0:08

So how did you start farming? I am

0:10

a fourth-generation farmer. Our

0:12

relatives migrated from

0:14

Europe and in South America. We

0:17

all ended up in Southwest Kansas, so

0:19

it's always been in our blood, kind

0:22

of a way of life and that's how I ended

0:24

up farming. Back

0:26

in 2015, Duane Roeth was farming 6,500 acres of

0:28

corn, wheat,

0:35

sorghum, and sunflower in Southwestern

0:37

Kansas. It had been a

0:39

wet year. Rainfall was seven inches above average

0:41

for much of the growing season. So

0:44

in December, when Duane got an invitation

0:46

to attend a meeting about water sensor

0:48

probes, he wasn't interested. He

0:51

was like everything else in farming. Somebody's always trying

0:53

to sell you something. Duane and

0:55

his family spent generations growing

0:57

crops without water measurement technology,

1:00

and he wasn't about to change, especially at

1:02

$1,000 a sensor. The

1:05

guy called me back the next day, he goes, how many probes

1:07

you want? And I said, you know, I'm not using any of

1:09

those probes. I go, you can stick those probes right up your

1:11

ass. I'm never going to use them. A

1:13

year later, the state of Kansas

1:15

launched its Water Technology Farm Research

1:18

Project, a three-year pilot

1:20

designed to test the latest water

1:22

conservation technologies on three working farms.

1:25

And Duane's farm, it was one of them.

1:27

Come the spring after the corn was planted, we

1:29

had about every damn soil moisture probe company on

1:31

our farm that you could think of on this

1:33

Water Technology Farm. I mean,

1:36

it engulfed it. Why the change

1:38

of heart? Duane had learned that those $1,000

1:41

probes would actually be free thanks

1:43

to a USDA program. So

1:45

he leaned in. Soon, the probes

1:47

were giving him a comprehensive picture of the

1:49

moisture content on 125 acres of

1:52

his soil. But in the early

1:55

hours of the morning, when he'd see

1:57

his neighbors' pivots watering their fields, he

1:59

didn't entirely what his sensors were

2:01

telling him. You could see one or

2:03

two lights start up in the early spring, those

2:05

strobe lights, and then boy, the next day there's

2:07

hundreds of them on. You know,

2:09

it's almost like a visual, like, okay, I probably ought

2:11

to get my pivot on him. Duane

2:13

resisted turning the water system on until

2:16

he noticed the leaves on his corn curling

2:18

up from heat and high winds, and

2:21

then he panicked. So he called

2:23

the sensor guy. I said, I'm gonna have to turn

2:25

these down, pivot's on. And that's when he said, I

2:27

want you to stop looking at those goddamn strobe lights. He

2:30

goes, I want you to go to the shop, I want

2:32

you to look at your computer. He goes,

2:34

the top six inches is a little dry. But he

2:36

goes, three and a half foot down, you've got a

2:38

solid profile. He goes, the gas tank's full. And

2:40

by that, he meant Duane had enough

2:42

water in the soil. Duane's

2:45

sensors were giving him accurate data, and

2:47

his yields that year were just as

2:49

good as previous years, saving him 10

2:51

inches of water. 10

2:54

inches may not seem like much,

2:56

but if farmers across southwestern Kansas

2:58

followed Duane's lead, all of those

3:00

water savings could help stabilize a

3:02

critical aquifer running under their land.

3:04

In the state of Kansas, agriculture

3:06

is king. The secret to this

3:08

land's bounty is not just the

3:11

soil that covers it, but also

3:13

something below ground that no one

3:15

can see, the Ogallala Aquifer, the

3:17

largest freshwater aquifer in North America.

3:26

The Ogallala Aquifer stretches from South Dakota

3:28

in the North to Texas in the

3:30

South, and it supports about $35 billion

3:33

in agriculture every year. But

3:35

ever since the Great Dust Bowl of the 1930s, when

3:38

the US government assigned farmers irrigation

3:40

rights that boosted well drilling into

3:42

the aquifer, its water levels

3:45

have been dropping. Today's irrigation technology is

3:47

able to pump out water that has

3:49

been in the aquifer for hundreds of

3:51

thousands of years, in just a matter

3:53

of minutes. A rate which

3:56

far outpaces how fast nature can replenish

3:58

it, threatening the overall sustainability. of

4:00

the aquifer and agriculture. And

4:03

there's another factor making the problem much

4:05

worse. Climate change. Shifting

4:10

seasons and extreme weather are threatening

4:12

food production well beyond Kansas. There

4:14

are about 50 harvests left in the

4:16

world. We have to get more food for

4:19

the world because there's more and more humans.

