, is the city or this town and what used to be there before the city was there?
Like, what did this look like 50 years ago? And, through the, just the sheer
volume of data, there's a massive need for artificial intelligence to, to automate
the process of making all the maps that are needed. and, uh, and, and the, the
other key thing, the other key point I'll make is that, When I was at the US
National Labs with a number of my colleagues who were actually now with,
with us at Impact Observatory, my head of science, Dr.
Amy Larson, my head of engineering. The three of us we worked together at, at,
uh, at Los Somos Lab, we've built systems to take satellite images and turn 'em
into maps. We built those types of systems for the US government at the
National Lab. We built those types of [00:13:00] maps. Mark and I built these
types of maps, at Decart Labs for Wall Street.
and the mission for Impact Observatory is to take technology that historically
was only available to the richest countries and the richest. Companies on Wall
Street. And now because of this moment with Commercial Cloud and with
publicly available satellite data and all the new commercial space, organizations
that are, companies that have, that are putting up their own satellites, there's this
tremendous moment to democratize this technology, to make it available to
every, everyone so that every country in the world.
Should be able to have the same sort of decision support tools, geospatial
decision support tools to plan out city growth, to be plan out essentially
sustainable development for the and sustainable livelihoods to understand
climate crisis and understand environmental risks. Be able to do that with the
same sort of quality data that the US government and the biggest companies
take for granted.
Danny: And tell us a little bit, ex, um, [00:14:00] kind of technically how it
works and obviously no trade sequence secrets need to be revealed, but, um, do
you contract with just commercial satellites and then that data gets pulled down
and it gets fed into a model? Or what does, what does that look like?
Steve: Yeah. So, um, so fundamentally, uh, and, and actually one of the reasons
I've had the opportunity to, to now do this, my second company, is that. Like 10
years ago. Actually even, yeah, about about 10, 12 years ago now. Um, at the
dawn of what we now think of as the deep learning era of ai, I was a principal
investigator at Almos Lab, saw the first work come out.
Tommy Pogi and Thomas's lab at mit, PO's lab, Tommy's lab. And um, we
breakthrough by.
Of the, just the, just the sudden leap forward,[00:15:00] with deep learning,
computer vision models, finally trained with sufficient amounts of data that they
actually suddenly started to work. there's a famous so young Lako who runs AI
at. at Meta Facebook slash meta, Lako was famous for having done a bunch of
work on, um, sort of looking at different size data sets, uh, especially with like
handwritten digits used by the US Postal Service, and, um, and then Hans at
Carnegie Mellon University written paper.
If you look at the amount of neurons in human brain and you ask, how much
data does it take to train a human to do every day? To to live, right? You try and
estimate how much language does a human is a human exposed to in a given
year, especially an infant is exposed to a year. How much do they see in a you
can [00:16:00] volume data? The situation until, basically, until the early two
thousands was that we were trying to train AI systems to do everyday tasks with
data sets that were like six orders of magnitude, a million times too small, and
wondering why they weren't working. Um, and trying to compensate for that by
trying to. Essentially guess or derived from first, first physics principles, what
the logical rules for a vision system or an audio system need to be to, to explain
language or to explain vision. and those systems worked somewhat, but they
never really worked the way we thought they would. And it's sort of dumb, but
the, the basic principles of how to train a convolutional neural network, um, and
the convolution in neural networks comes from early work in, um, uh, systems
neuroscience, going back to [00:17:00] Horace Barlow and, and some of the
early people doing recordings from human retina and then, and then visual
cortex, neural networks.
They were worked out in the fifties. they were always expected to be the other
type of architecture that could be the basis for ai, but they needed the amount of
data to make it work. So in the last 10 years, we finally had enough data. And
actually one of the reasons Impact Observatory has the chance to do what we're
doing is because while my team and I were National Geographic, we actually
followed the patent of what. Group at, uh, Caltech, at Stanford, follow the path
that had been taken by other people who had assembled very large training data
sets and then used that to show substantial progress in the ai. Um, and so my
team and I, we were able to leverage the network, national Geographic to create.
By far the world's largest training data [00:18:00] set for understanding for
human labeling of the earth surface.
And then we've leveraged that in the last few years now to produce what has
been independently peer reviewed to be the world's most accurate map of the
world.
Danny: So before that time, was it just a matter of a data sets being too small
and B, not enough compute power to even process those data sets?
