WEBVTT

37f3c81b-497c-477b-9942-2b3e8098f8c3-0
00:00:00.280 --> 00:00:01.440
Kate, Rich, welcome.

da7ed559-e464-423d-834f-babe9cc0eb81-0
00:00:01.440 --> 00:00:02.760
Great to see you both again.

78baacfe-4ce8-472d-9756-81fd07ae859c-0
00:00:03.080 --> 00:00:03.320
Kate, to you first.

8f52c595-44e0-4ffc-8e0f-b55404a5898d-0
00:00:03.320 --> 00:00:07.504
Talk to me, if you would, about
the role that AI is playing, and

8f52c595-44e0-4ffc-8e0f-b55404a5898d-1
00:00:07.504 --> 00:00:10.080
that it can play, in the climate
fight.

0c0b2fcd-6520-4716-942b-8e74897380e5-0
00:00:10.800 --> 00:00:11.040
Yes.

e9d2bf8c-14af-496e-bcce-9d979b5908a1-0
00:00:11.040 --> 00:00:14.553
Rich and I, and our teams,
worked together on a piece of

e9d2bf8c-14af-496e-bcce-9d979b5908a1-1
00:00:14.553 --> 00:00:18.005
research that we actually
published just before the COP

e9d2bf8c-14af-496e-bcce-9d979b5908a1-2
00:00:18.005 --> 00:00:21.703
kicked off, looking at the
opportunity for AI to accelerate

e9d2bf8c-14af-496e-bcce-9d979b5908a1-3
00:00:21.703 --> 00:00:25.093
existing technology and
solutions that we have in this

e9d2bf8c-14af-496e-bcce-9d979b5908a1-4
00:00:25.093 --> 00:00:26.080
decisive decade.

284c09cb-c02c-4f2a-ae14-3fdd83fc382c-0
00:00:26.320 --> 00:00:31.001
The headline from the study was:
by 2030, AI could enable 5-10%

284c09cb-c02c-4f2a-ae14-3fdd83fc382c-1
00:00:31.001 --> 00:00:35.390
emissions reductions globally,
again, by accelerating those

284c09cb-c02c-4f2a-ae14-3fdd83fc382c-2
00:00:35.390 --> 00:00:37.000
existing technologies.

12757778-6350-4cfa-8169-987c5cdb0a03-0
00:00:37.120 --> 00:00:39.709
To put that into context for
you, that's roughly the

12757778-6350-4cfa-8169-987c5cdb0a03-1
00:00:39.709 --> 00:00:42.640
equivalent of the emissions of
the European Union, so a big

12757778-6350-4cfa-8169-987c5cdb0a03-2
00:00:42.640 --> 00:00:43.520
opportunity space.

8cf8c42a-c811-483a-8ea1-ff983208fa44-0
00:00:43.800 --> 00:00:47.320
And we also identified three
areas where we see promise.

364ff389-01b4-4cf9-b32b-934880be26a0-0
00:00:47.480 --> 00:00:51.366
The first is around information,
using AI to better organize data

364ff389-01b4-4cf9-b32b-934880be26a0-1
00:00:51.366 --> 00:00:53.840
and provide insights that are
actionable.

a9e4e4c3-9f64-449d-9348-408e77bf14e4-0
00:00:54.000 --> 00:00:57.450
The second, optimization. This
is the real superpower of

a9e4e4c3-9f64-449d-9348-408e77bf14e4-1
00:00:57.450 --> 00:00:58.480
machine learning.

c59f5a34-8127-4464-8651-a2ab54c00efb-0
00:00:58.680 --> 00:01:01.825
How do we use optimization to
reduce energy use, to reduce the

c59f5a34-8127-4464-8651-a2ab54c00efb-1
00:01:01.825 --> 00:01:04.871
use of natural resources, which
also has a big double bottom

c59f5a34-8127-4464-8651-a2ab54c00efb-2
00:01:04.871 --> 00:01:05.920
line incentive to it.