4:21

We have to get it cheaper. We have to

4:24

get it in a way which is going to

4:26

survive adaptation to climate change. We have to do

4:28

it with much less water with a much lower

4:30

carbon footprint. And we have to do that all

4:32

at the same time. Astro Teller

4:34

is the captain of moonshots at Alphabet, who

4:37

we heard from earlier this season. And

4:39

there's no way to do all of them

4:41

except to bring technology to bear. Technology is

4:43

the only chance we have moving

4:45

from the sort of 20th century mechanical

4:48

view of farming to a

4:51

21st century view of doing that same

4:53

kind of work. That includes

4:56

technology like the soil moisture probes

4:58

on Dwayne's farm and OpenET, an

5:01

online platform that uses multiple satellite

5:03

driven models and Google Earth Engine

5:05

to track water consumption. And

5:08

also technology for distributing food to the people

5:10

who need it the most, all

5:12

while generating the least amount of waste possible.

5:15

Those solutions are all rooted in

5:17

better data, says Google's Emily Ma.

5:20

There was a McKinsey study that listed sort

5:23

of level of digitization of

5:25

all the industries. At the very bottom

5:28

was food and ag. We used to

5:30

have a saying on our team, silos

5:32

is for grain, not for data. Data

5:34

doesn't have as many benefits if it's

5:37

sitting in a corner somewhere, it needs

5:39

to be shared but

5:41

safely with great intention. And

5:44

we're better when that data helps us all

5:46

understand how the system works better. This

5:52

is where the internet lives, a show about

5:54

the unseen world of data centers. I'm

5:56

Stephanie Wong, I'm your guide to the people and

5:59

places that make up. the internet. In

6:01

our fourth season, we're exploring how data

6:03

centers are enabling a more resilient world.

6:06

So far in this season, we've

6:08

heard stories about how technologies run

6:11

by data centers are helping manage

6:13

wildfires and build more reliable electric

6:15

grids. In this episode,

6:17

a better food system. We'll

6:19

learn how data-driven predictive tools are

6:22

helping farmers use less water and

6:24

improve yields. We'll look at

6:26

how data is getting excess food to those

6:28

who need it most, and we'll confront

6:30

the paradox of hunger and food waste

6:33

existing at the same time in the

6:35

same places. Dwayne

6:44

Roeth always knew farming was a tough way of

6:46

life. Earlier in his career, when

6:48

he was farming with his dad and his brother,

6:50

there was one night when he almost pulled the

6:53

plug on it entirely. One night,

6:55

it even came to a point where my wife

6:57

and I were sitting in our dining room table

6:59

at our house where we lived in the country.

7:01

And we're trying to decide, you know, we knew

7:03

farming was really, you know, it was a way

7:05

of life, but it's a tough business. Like, okay,

7:07

if we're not going to pick up any ground,

7:09

maybe we should consider doing something else because, you

7:11

know, we're really majorly getting by.

7:14

But he stuck with it. And after

7:16

signing on to become one of Kansas'

7:18

first water technology farms in 2016, Dwayne

7:21

became an advocate for high-tech approaches

7:23

to water conservation, even

7:25

as other farmers doubled down on traditional

7:28

theories around water use. There's still some

7:30

mentality that, hey, the more water we pump,

7:32

the greater the yield. That is so untrue.

7:35

So pouring the water on it isn't

7:37

the correct solution. It's

7:39

giving the plant the water when it needs it.

7:45

In 2021, a year before one

7:47

of the worst droughts in Kansas

7:49

history, OpenET launched. It was

7:52

a public-private collaboration led by

7:54

NASA, the Environmental Defense Fund,

7:56

Desert Research Institute, Google Earth

7:58

Engine, and others. and

8:01

it aimed to help farmers know when,

8:03

where, and how much to irrigate through

8:05

satellite imaging and weather data. Duane

8:07

and his nephew Zion had plans

8:09

to start using Open ET last

8:12

July until a hailstorm destroyed their

8:14

target crop. So for

8:16

now, they're relying on their water probe

8:18

sensors to help preserve the Ogallala Aquifer

8:20

and save money. Every time

8:22

we turn that irrigation well on, it costs

8:25

us money. So we have natural gas, power

8:27

that irrigation engine, diesel to power that irrigation

8:29

electricity, power that irrigation engine. We're looking at

8:31

anywhere between $80 to $125 a day. There's

8:36

significant value to leaving that in the ground

8:38

to use that to a later date. Duane

8:40

knows it'll take some convincing for farmers

8:43

in Finney County to start thinking differently

8:45

about how they manage water, and

8:47

to hit the state's target of reducing water use

8:49

by 15% in western Kansas. Doing

8:53

so could actually double the life of the

8:55

Ogallala Aquifer in the region. Again, I

8:57

bring up soil and marshal probes. I

8:59

bring up Open ET. Twenty years ago,

9:01

we didn't have all this technology, but it's

9:04

about adoption. How do we bring the entire

9:06

industry into this? We've got to start managing

9:08

this a lot better than what we've been

9:10

doing with the technology that we have now.