Steve: Yes and yes. So fun fact, There are two data sets that are used by the
United States government that are essentially the gold standard of national
mapping, right? They are the best in the world. Um, and. And I'm good friends
with the people who make them, and I know a lot about how they make them
full.
Full disclosure, I've served on the Department of the Interiors, Lance Advisory
Group. I, I've worked, they've been my colleagues since I was at the National
Labs and they were USGS or, or elsewhere in US government. and. A lot of the
things about those national maps are [00:19:00] very out of date by today's
standards.
Um, the SGS example, um, the sgs does a tremendous job of making data at 30
meter resolution. They make a map every two years, and so listeners might be
surprised to hear the most advanced national map in the world is. And in, in
June, end of June of 2023, the most recently available U S G S map, official
map of the United States dates from 2019 before the covid epidemic.
So imagine trying to be a national government or a state government using
national data provided by your friendly national government, and you're trying
to make decisions about. Climate change and sustainable development, and
you're using data that's literally years old and um, and is updated every two
years.
And there's two year gaps between the data sets and, and the United States is the
most advanced at [00:20:00] this. Other countries have even longer gaps
between the data or the data might only be available at even course of spatial
resolution. SGS has, has a pretty nice computer set up. That was far ahead of its
time when it was created, but now is very small scale compute compared to
what's available in commercial cloud.
and the case with the U S D A, scape data layer is kind of even. Well, um, they,
they, they are struggling to be able to produce a map of all the different specific
crops grown across the United States every year, which is a sort of a unique to
the United States map of agriculture, that that takes them so long to process
using USDA's compute system that they're only able to produce it at the end of
the growing season.
So when we were at Decar Labs, we actually had the opportunity to basically to
revolutionize the way you can calculate, crop, do crop monitoring from space
and essentially in near real time. and now it impact observatory. We've cracked
how to do both that and [00:21:00] also, um, general land use, line cover change
so that it's not just, so Labs was known as an agricultural.
AI company or agricultural monitoring company. Impact Observatory is an AI
plus space company that's able to understand basically all the major land use,
land cover change processes that are happening at the different timescales that
are needed to understand what's happening in forests as well as what's
happening in cities and what's happening, um, in agriculture and wetlands and
all sorts of stuff.
Danny: Why does timescale matter so much as far as sustainability and
climate? Um, uh, kind of go,
Steve: there's two things to say here, and one of them is really pretty scary. So I
think part of framing of. know, there's lots of stories about how dire the climate
crisis is and how many decades now. The scientific community has been near
unanimous in [00:22:00] highlighting the dangers from human induced global
warming, um, and climate change.
and those, Risks are like, like I think the, the increasing prevalence, frequency
and, and um, and violence of national na uh, natural disasters like storms and
cyclones and things is, um, is like upon us now. And I think even everyday
people are starting to see, it's no longer theoretical now it's right on us, but
there's some other problems with just natural resource usage that are even more
pressing.
there are large parts of the United States where the agricultural choices that state
governments have made, the licensing choices people have made about where
they grow different crops and what type of crops to grow and how much
building development to, to sort of subsidize, not just promote, but actually
really subsidized with taxpayer money has resulted in a number of us
watersheds.
and, [00:23:00] um, forest systems where they're under tremendous pressure and
even before the climate change gets us, there's actually a lot of parts of the
United States that's at risk of serious, fundamental change because of natural
resource use. and again, remember like even the United States, the government
is using data that's years old and trying to make decisions, and the state
governments are even slower than the.
there's, there's an extreme need to understand the state of the world to
understand the trends that are impacting the world, to be able to have a
conversation with the planet. And, you know, ultimately in the future be able to
get into better forecasts of what can happen under different types of scenarios
of, of policy and scenarios of individual and collective action.
Um, that AI and space data, the combination of AI and space data powered by
the, the, the, the new capabilities of commercial cloud, [00:24:00] um, unlock.
you know, and really just in the nick of time, and, and you need to be able to see
some of the problems to really understand what's going on, to understand that,
oh yeah, all of this cotton is being grown. In a part of the country where it
wasn't there, just, you know, some of these things, some of these cash crops
have only been introduced in the last 10, 15 years in some places, and they're
drawing down aquifers at a rate that's totally unsustainable. and the, um, the
costs and the pollution of these practices are just being pushed to the taxpayer.
Right. We in the United States, we, we like to find market solutions to things,
but you can't have market solutions without information and public
transparency. and, uh, you know, and there's just been so much greenwashing of
impacts of people's [00:25:00] decisions that just, and that needs to change.