d50f5a36-8aed-4fa6-8bae-f421dd3e2c50-0
00:01:06.200 --> 00:01:09.444
And then the last is around
prediction, whether that's

d50f5a36-8aed-4fa6-8bae-f421dd3e2c50-1
00:01:09.444 --> 00:01:12.747
predictions around how wind
might come onto the grid 36

d50f5a36-8aed-4fa6-8bae-f421dd3e2c50-2
00:01:12.747 --> 00:01:16.168
hours in advance, but also in
the space of adaptation and

d50f5a36-8aed-4fa6-8bae-f421dd3e2c50-3
00:01:16.168 --> 00:01:19.589
resilience, giving us better
insights into flooding, into

d50f5a36-8aed-4fa6-8bae-f421dd3e2c50-4
00:01:19.589 --> 00:01:20.119
wildfire.

6aa81cc0-f3e6-4872-aff7-8beb57c9acb2-0
00:01:20.320 --> 00:01:23.195
So, a real basket of solutions
that, again, are here today,

6aa81cc0-f3e6-4872-aff7-8beb57c9acb2-1
00:01:23.195 --> 00:01:25.880
they're tangible, and they can
accelerate our progress.

9f962dd5-4fab-4b02-9d09-691ea9e5927c-0
00:01:26.320 --> 00:01:29.473
So, a lot of opportunities,
Rich. The solutions, optimism

9f962dd5-4fab-4b02-9d09-691ea9e5927c-1
00:01:29.473 --> 00:01:31.920
around this, but how can
businesses identify

0a252621-214e-4ae0-b77f-c69d952f3087-0
00:01:31.920 --> 00:01:34.964
and then use AI to get the most
to actually achieve their

0a252621-214e-4ae0-b77f-c69d952f3087-1
00:01:34.964 --> 00:01:38.113
climate ambitions? It's not just
a question of grabbing AI,

0a252621-214e-4ae0-b77f-c69d952f3087-2
00:01:38.113 --> 00:01:40.160
chucking loads of money at it,
surely.

fb4d8c36-a96a-44f8-905f-25977e2d9090-0
00:01:40.600 --> 00:01:41.600
Well, it's a couple of elements.

fffc7810-49e4-4b56-8b17-87206592b967-0
00:01:41.640 --> 00:01:44.844
First, it's really critical to
pick the critical opportunity

fffc7810-49e4-4b56-8b17-87206592b967-1
00:01:44.844 --> 00:01:45.160
areas.

143edd0b-cd34-41ac-a808-6f219d9856b8-0
00:01:45.920 --> 00:01:48.360
The phrase we use internally is
'outcomes at scale'.

b98d2bb2-9c00-4a2e-b2c0-d9f14af4543d-0
00:01:48.360 --> 00:01:51.494
What are the places that, if you
put it in, and you can make it

b98d2bb2-9c00-4a2e-b2c0-d9f14af4543d-1
00:01:51.494 --> 00:01:53.600
work, you can really drive broad
outcomes.

612c683d-2c29-4205-8c96-5191a8210f43-0
00:01:54.000 --> 00:01:57.761
The second is a huge agenda on
people and how do you bring

612c683d-2c29-4205-8c96-5191a8210f43-1
00:01:57.761 --> 00:02:01.842
people along so they understand,
there's reskill, organizations

612c683d-2c29-4205-8c96-5191a8210f43-2
00:02:01.842 --> 00:02:02.480
are built.

2912c778-26a8-4ef3-89f3-020b2a2c8547-0
00:02:02.720 --> 00:02:05.905
But the two points I'd like to
quickly highlight are data and

2912c778-26a8-4ef3-89f3-020b2a2c8547-1
00:02:05.905 --> 00:02:09.192
partnerships that I think have
been so integral to this. Rather

2912c778-26a8-4ef3-89f3-020b2a2c8547-2
00:02:09.192 --> 00:02:12.275
than talking about them in
theory, I'll give just two quick

2912c778-26a8-4ef3-89f3-020b2a2c8547-3
00:02:12.275 --> 00:02:13.199
examples. On data,

16974d67-a2a9-448a-a8d2-5a6694af7ef4-0
00:02:14.280 --> 00:02:18.677
we built CO2 AI, which was a
platform, that is now a SAAS

16974d67-a2a9-448a-a8d2-5a6694af7ef4-1
00:02:18.677 --> 00:02:19.360
platform.