9:13

So you've passed your farming on to your nephew Zion,

9:16

and now you spend your time advocating for

9:18

conservation. In 2021, you also

9:21

won the Leopold Conservation Award for that

9:23

work. And you were pretty

9:25

emotional when you received it. How were you

9:27

feeling at that moment? It

9:30

just gave me confidence that I

9:32

was doing the right thing. When

9:34

you're doing this, you're questioning yourself like, are you doing the

9:37

right thing? You know, I'm glad

9:39

Zion, I not only was able

9:41

to successfully transition the farm to him,

9:44

he's just as passionate or more than

9:47

I am about water conservation. Now

9:50

he has a little baby girl,

9:52

she's probably 15, 16 months now,

9:54

you know? And she

9:57

might not farm, she might. But

9:59

you've got to have water. Technology

10:01

alone won't save agriculture. We need

10:03

more people like Dwayne encouraging fellow

10:05

farmers to adopt new practices. But

10:08

a wave of new tech solutions are starting

10:10

to transform the food system, helping

10:12

with far more than water conservation. Astro

10:15

Teller has been working with different teams

10:18

at X to pursue moonshots for boosting

10:20

the sustainable production of food. It's

10:22

all about putting intelligence into things so

10:24

that the machines that we build can

10:26

be working for us faster, smarter, more

10:28

efficiently so that we can solve all

10:30

these problems at the same time. When

10:33

you explore the different solutions across

10:35

agriculture, what are the most acute

10:37

problems you see? Let me give you an interesting

10:40

one in the ocean space that's

10:42

about food. Humanity gets about

10:44

two and a half trillion dollars a year

10:46

from the oceans, and we're destroying the oceans

10:48

faster than we're destroying either the land or

10:50

the air. It is essentially

10:52

like the sink for us that's like

10:55

pulling all of humanity's badness into it,

10:57

but it can't take much more. Somehow,

11:00

we need to get more goodness

11:02

from the oceans while simultaneously

11:04

start to revive the oceans.

11:07

There's no way we're going to do

11:09

both of those at the same time

11:11

except to take automation to the oceans,

11:13

which essentially hasn't happened yet. One of

11:15

the places that we think is ready

11:17

for that is in the aquaculture space.

11:20

Being able to farm fish

11:22

instead of do wild fishing

11:25

is also really important for

11:27

humanity because a pound of

11:29

fish meat is one-eighth, one

11:32

part in eight, of the carbon

11:34

footprint of a pound of beef. So

11:36

it is an enormous issue for

11:39

us to help the world move towards

11:41

fish protein, which by the way is

11:43

the fastest growing protein in the world,

11:45

but in an unsustainable way because still

11:47

more than half of fish is caught

11:49

and not raised. In order to do

11:52

that, you would have to make these

11:54

farms for growing and harvesting the fish

11:57

much more efficient by bringing technology to

11:59

them. working on this for

12:01

a number of years. This is our project title.

12:04

Titles mission is to bring greater

12:06

visibility beneath the ocean surface. The

12:08

team deploys cameras, sensors, and AI

12:10

powered software to provide deep insights

12:13

into the health of fisheries. Where

12:15

you could do things like look at the

12:17

fish, be able to move around in a

12:19

huge pen where there are hundreds of thousands

12:21

of salmon, let's say, and understand

12:23

things like how well are they growing?

12:25

Are they being underfed? Are they being

12:28

overfed? Do they have any health

12:30

problems? Title deployed its technology with

12:32

Mowi, a company that farms salmon in

12:34

the waters off Norway. Mowi

12:37

can now detect and interpret fish

12:39

behaviors and environmental factors like temperature

12:41

and oxygen levels and track changes

12:43

over time. These insights

12:45

have improved decision-making about fish welfare,

12:48

health, and feeding. The standard

12:50

of care in a fishery right

12:52

now is to, for 300,000 fish, you go, you

12:56

scoop them up with your hands, put

12:59

like 20 fish on the scales, you look at it

13:01

with your eyes, and you're like, yeah, they seem okay.