Danny: My wife is a hydrologist and she actually does, uh, hydrologic
modeling, and she always reminds me that if the data is only updated once a
year, then suddenly your a hundred year events aren't accurate anymore. They
may actually be 10 year or five year events as. Kind of, uh, data changes more
quickly and quickly.
Is that kind of an issue here as well, is just that the models will break if we don't
update the data quickly enough.
Steve: Oh, absolutely. And um, and until recently there wasn't enough compute
in most of academia to even handle data. If you got, if you had models that
could produce, you know, um, more data, more frequently than annual. Um,
that, uh, you would struggle to do the computation unless you were at a fancy
national lab with access to a gold plated supercomputer.
Um, and if you were, and, and even if you did that, the state governments and
the regulatory agencies struggle to know [00:26:00] what to do with that data.
Um, so there's, there's been tremendous barriers to trying to get the data in front
of the right people with the, in a way that is not just. Like, you know, these are
very large amounts of compute.
There's a lot of smarts, goes into making the science. Um, these things have all
been tested and verified now, um, to the, you know, in a, in a very, uh, you
know, in a big, big, uh, well verified multifactor sort of thing. at the end of the
day, you're trying to convince a decision maker to make to, to, you're trying to
provide, not convince, you're trying to provide the decision maker with the best
data that they can have to make a, a good decision.
Um, and you're trying to also inform the stakeholders of that decision maker so
that they can judge whether the decision maker is making the right choices. Um,
and so getting back to that idea of democratization and, and what I like to think
of as the opposite of greenwashing. So I think we all, you know, greenwashing
is the practice of, you know, [00:27:00] essentially people self-reporting that
everything's fine.
Or, um, relying on old data that doesn't show the actual impacts of stuff too.
Again, assert that things are fine. the opposite of greenwashing isn't sort of a
well known term. So actually I. I personally like the term provably green, can
we actually suggest, can we actually show that certain actions, certain impacts
are provably more sustainable, more green, more, you know, just better ideas
than others.
Um, and if you can do that, then the markets, the political system can actually
act to reward the people in the organizations and the companies. So, The, the,
the data, the, um, the right sorts of insights, simple enough so that people can,
can absorb them, um, and help identify the, the folks who are doing the right
[00:28:00] things.
And if somebody figures out like the right way to preserve forest or the right
way to re rewild and reintroduce forest into a watershed so that everybody
downstream has better water, more water. Higher quality water. Um, if
somebody figures out schemes to do that and they work, how do you, how do
you allow that person, that group to show what they've done and then get the
word out so that other groups can, um, can find a version of that approach that
works for them?
And then similarly, you can have feedback loops that help people adapt. To the
climate change that some of the climate change is now baked in. There's been in
a sufficient decades of delay that some of this is gonna happen anyway, but we,
but we should not be in a situation where there's just doom and gloom.
There's still so much that people can do to head off the worst of climate change
and to have a happy, sustainable future. Um, that's a lot better. Has a world
that's actually better [00:29:00] shape than it. And, and to do that, we need the
right decisions at the right time, quickly enough so that people can actually get
feedback on what they're doing.
Danny: Are there any verification, verification, certificates these days that are
geospatial versus just going out and, um, looking for guaranteeing that carbons
being sequestered or anything like that? Is anybody doing geospatial
verification?
Steve: Yeah. So there's a number of programs around the world to try and have.
Credits and, uh, to give people credit for things. So one of the best systems that
works well, um, is the, is the US, um, framework of what we call wetland
mitigation banking. So if, if some developer wants to, um, you know, do some
construction chain, drain, some wetland to put in some whatever, [00:30:00]
way of saying, okay, this is the number of acres of wetland that you're
converting. Um, you here's a regulatory framework and a structure that already
exists that says, okay, you wanna develop those wetlands. Here's some nearby
wetlands that for every acre you get rid of, you convert, you are going to invest
to restore and preserve one and a half acres or two acres for every acre that you
took away.
Right. And this is a system that's like going well, as far as I can tell it, it looks
like it's actually doing well. There are some other systems that have had a
number of sort of, you know, have, have had some bad press recently because
there's been some issues implementing them. Um, in particular forest carbon
credits.
I've had some issues because it's very complicated to measure the amount of
carbon all the way through the forest. You know, like when you think about
everything in the understory and the root system and the soil, and then the
actual layers of canopy and all sorts of [00:31:00] stuff going on in forest, it's
very complicated.