55faf215-213a-4f5f-b335-5a952b8cf1b5-0
00:02:19.520 --> 00:02:23.376
It is its own little company now
that's out there with the best

55faf215-213a-4f5f-b335-5a952b8cf1b5-1
00:02:23.376 --> 00:02:26.810
data in the world about
emissions, so that you can track

55faf215-213a-4f5f-b335-5a952b8cf1b5-2
00:02:26.810 --> 00:02:30.486
emissions not just in your own
organization, where you often

55faf215-213a-4f5f-b335-5a952b8cf1b5-3
00:02:30.486 --> 00:02:33.680
have good data, but across your
entire supply chain.

eed3ff74-c18d-4a18-b79f-64b96f5d7851-0
00:02:33.880 --> 00:02:35.680
We worked with a healthcare
company.

b8f0c4ba-11c2-45ba-823f-09de9ed6d199-0
00:02:35.840 --> 00:02:41.487
They worked with a healthcare
company that could use the 50

b8f0c4ba-11c2-45ba-823f-09de9ed6d199-1
00:02:41.487 --> 00:02:46.663
times more insight and data to
get to 20% more precise

b8f0c4ba-11c2-45ba-823f-09de9ed6d199-2
00:02:46.663 --> 00:02:51.840
predictions and exceed very
ambitious targets by 120%.

1ab4c46f-b078-467d-a12f-5e625e42d6d9-0
00:02:52.120 --> 00:02:55.501
The other example is one from
Google. It's a personal favorite

1ab4c46f-b078-467d-a12f-5e625e42d6d9-1
00:02:55.501 --> 00:02:58.721
because I fly so much. They
teamed with Breakthrough Energy

1ab4c46f-b078-467d-a12f-5e625e42d6d9-2
00:02:58.721 --> 00:03:00.600
and American to take on
contrails.

97267f04-9245-48aa-a3ed-e0027ae6a83f-0
00:03:00.600 --> 00:03:04.442
People don't realize, but the
little lines we see in the sky

97267f04-9245-48aa-a3ed-e0027ae6a83f-1
00:03:04.442 --> 00:03:08.221
coming out the back of planes
are responsible for more than

97267f04-9245-48aa-a3ed-e0027ae6a83f-2
00:03:08.221 --> 00:03:11.560
30% of the emissions, the GHG
impact of an airplane.

2392cc03-c412-413c-a137-41f498ec0a80-0
00:03:11.720 --> 00:03:13.920
When I first learned this a few
years ago I was shocked.

73c1f8f0-4ec6-46ca-a24c-6e45f98bdb1c-0
00:03:14.320 --> 00:03:17.720
And that's actually one we can
really tackle quickly.

05e79ff9-3bb4-48b9-8156-72b46f4e9cc9-0
00:03:17.720 --> 00:03:21.880
It doesn't require a whole new
technology supply and using AI,

c6003feb-a0ea-4623-ab05-b612e9a7940c-0
00:03:21.880 --> 00:03:26.055
they worked with Breakthrough
and American to design new tools

c6003feb-a0ea-4623-ab05-b612e9a7940c-1
00:03:26.055 --> 00:03:30.032
for pilots to be able to use
that seem to be able to reduce

c6003feb-a0ea-4623-ab05-b612e9a7940c-2
00:03:30.032 --> 00:03:33.280
those emissions, the contrail
part, by over 50%.

ca645e98-dcac-4f3a-a654-9a3cdaa92d3d-0
00:03:33.440 --> 00:03:35.320
It's incredible and it's fast.

f79974aa-bd0b-4311-8079-181f326e792a-0
00:03:35.800 --> 00:03:39.568
Maybe just to pick up on this, I
think back to this theme of AI

f79974aa-bd0b-4311-8079-181f326e792a-1
00:03:39.568 --> 00:03:43.454
as an accelerant, and being able
to use it on top of our existing

f79974aa-bd0b-4311-8079-181f326e792a-2
00:03:43.454 --> 00:03:45.280
technology stack and solutions.