13:03

They're like getting chubby, I guess. And you put them

13:05

back in the water and you use that as a

13:08

surrogate for how those 300,000 fish are doing. We

13:12

can do wildly better. We are

13:14

doing wildly better with technology. So

13:16

that's a very specific example about

13:19

using technology and AI in the

13:21

cloud to facilitate something

13:24

really concrete in the real

13:26

world, getting better to

13:28

help humanity, in this case to

13:30

eat, and to get the carbon

13:32

footprint of our ability for humanity

13:34

to survive on this planet down at the same

13:36

time. We need

13:39

to produce far more food with

13:44

much less environmental damage. But

13:46

even if we build a truly sustainable

13:48

agricultural system, it only addresses half of

13:50

the problem. We still have

13:53

to deal with food waste and food

13:55

access, two problems that

13:57

deeply bother Google's Emily Ma. States,

14:00

there's like 50 million people who are

14:02

food insecure. And yet we

14:05

throw out something like 145

14:08

billion meals every year. That's enough to

14:10

feed all those people three meals a

14:12

day for three full years. So what

14:15

the heck is going on? Food

14:23

waste in the US alone causes a

14:25

lot of emissions, roughly as much as

14:27

the annual emissions from 42 coal-fired power

14:29

plants, according to the EPA. Meanwhile,

14:32

one in six people in the US

14:34

rely on food assistance. It's

14:36

a vexing problem that Stephanie Zydek saw

14:38

worsen in 2020. Can

14:40

you take me back to the start of the

14:43

COVID outbreak? You were at

14:45

the center of two really extreme problems,

14:47

disruptions to food supply chains and an

14:49

increasing need for food assistance. With

14:52

the pandemic, there were farmers

14:55

who were literally pouring milk down

14:57

the drains because they didn't

14:59

have the supply chain to get it out.

15:01

These trucks here behind me are

15:03

empty of milk, but that's not for

15:05

a lack of production. This farm here in West Bend has

15:07

been forced to dump 25 to

15:09

30,000 gallons of milk each day as

15:12

a part of coronavirus. And

15:14

that is heartbreaking in general, but

15:16

heartbreaking for people who are now

15:18

in need of something as nutritious

15:20

and sustaining as milk, and

15:22

they can't get it because they can't afford it or they couldn't

15:24

buy it at a grocery store or

15:27

any number of reasons. Stephanie

15:29

is the vice president of data and

15:31

analytics at Feeding America, a national

15:33

network of nearly 200 food

15:35

banks and 60,000 pantries.

15:38

It distributes 6 billion pounds of food

15:40

a year. And during the

15:42

pandemic, she witnessed a breakdown of the

15:44

food system. As the shelves of

15:46

grocery stores and pantries emptied, food

15:49

suppliers threw out millions of tons of

15:51

perishable food. The most recent survey by

15:53

the Census Bureau found that more than

15:55

18 million American adults

15:57

said they sometimes or often. often

16:00

didn't have enough to eat in the

16:02

past week. Food banks

16:04

are stretched to the brink.

16:06

Disruptions to supply chain dramatically

16:09

impacted donations in particular to

16:12

food banks and food pantries. So

16:15

that upended things, or even for

16:17

purchases that meant much longer waits

16:19

for those supplies to arrive. It

16:21

felt like every day there was

16:24

some brand new challenge. Government

16:26

food programs surged in the early part of

16:28

the pandemic. But by 2022, food insecurity

16:30

was still rising and

16:33

rates remain high today. But back when

16:36

farmers were letting fields go fallow or

16:38

pouring milk down drains, Stephanie

16:40

realized that feeding America didn't just

16:42

have a supply chain problem, it had

16:44

a data visibility problem. We

16:46

had been in the pandemic for a little while

16:49

and it was like May and people

16:51

wanted to be able to communicate here's

16:53

what we as a network

16:55

have rallied in this moment of crisis

16:59

and what's still needed. And

17:01

I said to them, I said, I don't

17:03

have the data. We collect the data

17:05

quarterly. The next time you're gonna

17:07

be able to get data is July. In

17:10

a moment where you're needing to

17:12

react on a literal day-to-day basis,

17:14

it's an unacceptable answer. Feeding

17:20

America had already been talking to Google

17:22

about data tools to help connect with

17:24

independent food banks. But suddenly

17:26

the need was acute. Food

17:28

bank demand had grown 60% during

17:31

the pandemic. So Stephanie

17:33

started re-imagining feeding America's

17:35

entire data ecosystem. What

17:37

would it look like to connect

17:39

our data systems with their inventory

17:42

management systems or their ERP systems

17:44

such that we could get daily

17:46

data, daily transactional level data about

17:49

the food that they were sourcing and

17:51

distributing. So we've been working ever since

17:53

then to make these automated data connections.