Um, and the, the state of the art for measuring the carbon in a forest, Um, is
something that hasn't changed much since the, since the thirties, I think. You go
out the forest and you literally hug it. You measure the width of the trunk at the
diameter of the trunk. At breast height is the actual terminology.
So you literally go out there and hug the tree and figure out how big it is at the
trunk. Um, and then there's all this formula and stuff, so it still ends up with
some estimate. Um, people have tried to figure out how to do this from space or
from the aircraft. From aircraft. There's some pretty good estimates.
Are they precise enough to really justify like, like if they were really honest
with their Arab bars being a physicist and a mathematician from back in the
day, I understand Arab bars and you know, like, um, you know, if you're really
honest about the Arab bars on carbon credits, they're kind of loose at the
moment.
Um, so. That bit of carbon credits is difficult. The [00:32:00] easier bit that you
can do from space is to say, well, okay, somebody spent a lot of money to go
get red certification or Vera certification for their forest. Those are two of the
big sort of schemes. Um, R E D D and Vera v e r r a. Um, you've gotten your
estimate of the carbon in your forest, who's now watching to make sure that that
forest is still there.
Danny: Yeah, no one.
Steve: And in, and in fact there was a famous scandal from 20 years ago called
Sino Forest, about a forest in China that, that sold a bunch of carbon credits.
And then later on when people went out there, they discovered there was no
forest. So just no forest at all. So, you know, there's a a, there's a very pressing
need for providing the transparent monitoring systems to go along with.
Programs so that investors, politicians, everyday people can feel confident
[00:33:00] that, again, taxpayer money's not being wasted, systems are actually
being put in place. And now you can fairly talk about. Trying to estimate the
actual impact that industry's having on the on, on the public natural resources,
um, and, you know, downstream pollution, atmospheric pollution, these other
things, and start to actually put the costs onto the industries that are generating
these hazards for society.
Danny: Besides the verifications that we just kind of spoke about, are there any
other needs, um, as far as kind of mitigating climate issues that you believe are
currently underrated or not rated at all?
Steve: Yeah, so, you know, a lot of the stuff in in climate, in the climate crisis
that gets most of the press, um, has been around, um, energy production and,
um, atmospheric sources of, you know, global. Greenhouse gases that
[00:34:00] change the climate, that long term change, the that, that are causing
global warming. Um, you know, and identifying what are the biggest polluting
facilities.
Okay. Now, in addition to that work, and there's been a lot of progress on that,
including some space-based work. So there's a satellite called methane sat by
the Environmental Defense Fund, and then there's GT sat and there's a few of
these satellites that try and, um, uh, that are. Greenhouse gases emitted from
different parts the earth. After energy production and transportation. The next
biggest chunk of stuff that we need to better mon monitor and manage is the
global agricultural systems and basically the extractive, um, uh, minerals and
also the forest management systems all around the world. Um, and that area
[00:35:00] I think is. pun intended.
With respect to agriculture, for much better monitoring, um, the US knows how
to do it. Canada, to a lesser extent, has sort of done this in the past. Uh, Europe
is doing it nowadays, but most of the world does not have the sort of
agricultural monitoring systems that the US that. This area is very amenable to,
to, um, help from AI powered space data analysis. And so lot of my work, my
team and work over the has that area.
Danny: Was there a turning point or kind of a tipping point for you as far as
when you stopped pointing your eyes upwards towards space and became more
unconcerned with Earth and the climate?
Steve: Um, I'm so sorry. Can you just repeat that? The audio tracked off just a
sec.
Danny: Yeah, no problem. Uh, so were there any points, um, kind of [00:36:00]
tipping points where you stopped, um, kind of caring as much about space and
wanted to focus your kind of time and energy on solving climate issues and
problems on the earth?
Steve: Yes. So, um, all of this became very real for me. Back in the year 2000
when the national lab I was at, almost burned down. Um, so there was a
catastrophic wildfire that hit New Mexico back in 2000. It was sparked by a
controlled burn that got outta control by the US Park Service on federal land.
Um, The forests had been managed, so US forests are managed in a way that
emphasizes fire suppression. And um, and one thing that a lot of people maybe
don't realize about US forests is that US forests, Even the ones that look healthy
and green and [00:37:00] very lush and ent are actually very different from the
way the forests looked when the first Anglo explorers and, and before them, the
first Hispanic explorers went across the, the, the country and encountered all
the, the indigenous Native American groups that were living here already for,
um, and the forests used to be, Um, is just very different from how they are
now.