8a46a73a-5820-47b7-b603-14d6283149f2-0
00:03:45.280 --> 00:03:47.160
So I love the contrails example.

d067e830-1ec9-4b13-bbf8-00300b1bed51-0
00:03:47.360 --> 00:03:49.954
The other one I would highlight
is work that we've been doing

d067e830-1ec9-4b13-bbf8-00300b1bed51-1
00:03:49.954 --> 00:03:51.000
that we call Green Light.

82e587a4-eda6-4375-88b4-114199eeb131-0
00:03:51.160 --> 00:03:54.167
This is an initiative where our
Google research team has been

82e587a4-eda6-4375-88b4-114199eeb131-1
00:03:54.167 --> 00:03:56.980
working with city traffic
engineers to give them insights

82e587a4-eda6-4375-88b4-114199eeb131-2
00:03:56.980 --> 00:03:59.599
into how to better sequence
traffic lights in a city.

885c3e42-1805-4f18-8846-44aa1d39d12f-0
00:03:59.760 --> 00:04:03.224
Because what we know is, not
only is it really annoying to

885c3e42-1805-4f18-8846-44aa1d39d12f-1
00:04:03.224 --> 00:04:06.806
sit at red lights, there's 29
times more emissions happening

885c3e42-1805-4f18-8846-44aa1d39d12f-2
00:04:06.806 --> 00:04:08.920
when cars are idling at red
lights.

2d94ceff-f719-4eed-a80a-367b6dd391df-0
00:04:09.240 --> 00:04:12.372
So if you can create that Green
Wave effect that we all love,

2d94ceff-f719-4eed-a80a-367b6dd391df-1
00:04:12.372 --> 00:04:15.605
where you feel like you're just
hitting green light after green

2d94ceff-f719-4eed-a80a-367b6dd391df-2
00:04:15.605 --> 00:04:18.838
light, that is not only a better
experience as a driver, but it

2d94ceff-f719-4eed-a80a-367b6dd391df-3
00:04:18.838 --> 00:04:20.959
also reduces air pollution and
emissions.

094f1eaf-a6a1-4fe8-80a1-b6805d76416f-0
00:04:20.960 --> 00:04:24.955
What we've found in tests with
an initial 12 cities is we can

094f1eaf-a6a1-4fe8-80a1-b6805d76416f-1
00:04:24.955 --> 00:04:28.822
reduce starting and stopping
events by 30% and emissions by

094f1eaf-a6a1-4fe8-80a1-b6805d76416f-2
00:04:28.822 --> 00:04:29.080
10%.

090885a4-47d7-48db-8b2e-f2354e963e0a-0
00:04:29.240 --> 00:04:33.083
So we're doing that in
Bangalore, in Rio, in Haifa, and

090885a4-47d7-48db-8b2e-f2354e963e0a-1
00:04:33.083 --> 00:04:35.280
coming to many more cities soon.

440ea688-8728-4ad4-af14-8d14d0be51f5-0
00:04:35.720 --> 00:04:40.880
The joy we feel when we run the
green lights, we can also feel

440ea688-8728-4ad4-af14-8d14d0be51f5-1
00:04:40.880 --> 00:04:46.040
climate joy, as well as joy for
just getting somewhere faster.

8d45a984-eb46-4cf7-9696-22bb7b2aca92-0
00:04:46.040 --> 00:04:46.720
Obviously there's so much
excitement and

1cf4af5f-296b-479b-91fe-bc46b58a33ed-0
00:04:46.720 --> 00:04:49.493
there are lots of opportunities.
I don't want to sort of bring a

1cf4af5f-296b-479b-91fe-bc46b58a33ed-1
00:04:49.493 --> 00:04:52.266
downer on the conversation, but
we do have to identify the risks

1cf4af5f-296b-479b-91fe-bc46b58a33ed-2
00:04:52.266 --> 00:04:52.479
here.