18:00

Stephanie is an analyst who loves playing

18:02

with data. She used to crunch

18:04

numbers for the insurance industry. When she

18:06

joined Feeding America in 2011, building

18:09

better data tools was a huge priority.

18:12

But she still faced limitations. In the

18:14

early days, we were publishing a lot

18:16

of these reports that we had kept

18:18

collected in PDF. Some

18:20

of that had to do with staffing constraints. But

18:23

as an analyst, I thought, wow, that's really

18:25

limiting what I can then do with that

18:28

information. In episode one, we

18:30

met Jorge Rivera from the Advocacy Group

18:32

One, which had a similar problem. His

18:35

team often commented on sprawling PDF

18:37

reports about global development issues from

18:40

other organizations, but was

18:42

constrained by limited access to the underlying

18:44

public data. There would be a

18:46

report, we would write about it, we would say,

18:48

this is good progress, this is not very good

18:51

progress. But we didn't have the capacity

18:53

to actually go back to the original data and ask,

18:55

is this the right way to think about this issue?

18:57

Is this the right way to analyze the

18:59

data? And Stephanie faced a similar dilemma.

19:04

What are some of the questions about

19:06

food that were difficult to answer with

19:08

limited data? We are asking really important

19:11

questions about food access. And

19:13

that's a really challenging concept to

19:15

analyze. It is a problem

19:17

of geography or proximity,

19:19

how far or how long does

19:21

it take a person facing hunger

19:24

to get to a charitable food

19:26

assistance site? How many hours or

19:28

days of the week is that

19:30

particular charitable food site open? Does

19:32

that meet the needs of people?

19:36

Do they have a nine to five job? And so

19:38

pantries that are open from one to five every

19:40

day don't work for them. Or do

19:42

they have weekend or evening jobs? And

19:44

so the pantries that are only open in

19:46

the evenings don't work for them. Like

19:48

really trying to think through what are some of

19:50

those. And then thinking about access

19:53

to the kinds of food that people want.

19:56

You know, people from different cultures value

19:59

and are familiar. with different kinds

20:01

of foods. And that's

20:03

something that's been really challenging to

20:05

know is how do we know

20:07

what people in an individual community

20:09

need and then

20:11

how might we source and then

20:13

distribute to meet those needs. Since

20:16

joining Feeding America in 2011, Stephanie

20:18

has made progress. She

20:20

developed better approaches to collecting, standardizing

20:22

and analyzing data, bringing new insights

20:24

around staffing food banks and where

20:27

to send food. But

20:29

even a large nonprofit like Feeding America

20:31

can't invest in a large data science

20:33

team. So it started

20:35

partnering with Prem Ramaswamy of Google's

20:37

Data Commons, who we met in

20:39

the first episode. Data Commons

20:41

is an open source project which

20:44

aims to make all the world's

20:46

public data universally accessible and useful.

20:49

My job is to understand and

20:51

empathize with user needs and hopefully

20:53

help solve those problems usually through

20:56

online technology products and tools. And

20:59

when you started combining the public data

21:01

sets under Data Commons with data from

21:03

Feeding America, what did you find? For

21:05

Feeding America, they have this concept called

21:07

the meal gap index. For every census

21:09

tract in the US, so like every

21:11

county is made up of multiple census

21:14

tracts, like hundreds of census tracts, they

21:16

have the concept of a

21:19

meal gap index, which is how hungry are

21:21

people in that county? What is the food

21:23

need in that county versus the ability to

21:25

get that food? So you can imagine

21:27

that like, it's hard to say what food need

21:29

is based on something like poverty, or

21:32

the poverty level, because the poverty level is one

21:34

number across the US, but that might look different

21:36

if you're in New York City versus in a

21:38

farming town, where the cost of

21:40

goods are very different. And so they

21:43

have this concept of the meal gap index,

21:45

it's their proprietary number. And

21:47

once they converted that number into the Data

21:49

Commons format, they're actually able to compare that

21:52

with all the other stuff that's already in

21:54

Data Commons. Now, this is a really cool

21:56

point about Data Commons, because it brings about

21:58

network effects of data. And what

22:00

they were able to do with this is

22:02

quickly see, for all the

22:04

counties in the US, what is

22:07

the meal gap index in one axis? And on the

22:09

other axis, what if I put the

22:11

IPCC temperature models, right? Like, is it gonna

22:13

be greater or less than two degrees centigrade

22:15

and by how much? And

22:17

if I look at that on a graph, what you find is all

22:20

the counties that are gonna have greater than

22:22

two degrees centigrade are actually already experiencing a

22:24

lot of food insecurity. And then you dig

22:26

in a level deeper and you see a

22:28

lot of those counties are farming in agricultural

22:30

counties today. And this sort of

22:33

makes sense as a story now, the farming

22:35

in agricultural counties are already suffering from the

22:37

weather events of climate change, which is affecting

22:39

their economics, which is affecting their employees, which

22:42

is causing more hunger in those places. And

22:44

so you're seeing these walk through, now how

22:46

do you take that from there to

22:48

convincing a lot of those folks about the policy

22:51

actions that need to happen to help their constituents?

