0:00

For most businesses looking to

0:02

create value, I would say look through

0:04

the applications and whether the unique things

0:06

you can build with AI that were

0:09

not possible just one or two years

0:11

ago. Welcome to Radio Davos, the podcast

0:13

from the World Economic Forum that looks

0:16

at the biggest challenges and how we

0:18

might solve them. This week, we're looking

0:20

at how businesses are deploying artificial intelligence.

0:23

We're seeing a real shift from

0:25

companies experimenting with AI in small

0:27

pilots to actually scaling it across

0:30

their businesses. The World Economic Forum

0:32

has looked at how AI is

0:34

being used across a wide range

0:36

of sectors and has published the

0:38

findings to help companies deploy the

0:40

technology safely, responsibly and profitably. If

0:42

you walk in a factory that's

0:44

truly advanced today, you'll have autonomously

0:46

guided vehicles everywhere, delivering different pieces

0:48

just in time for when that

0:50

specific part is needed, those are

0:52

all great examples of how they're

0:54

using machine learning AI and AI

0:56

agents on the factory floor. We hear

0:59

from one of the world's most influential

1:01

AI researchers, and from Siemens, a company

1:03

that makes the brains for other manufacturers

1:05

to use AI in their factories. The

1:07

advantage you have on going digital and

1:09

AI is that you can... Learn from

1:11

your mistakes way faster. You run scenarios.

1:13

Not one or ten or hundred times,

1:15

but thousands or millions of times. I'm

1:17

Robin Pomerote, the World Economic Forum. And

1:19

with this look at how AI is

1:21

changing industry and everyone in it. AI

1:23

was the domain of the nerds in

1:25

the basements, which would come up and

1:27

look at a problem and go back and

1:29

calculate it. Now everybody can. That makes a

1:32

huge difference if you put it in the

1:34

hand of everybody. This is Radio Davos. Welcome

1:37

to Radio Davos where we are once

1:39

again talking about artificial intelligence and I'm

1:42

joined at the start of this episode

1:44

by Kathy Lee. Kathy, how are you?

1:46

I'm good, how are you Robin? Very well, thank

1:48

you. Remind us, Kathy, what your

1:50

important job is at the World Economic

1:52

Forum. I had the AI Data and

1:55

Metiverse work at the World Economic Forum.

1:57

We're going to be talking about these

1:59

white papers. published, most of them

2:01

were published in Davos I think

2:03

in the annual meeting, this series

2:05

is called AI in action beyond

2:07

experimentation to transform industry. Tell us

2:10

what, you know, in a nutshell

2:12

what these papers set out to

2:15

do. In January, we launched this

2:17

whole series, as you just mentioned,

2:19

which takes a deep dive into

2:22

how AI is transforming different industries.

2:24

AI is evolving at an incredible

2:26

pace and with so much changing.

2:29

We wanted to capture where things

2:31

stand today while also keeping an

2:34

eye on what's next. This is

2:36

a 10 paper series developed

2:38

with Accenture, BCG, McKinsey and

2:40

the University of Oxford, covering

2:42

a mix of broad industry

2:44

trends and more focused deep

2:47

dives into specific sectors and

2:49

regions. Right, so there are other

2:51

papers in this series covering, could you

2:53

give us an idea of what they

2:56

cover? Healthcare, financial services, media,

2:58

transport, advanced manufacturing. and

3:01

critical topics such as

3:03

energy, cyber security, and

3:05

AI's industry development in

3:07

China. Industries have been using

3:10

artificial intelligence for many, many years.

3:12

It's not as if it's this

3:14

brand new technology, but for most

3:16

of us, kind of civilians, it

3:18

was the launch of ChatGBT at

3:20

the end of 2022 that really

3:23

brought it to our attention, this

3:25

amazing thing. And there was maybe

3:27

a hype bubble for... you know, the

3:29

rest of that year, 2023. And

3:31

now I think everyone, industries and

3:33

just normal people who have access

3:35

to anything like CHATGBT or any

3:37

of those other things, many of

3:39

which are free, what to actually

3:41

do with this? So you've got

3:44

big, big companies who maybe have

3:46

been using it before, and these

3:48

papers, I suppose, are designed to

3:50

help them get an idea of

3:52

what other people are doing with

3:54

this. evolving technology. You're absolutely right.