71c51b83-339f-4b3e-961c-6d5ae775329e-0
00:04:52.480 --> 00:04:54.160
What are the potential risks of
AI?

3d9ede3b-1e37-41b7-b0d6-81559a4f97f3-0
00:04:54.560 --> 00:04:59.291
How can you overcome them as
leaders, or at least keep them

3d9ede3b-1e37-41b7-b0d6-81559a4f97f3-1
00:04:59.291 --> 00:05:00.080
in check? 

edd30117-55b5-496b-aa36-748d87816238-0
00:05:00.080 --> 00:05:02.550
Indeed. With this paper, we
certainly focused on the

edd30117-55b5-496b-aa36-748d87816238-1
00:05:02.550 --> 00:05:05.254
opportunity space and these
real, tangible, examples that

edd30117-55b5-496b-aa36-748d87816238-2
00:05:05.254 --> 00:05:06.560
Rich and I have been giving.

24dafdbf-f54a-4e98-a235-857b9f8064e9-0
00:05:06.760 --> 00:05:10.073
But also, we thought it was
really important to look at the

24dafdbf-f54a-4e98-a235-857b9f8064e9-1
00:05:10.073 --> 00:05:11.840
environmental impact side of AI.

253c6b36-6688-4766-b61b-0bd472c74b4b-0
00:05:11.840 --> 00:05:14.808
We want to ensure that the
opportunities we're unlocking

253c6b36-6688-4766-b61b-0bd472c74b4b-1
00:05:14.808 --> 00:05:18.193
far outweigh the impact, but we
wanted to look at the literature

253c6b36-6688-4766-b61b-0bd472c74b4b-2
00:05:18.193 --> 00:05:20.120
and really share what we know
today.

aa7eee7c-3a38-4dda-b848-7093bab396ca-0
00:05:20.640 --> 00:05:25.604
What we found is that in 2022,
data centers globally, which

aa7eee7c-3a38-4dda-b848-7093bab396ca-1
00:05:25.604 --> 00:05:30.652
enable these AI solutions, made
up about .1 to .2% of global

aa7eee7c-3a38-4dda-b848-7093bab396ca-2
00:05:30.652 --> 00:05:31.480
emissions.

59a8c965-9899-46c4-bad4-a4515ade59ee-0
00:05:31.720 --> 00:05:35.843
And then from a water
consumption basis in the US,

59a8c965-9899-46c4-bad4-a4515ade59ee-1
00:05:35.843 --> 00:05:39.240
data centers are about .06% of
water use.

4d16839d-8b4e-4b7f-b13f-f57199b66f9e-0
00:05:39.600 --> 00:05:41.480
So it's still a relatively small
footprint.

d3236c19-0eda-4a48-a171-754ad073ec7d-0
00:05:41.720 --> 00:05:44.522
But also we need to be
thoughtful because AI is at an

d3236c19-0eda-4a48-a171-754ad073ec7d-1
00:05:44.522 --> 00:05:47.739
inflection point and what we
found is that there isn't really

d3236c19-0eda-4a48-a171-754ad073ec7d-2
00:05:47.739 --> 00:05:50.852
good information yet on what we
think that future footprint

d3236c19-0eda-4a48-a171-754ad073ec7d-3
00:05:50.852 --> 00:05:51.320
might be.

d6442dea-ff29-4b55-89a1-b76da17f6a6a-0
00:05:51.320 --> 00:05:54.539
So that's an area that requires
more research, but also there

d6442dea-ff29-4b55-89a1-b76da17f6a6a-1
00:05:54.539 --> 00:05:57.291
are tested best practices for
designing data centers

d6442dea-ff29-4b55-89a1-b76da17f6a6a-2
00:05:57.291 --> 00:06:00.146
efficiently, for how we train
and optimize the machine

d6442dea-ff29-4b55-89a1-b76da17f6a6a-3
00:06:00.146 --> 00:06:02.120
learning models, the hardware we
use.

d9f9f0cc-1378-40e1-96ee-4c3345b1c0de-0
00:06:02.280 --> 00:06:04.840
And then of course, crucially,
carbon-free energy.