22:53

Data Commons collects and standardizes the

22:56

world's public data. And with

22:58

generative AI, users can ask

23:00

questions of that data, democratizing data

23:02

analytics for any size organization. In

23:05

the case of Feeding America, it can

23:07

answer critical questions about food distribution. So

23:16

I know you're still early in your partnership with

23:18

Data Commons, but does it feel like a new

23:20

world is opening up? Yes, it

23:23

definitely does. At first, the big new world was,

23:25

oh my gosh, we just have access to daily

23:27

data and that felt like a big deal.

23:30

But this whole, you know, Gen

23:32

AI world is going to

23:34

evolve. So I feel like in

23:36

some ways we're all a little bit along for the ride in that

23:39

journey, but I'm really excited for the

23:41

possibility. And what are those

23:44

possibilities for Feeding America? Going

23:46

back to the disruptions during COVID, can you

23:48

give an example of how it might make

23:50

food supply chains more transparent? I

23:52

know you like to use mangoes as an example. Sure,

23:55

so a lot of the data that we have

23:57

right now is at a

23:59

higher cap. categorized level. So we

24:01

are tracking how much

24:04

produce overall a food bank

24:06

may source and distribute.

24:09

But you can imagine mangoes are different than

24:11

watermelons are different than lettuce and different than

24:13

potatoes. And so having some

24:15

visibility into mangoes would be really helpful

24:18

in the case where we

24:20

have a mango farmer that we're working with or

24:22

would like to work with us. But

24:24

we don't know where is that food needed.

24:26

Are there a number of food banks who

24:28

already have a fair amount of mangoes and

24:31

they aren't in need of any more?

24:34

What about the cultural considerations for mangoes?

24:36

Not everybody might prefer to have mangoes.

24:38

And so we want to go approach

24:40

mango farmers, for example, and say we'd

24:42

like to be able to take and

24:44

work with you to take the extra

24:46

product and bring that to people facing

24:48

hunger. But to know the

24:51

how and the where and the when and

24:53

the transportation, you know, these are

24:55

the challenges that we face with

24:57

a lack of specificity of data. So we're working

24:59

with our Google partners to explore how might we

25:01

use some of this new technology

25:04

with machine learning and

25:06

generative AI to help

25:08

us categorize that those

25:11

freeform text descriptions and,

25:13

you know, for millions of transactions

25:15

in a given year. And how can we process

25:18

those to make the visibility of that

25:20

data usable? If I

25:22

sent somebody some type

25:25

of analysis today or a resource or

25:27

a dashboard or some kind of report, and they

25:29

had to sift through all of the various versions

25:32

of tomatoes or milk or whatever it is, I'm

25:36

sure people might start throwing tomatoes back at me. If you

25:42

can make it easier rather than having to sift through a

25:44

ton of data, I think we'll all be

25:46

a lot more data informed in our in

25:48

our work. I do hope

25:50

that the world's data in data

25:52

commons means that this previous

25:55

world we lived in where you had to

25:57

spend a lot of money to be able

25:59

to do anything. with data is no

26:01

longer. I want to make sure that

26:03

you don't have a competitive advantage just because you

26:05

happen to have a large amount of data to

26:07

start with. By making that data

26:09

available open source off the bat, I think

26:11

we're leveling the playing field for the world

26:14

to create these new algorithms, these new

26:17

machine learning algorithms, these large language models

26:20

in your localized context for your

26:22

specific problem. How do we help

26:25

food banks across the country service

26:27

their clientele better? This

26:29

is an example of us creating a project

26:31

where we didn't have to worry about the

26:33

data center, the storage, the

26:35

fundamental elasticity to be able

26:37

to set up multiple machines

26:39

quickly, which allows a very,

26:42

very tiny team to

26:44

do this level of scaled work. At this

26:46

point, the number of data points we have

26:48

in Data Commons is over 250 billion

26:51

data points. One of the largest databases

26:53

in the world is the Federal Reserve

26:55

Economic Database in St. Louis. This is

26:57

five times the size of that. And

27:00

we're not even like hiccuping yet in

27:02

terms of scale. The

27:04

speed to which people will have access to

27:06

data is going to be really incredible. That

27:09

excites me because if we can really serve

27:12

people more quickly, more effectively for

27:14

what they need and want, I

27:16

think that makes the work that

27:18

we do so important. There's a

27:21

huge amount of data challenges in

27:23

the food system. And so I'm

27:25

very, very excited about what's possible.