In particular, the most US forests are totally overgrown. Um, they're missing a
lot of their species of animals. That evolved, co-evolved with the forest to
regulate the forests and essentially make them safe for themselves. Forest down
every five years the. so anyway, so when this gigantic catastrophic wildfire
happened, that burned down, I think 50 square miles of forest and almost took
[00:38:00] out the US' premier nuclear science lab and almost caused a major
radiation scandal. The heroic efforts by the forest fire by the firefighters stopped
that didn't stop 10% of the town of Los Alamo.
And, um, it didn't stop everybody in town having to become a refugee. 10 days
or something. Um, and uh, afterwards I volunteered to help with the data
analysis using aircraft imagery and satellite imagery, and we were able to sort of
show, it was a very early demonstration of using AI to map out the effect of the
fire and fun fact when I did the analysis, And I was mapping out the burn scar,
the, the scar left behind by the fire.
And the fire was so intense cause the forest was so overgrown. The fire was so
intense that it didn't just burn the forest. It actually burned into the soil and
destroyed the seed bank for the forest. So it like turned the whole thing into
glass. Um, [00:39:00] when I did that analysis, the AI kept saying that there was
this, And it just, I couldn't, I couldn't, I played with the ai, I was trying to get it
to like, fix this obvious mistake and it wouldn't do it. So in the end, I called up
the ranger station on the other side of the, of the mountain and asked them, you
know, like, was there a fire over there a few months before the big one that, that
got outta control and took out the town, um, and caused, you know, a billion
dollars worth of damage to New Mexico and fire Marsh over there was like all
sheep.
Yeah, we had a little controlled burn that got outta control, but it happened on
the other side of the mountain. Same mountain, but just on the other side of it.
And so the word didn't get up. So you know, like there was this like false, there
was this early warning sign that the conditions were now so extreme that maybe
policies of controlled burn is actually have to be tightened.
To be safely executed, [00:40:00] that if that had gotten out, it would've
potentially held off this giant wildfire. And then, and then the rest of the story
is, so I've had the firsthand experience of being a refugee from, from natural
disaster. That was the inciting incident for me to get into this sort of like doing
this, this main focus in my career.
And 10 years later, There was a second catastrophic fire that almost burned on
Los Lamas again. Um, and the first time my analysis of the forest after the first
fire showed me that there was still a ton of forest that hadn't burned yet, and the
rest of the forest was in no better shape than the that burned.
And so I decided to move my family and I outta Los
Danny: Wow.
Steve: first. I was a refugee. We. We were actually in Santa Fe where we were
safe, and we actually took in refugees from Los Alamos that went down a
second time. Um, and that sort of getting ahead of the climate change, getting
ahead of the [00:41:00] natural resource and the, the, um, the natural disaster
risk, um, and making better decisions so that you are no longer like, have to be a
refugee again.
Um, Is very personal to me because I've actually like lived it a couple times
now. And the last like sad fact about that is that when the first fire hit hit in
thousand, it was by far the largest wildfire that New Mexico state of New
Mexico had ever seen. When the second fire hit in 2010, which was bigger than
the first one, the first fire was no longer even in the top 10 for the state. So the
degree to which the United States is, you know, like we, the people make jokes
about the poo frog in boiling water. And if you turn up the temperature slowly,
you don't notice it until the cooked. I think a lot of people, we forget how much
things have changed in just the last 20, 30 years. Um, there's 50 years of space
[00:42:00] data in the can, and this is the moment where we need to bring that
data to life so that people can have a conversation mediated by AI with the
planet and understand really the risks and the threats, the opportunities for. To
make better decisions and actually try and reach a sustainable future.
Danny: You've had an incredible career and one that I assume is far from over.
Do you have any advice for, for young people that, you know, want to, um,
work on big projects or with, um, incredible organizations that, and also what's
your advice on building incredible teams?
Steve: Right? So, um, So I think the, the, the first bit of advice that I, um, I
often tell students is learn some programming languages, right? [00:43:00]
Programming languages is the interface between data and, and, and ourselves.
Okay. Right. That's the natural language for talking. And then from the sensors
to talk sensor collect data and programming languages is the access.