3:56

We understand we know that we're

3:58

still in the middle. this AI

4:00

hype cycle, but what's really

4:02

exciting is that we're seeing

4:04

a real shift from companies

4:07

experimenting with AI in small

4:09

pilots to actually scaling it

4:11

across their businesses. The report

4:13

highlights how AI is no

4:15

longer just about efficiency. It's

4:17

becoming a core driver for

4:19

growth and transformation. And some

4:21

of the numbers really stand

4:23

out. Companies that lead in

4:25

AI adoption are the outperforming

4:27

their peers by 15% in

4:29

revenue generation is a figure

4:31

that's projected to more than

4:33

double by 2026. And this

4:35

expected revenue growth could contribute

4:38

some figure between 7.6 to

4:40

17.9 trillion dollars. to the

4:42

global economy by 2030 with

4:44

the upper end of the

4:46

range achieved through people-centric solutions.

4:48

Tell us about this first

4:50

paper then of these 10.

4:52

It's already out there. People

4:54

can get it on our

4:56

website. It's the cross-industry one

4:58

kind of summarising everything. Tell

5:00

us some of the highlights

5:02

from that. The paper was

5:04

developed with Accenture. It takes

5:07

a closer look at how

5:09

AI is moving beyond the

5:11

testing phase and into real-world

5:13

impact across industries. So some

5:15

of the figures, as I

5:17

just called earlier, really demonstrate

5:19

that transformative impact of AI.

5:21

It also dives into what

5:23

it takes to make AI

5:25

transformation work. Things like having

5:27

the right digital infrastructure, strong

5:29

cyber security. and the workforce

5:31

that's ready to work with

5:33

AI. These are the errors

5:35

that will determine which companies

5:38

really get ahead. So it's

5:40

beyond hype and it's getting

5:42

beyond experimentation. Companies are really

5:44

implementing this. Later in the

5:46

program I'll be talking with

5:48

our colleague Kiva All Good

5:50

You Heads, the Advanced Manufacturing

5:52

Center at the World Economic

5:54

Forum, and we'll be hearing

5:56

from some... from the engineering

5:58

company Siemens about how that

6:00

huge company is using. He'll

6:02

be giving us examples of

6:04

what they're doing on the

6:06

ground with artificial intelligence. But

6:09

before that we have another

6:11

interview a colleague of mine

6:13

spoke to Andrew Un. Could

6:15

you tell us something about

6:17

him? He's someone, he's been

6:19

on Radio Davos before, remind

6:21

us who Andrew Un is

6:23

and what he's going to

6:25

talk about? Andrew, we work

6:27

with him very closely. He's

6:29

a legendary AI researcher and

6:31

he runs AI fund now.

6:33

He also was part of

6:35

numerous AI startups, including Caserab,

6:37

which is a household name.

6:40

Andrew Un goes into how

6:42

AI is evolving, not just

6:44

in terms of technical advancements,

6:46

but in how businesses are

6:48

shifting are shifting from us.

6:50

theoretical discussions to real practical

6:52

applications. He highlights how AI

6:54

has been using industries beyond

6:56

tech, from optimizing pricing strategies

6:58

to streamlining legal document processing

7:00

and even facilitating cross-border trade.

7:02

He also talks about a

7:04

key shift happening right now,

7:06

moving beyond the AI hype

7:09

and focusing on tangible business

7:11

value. One of his most

7:13

interesting points is that AI's

7:15

biggest impact isn't just in

7:17

the foundation models we hear

7:19

so much about, but in

7:21

the applications built on top

7:23

of them. Another important takeaway

7:25

is the growing need for

7:27

AI literacy across all professions.

7:29

Andrew makes a strong case

7:31

for why simply being an

7:33

AI user isn't enough anymore.