27:28

Emily Ma is the Google sustainability expert

27:30

we heard from earlier. Emily

27:32

has a background in robotics. Over

27:35

the last decade, she worked on many

27:37

different projects inside Google, from kite-based wind

27:39

energy to net zero buildings to Google

27:42

Glass. And in 2013,

27:44

while running global operations for Google

27:46

Glass, she started thinking more about

27:48

food systems. I was

27:51

running Google Glass operations, deploying the

27:53

first 50,000 units of Google Glass.

27:56

And the Google food team pinged

27:58

me and said, hey, what do

28:00

you think about Google Glass in farmer's

28:03

fields? Or what do you think about

28:05

Google Glass in the back kitchen of,

28:07

you know, shade panisse, right? Like what

28:09

could we do with this incredible technology?

28:11

And I met all

28:13

these folks that were really curious about

28:15

technology and how it interplayed with food

28:17

and ag. And, you know, a

28:19

lot of other friends came to me and

28:21

asked, well, is there something we can do

28:23

with innovation and information to help

28:26

tackle this? And we

28:28

had to make a run at it. Over

28:32

the next five years, Emily spent a

28:34

lot of time with food sustainability advocates

28:36

and other technologists who were passionate about

28:39

slashing waste and increasing food access. The

28:42

more she immersed herself, the bigger the opportunity

28:44

felt. It's interesting how

28:46

interconnected the food system is. And, you

28:48

know, hat tip to you, Chef Andrew

28:50

Zimmern, you taught me that the food

28:53

system is like a Mobius strip. So

28:55

you could start in food waste and end

28:57

up in food justice. You can start in

28:59

food justice and you can end up in,

29:01

you know, climate. You can start in climate.

29:03

You can end up in data and technology.

29:05

And so we're kind of very much interconnected

29:07

and we have to allow for these topics

29:09

to intersect and have those

29:11

fluid conversations with everybody who participates

29:13

in the food system, which ultimately

29:16

is all 8 billion people on earth. That

29:18

research led to Project Delta, a project

29:20

that started at X, which used computer

29:22

vision and machine learning to analyze the

29:24

food being tossed out in Google facilities.

29:27

By 2023, their efforts

29:29

had saved millions of pounds from the landfill.

29:31

That project later expanded with a

29:33

mission of organizing the world's food

29:36

information to help grocery stores, food

29:38

distributors, and food banks cut waste

29:40

in their supply chains. And

29:42

as Emily got deeper into the project,

29:44

she discovered that the solutions, at

29:46

least within the walls of Google, were both

29:48

high tech and low tech. So

29:50

I know this is a technology talk, but

29:52

at the end of the day, especially when

29:55

it comes to climate, technology

29:57

is a supporting actor. It is

29:59

not necessary. the solution. And

30:02

I kid you not, the best solution to

30:04

reducing food waste is to make the spoons

30:06

smaller. Really? I love that.

30:08

I know. It's or make the bowl

30:10

shallower. It's pretty interesting. The bowl I

30:12

have, I can only hold one ladle

30:14

of soup. And if I want more,

30:16

I can always go back and get

30:18

a second ladle, but it forces me

30:20

to finish that bowl first. And that

30:23

is what makes all the difference. So

30:25

I think as we sort of look at

30:27

the next 10 years, 20 years, 30 years

30:30

of humanity's biggest challenges,

30:32

I think as technologists,

30:34

we do have to be really thoughtful

30:36

about what our role is and

30:38

to be really, really good stewards

30:40

of a broad array of solutions. Some of them

30:42

will include data and tech and some of them will

30:45

not. Fascinating. So what about the

30:47

high tech solutions? Where did you see the

30:49

most impact there? This is

30:52

so interesting. I think in 2017, 2018, when I started digging

30:55

around with computer vision and food

30:57

waste, gosh, like some of the things that we

30:59

worked on back then are still mind

31:01

blowing to me. Like we can tell whether a

31:03

strawberry has a defect before

31:06

a human eye can see it like

31:08

four or five days in advance. Oh, that's wild. How

31:10

do you do that? It is

31:12

called hyperspectral imaging. We used artificial intelligence

31:14

to figure out the right frequencies of

31:17

light to illuminate a

31:19

strawberry so that you can see

31:21

the part that is rotting faster and

31:24

better with greater contrast. You can only

31:26

do that when you have incredible amounts

31:28

of compute. That was a very fun

31:30

project. And what about

31:32

the use of generative AI? How has that come into

31:34

play for food? I am

31:36

so tickled by the

31:38

last 12 months of gen

31:41

AI and how just

31:43

the world of gen AI has

31:45

unlocked the creativity of so many

31:47

people, including all these incredible

31:49

folks who are trying to feed the country.