The way I would've suggested to learn programming five years ago, um, has, is
going through a bit of a revolution at the moment with AI co-pilots. So it turns
out that of all the things that you can stick AI onto, that you can ask AI to help
us with learning to program is actually turns out to be one of the best things that
you can do with it.
So there's some really interesting copilot technologies that developed. In the
generic sense, I'm not referring to any particular product, but just the, just the
ability of [00:44:00] having AI Buddy who can help you learn program. There
are some tremendous systems coming up and this, it's never gonna be as easy to
get into data science as it will be.
Um, I also think that, um, anybody who wants to do this type of stuff, you need
to get your programming skills up. Um, especially languages like Python that
are basically the standard language for data science nowadays. Um,
understanding how to program in something like Python, understanding how to
use the AI powered tools to help you learn and then, uh, quickly adapt to new
things necessary.
In some ways, the simplest jobs in, in data analysis are gonna get eaten by the
ai, and the humans are gonna be needed only for the more interesting jobs,
which is good for the humans, but it's gonna, it's gonna be problematic for
people trying to get started industries. Um, I think the key to building good
teams is to have good mission if you giving your [00:45:00] team important.
Impactful mission, then you are more likely to find people to, to attract the right
sorts of people who wanna do that for the right reasons. Um, I've seen
organizations that are more mission driven, like National Geographic Society
and Los National Lab. I've seen organizations that are more driven purely by
commercial objectives. I know which ones, which sort of organization I wanna
work with. And Impact Observatory is designed to be a mission driven
technology company that, um, sort of leverages this democratization of. Fund
ourselves by providing premium products to industry and G 20 governments,
and then use some of the proceeds of the company to provide public good data
sets that will help everybody, um, including the public data that we've already
released through the UN and through our cloud partners that have, um, you
know, really started to change the way that [00:46:00] the, that the world
understands itself.
Um, so do not, yeah. So anyway, so bottom line is mission. Um, Uh, the
National Labs are jewels in the US system. Um, if you, if anybody has the
chance to take a student position or a postdoc position at a US National Lab,
any of the national labs, I strongly recommend those. It's fantastic places to
meet other people who are attracted to mission and to be exposed to some of the
world's best technology.
Um, And, but I also think that we're, we're now in a moment in world history
where digging our way out, digging ourselves out of the climate crisis is a team
sport. It is not enough to just think this is just gonna be solved by some
professors at Harvard and Stanford and Caltech and mit, or this is just gonna be
solved by people just by. Who are in [00:47:00] Silicon Valley. Um, it is a, it's a
team sport. You need people from lots of different points of view, including
from conservation and from indigenous communities and, and local knowledge
all coming together in a way that. Enables people to bring the best of their
experience and their, their knowledge to bear in a way that's mutually respectful
and builds on each other and doesn't try.
One group doesn't try to dominate everybody else.
Danny: Always our final question. What's the hardest? You've laughed
recently.
Steve: Um, Uh, yeah. When you, when you work on climate change and
climate crisis, it's important to.
Danny: I.
Steve: Don't sink too far into anything that's bad news. Um, I enjoy watching,
um, comedies. I enjoy watching, uh, TV shows that are, you know, streaming
shows that are funny. Um, some of [00:48:00] the best laugh I've had recently
actually is, is when I, about once a year I rewatch most of the Silicon Valley TV
show.
I find that. it is surprisingly insightful and there are a number of real world
stories that are behind the plot lines in Silicon Valley. And um, so anybody
who's like, don't watch Silicon Valley is a how to, but don't be surprised if
you've watched it to discover that when you're in the startup world that things.
Danny: Yeah, it's a fantastic show. Well, Steve, thanks very much for coming
on our show. If listeners are interested in learning more about what you're
doing, where can I point them to online?
Steve: Yeah, so, um, impact Observatory, um, uh, website is impact
observatory.com. You can find us on Twitter, you can find us on LinkedIn. It's
easy to find me Steven Brumby on LinkedIn and on the, all the different
[00:49:00] inner tubes of the, uh, of the Internets. So, um, Uh, you know, and
we look forward to engaging with people.
I think, um, this new wave of AI is super exciting. I think prompt engineering is
a great thing for people to,
glad for this opportunity to be on, um, prompt Pod today.
Danny: Excellent. Well, Steve, thanks again for coming on.
Steve: Thank you, Danny.
Danny: And thanks to you, my dear prompter for tuning in, and I hope you
enjoyed this conversation as much as I have. If you enjoyed the show, please
consider subscribing and leaving a good review. Take care and always be
prompting.