dd10b1f2-d3ad-4188-af88-eb4a2700d486-0
00:06:05.000 --> 00:06:08.393
We have a goal at Google, that
by 2030, we want to run all of

dd10b1f2-d3ad-4188-af88-eb4a2700d486-1
00:06:08.393 --> 00:06:11.842
our offices and data centers on
24/7 carbon-free energy on all

dd10b1f2-d3ad-4188-af88-eb4a2700d486-2
00:06:11.842 --> 00:06:13.320
the grids where we operate.

bd0a066c-0205-400d-92a9-96b598fd66fe-0
00:06:13.320 --> 00:06:15.040
We're 64% of the way there.

e2ddfeed-f38d-4913-9806-7ec90d91875a-0
00:06:15.320 --> 00:06:17.360
So we need to be thoughtful
about the impact as well.

1956fa7b-937c-4cb1-9539-adfc61ca0c64-0
00:06:18.000 --> 00:06:21.003
I also want to highlight this
because when people talk risk,

1956fa7b-937c-4cb1-9539-adfc61ca0c64-1
00:06:21.003 --> 00:06:24.006
they tend to think about the
downsides of AI that we need to

1956fa7b-937c-4cb1-9539-adfc61ca0c64-2
00:06:24.006 --> 00:06:24.400
manage, 

b1e983c5-4df8-432b-b077-ce8d63fd0739-0
00:06:24.400 --> 00:06:27.957
as Kate said. AI can make an
enormous difference on

b1e983c5-4df8-432b-b077-ce8d63fd0739-1
00:06:27.957 --> 00:06:29.120
addressing risks.

c8560be0-c23f-4f25-af45-c2dec67dc4e5-0
00:06:29.320 --> 00:06:32.453
I mean, we think about what's
coming, not just for mitigation,

c8560be0-c23f-4f25-af45-c2dec67dc4e5-1
00:06:32.453 --> 00:06:33.200
but adaptation.

743ccf40-6776-48ca-b77e-4845cc63cad4-0
00:06:33.640 --> 00:06:36.946
There's a wonderful example from
Google on identifying floods

743ccf40-6776-48ca-b77e-4845cc63cad4-1
00:06:36.946 --> 00:06:39.613
before they happen to give
people time to protect

743ccf40-6776-48ca-b77e-4845cc63cad4-2
00:06:39.613 --> 00:06:40.200
themselves.

862ae6c8-6f73-4da1-9466-595e55191e3e-0
00:06:40.440 --> 00:06:44.507
We've been working with cities
around the world, like Lagos, on

862ae6c8-6f73-4da1-9466-595e55191e3e-1
00:06:44.507 --> 00:06:48.321
how they use AI to optimize
their investments to reduce the

862ae6c8-6f73-4da1-9466-595e55191e3e-2
00:06:48.321 --> 00:06:49.720
risk of climate impact

f3081fdd-6766-4cae-a1a2-53387db87290-0
00:06:49.720 --> 00:06:50.400
down the road.

46522854-7eed-4df8-8747-07cb487b7bef-0
00:06:50.560 --> 00:06:54.011
There'll be tons of opportunity
to help farmers reduce risk in

46522854-7eed-4df8-8747-07cb487b7bef-1
00:06:54.011 --> 00:06:55.600
their agriculture operations.

8553eed4-51f6-4e66-bfc3-de02601faeea-0
00:06:55.840 --> 00:06:56.120


ccb0873a-73ed-4227-a092-4199e3e134bf-0
00:06:56.120 --> 00:07:01.744
There are negative risks to
manage, but there are enormous

ccb0873a-73ed-4227-a092-4199e3e134bf-1
00:07:01.744 --> 00:07:07.560
risks from climate that AI will
be critical to help address.

b436423e-c05d-421a-b783-65306a8b5cda-0
00:07:07.560 --> 00:07:08.040
Thank you so much.

9e29c59f-0fac-47f7-ac6d-c1b0520896ac-0
00:07:08.040 --> 00:07:12.480
Thank you, Georgie.