31:51

I was touring the food

31:53

bank in Houston and they

31:55

were packing up food for individual

31:58

neighbors. So they called. their customers'

32:00

neighbors, and neighbors can

32:02

order their groceries and come pick them

32:04

up. It was interesting. I

32:07

took a picture of a basket of food, and

32:10

I said, based on what you

32:12

see here, can you give me

32:14

three low-sodium recipes that are culturally

32:16

appropriate for someone from Ethiopia? And

32:18

it did. It's like, holy crap,

32:20

right? That is so cool, because the

32:23

part that's hardest, especially in hunger relief,

32:25

is providing people with the

32:27

dignity of choice. And

32:29

then also, even with ingredients, it doesn't mean

32:32

that they're well-nourished, right? Because if

32:34

they don't eat the food and they don't like the food,

32:36

they're not going to benefit

32:38

from that social

32:40

service. And so my

32:42

hope for the world is everybody really,

32:45

truly gets to eat the

32:47

food that they want to eat that helps

32:50

them be a better human being and

32:52

achieve their goals and to achieve their

32:54

dreams. And it's tools like Gen. AI

32:56

that actually start to do that, not

32:58

just for food bankers, but for everyone,

33:01

right? That's

33:03

such a fun example. We've also heard

33:06

about how AI is helping food banks and

33:08

grocery stores track vast amounts of food and

33:10

helping us grow more crops with fewer resources.

33:13

So for the solutions that do

33:15

require sophisticated technology, how do you

33:17

think about the underlying computing infrastructure

33:19

that makes it possible? We

33:22

are incredibly fortunate to

33:24

be living in this era. It's a

33:26

little bit like, you know, at the

33:28

beginning of every meal, the Japanese will

33:30

say, itadakimasu. And that means

33:32

that they are acknowledging

33:35

everything that went into the meal in

33:38

front of them, right? And so whether

33:40

it was the farmer or the fisherman

33:42

who caught the fish or produced

33:44

the fish, the truck driver who had

33:47

to drive it to where it needed to be, the

33:49

kitchen staff who had to prepare it, the chef

33:51

who prepared it, and then the waiter who presented

33:53

it, or, you know, the mother

33:55

who put it in front of the child,

33:57

right? They say it because they acknowledge everything

34:00

behind the scenes. that's not necessarily apparent. Maybe

34:02

sort of in a poetic

34:04

way, I wish we had the

34:06

same kind of reverence for

34:10

the technical infrastructure behind the internet

34:12

because it truly is astounding. It

34:14

took three, four

34:16

decades of science and engineering to

34:19

get to the point where our data

34:21

centers operate the way they do with

34:23

things like tensor processing units and GPUs

34:26

that operate efficiently using

34:28

as little water and energy

34:30

as possible. Like, it really truly is

34:33

magical. I would love to see people

34:35

be able to peek behind the hood and be like,

34:37

wow, this is the engine

34:39

that gives us this magic. That's

34:53

it for the third episode of this season of Where

34:55

the Internet Lives. Thanks for sharing the magic with us.

34:58

Coming up in our fourth episode, we'll

35:00

hear from scientists and public health officials

35:02

who are using AI to overcome invisible

35:05

threats like extreme heat and air pollution.

35:08

And we'll hear about a satellite that

35:10

could revolutionize our understanding of methane leaks.

35:13

In this episode, you heard from

35:15

Emily Ma, Dwayne Roath, Astro Teller,

35:17

Stephanie Zydek, and Prem Ramaswamy. Where

35:20

the Internet Lives is produced by Latitude Media

35:23

in collaboration with Google. You

35:25

can subscribe to the show anywhere you access your

35:27

podcasts, and please give us a rating if you

35:29

are enjoying our journey together. If

35:31

you want to learn more about how Google's data

35:34

centers are benefiting communities around the world, click the

35:36

link in the show notes. You

35:38

can also spend more time with Stephanie Zydek

35:40

as she works with data and food distribution

35:43

at a food pantry in our mini documentary.

35:46

I'm Stephanie Wong, thank you for listening. Thank

35:58

you.