7:35

He believes that the professionals

7:37

who develop a deeper technical

7:40

understanding will be the one

7:42

who truly maximize AI's potential

7:44

in their views. It's a

7:46

really insightful discussion and I'm

7:48

super excited for listeners to

7:50

dive into it. This is Andrewung

7:52

talking to my colleague Anna Bruce Lockhart

7:54

in Davos in January. Andrewung, managing general

7:57

partner of AI. Here at Davos you

7:59

recently spoke about artificial general intelligence. How

8:01

do you see AI evolving to better

8:04

understand and emulate human reasoning? There's a

8:06

lot of excitement about AI and as

8:08

AI improve this reasoning, the set of

8:11

tasks that AI is able to do

8:13

is expanding. And the reason this is

8:15

exciting is because this also expands the

8:18

set of applications that are now possible

8:20

by building on top of AI. I

8:22

think intelligence and isolation is not that.

8:25

useful is when we find an application

8:27

such as can we use this to

8:29

screw out cross-border trade by improving compliance

8:32

or can we use this to drive

8:34

pricing analytics or can we use this

8:36

to help process legal documents is those

8:38

applications where the value is and so

8:41

as AI becomes more intelligent and more

8:43

able to generate sensible responses the effectiveness

8:45

which we should have built these applications

8:48

also gross. So which industries you believe

8:50

are underutilizing AI? AI is a general

8:52

purpose technology, meaning it's useful not just

8:55

for one or two things, but for

8:57

many different things. And I think adoption

8:59

of AI has been progressing across all

9:02

industries, but the industry is there are

9:04

more digital, are probably ahead. So tech

9:06

is clearly very digital, embracing AI. I

9:09

think finance, financial services, maybe healthcare, these

9:11

are industries, a little bit more digital.

9:13

In contrast, industries such as natural resource

9:16

extraction. It turns out that in the

9:18

last decade, that become quite digital as

9:20

well, but then maybe the IT staffing

9:23

and the DNA and the skill set

9:25

isn't as advanced as financial services, but

9:27

I think compared to 10 years ago,

9:29

all industries are much more digital than

9:32

they have been. So I feel like

9:34

there is adoption of AI free from

9:36

having across all sectors. How can businesses

9:39

use AI to innovate? Where are the

9:41

biggest opportunities? This is what I think

9:43

of as the AI stack. At the

9:46

lowest level is the semiconductors and then

9:48

the cloud companies and then the the

9:50

large AI model trainers or the foundation

9:53

model companies. And it turns out there's

9:55

a lot of excitement in buzz about

9:57

these technology companies, and that's fine, nothing

10:00

wrong with that. It is exciting. Almost

10:02

by definition, though, there's another layer of

10:04

the AI stack that's gonna be even

10:07

more valuable, which is the applications we

10:09

built on top. because frankly we need

10:11

the applications to generate even more revenue

10:14

so they can afford to pay these

10:16

technology providers that the media writes about

10:18

so much. So for most businesses, looking

10:21

to create value, I would say look

10:23

to the applications and whether the unique

10:25

things you can build with AI that

10:27

were not possible just one or two

10:30

years ago. The conversations at web this

10:32

year were very encouraging. Last year there's

10:34

a lot of hype and unnecessary fears

10:37

about what AI calls on? The answer

10:39

is no, by the answer is no,

10:41

by the way. I think this year

10:44

the hype is still a little bit

10:46

there but it's died down substantially but

10:48

I find myself seeing a lot more

10:51

business implementation conversations where businesses will be

10:53

excited about you know AGI and intelligence

10:55

and all of these things but then

10:58

they'll sit down and say okay great

11:00

now let's figure out concretely what work

11:02

needs to be done in order to

11:05

drive a valuable business use case. This

11:07

year's we've had so many businesses looking

11:09

concretely into what to do, that I

11:12

have a prediction for next year's work,

11:14

which is that we'll be back and

11:16

we'll hear a lot more success stories

11:18

from many businesses. And so that feels

11:21

really good to me. Because A. has

11:23

a general purpose technology. is applicable to

11:25

so many different things. So I'm seeing

11:28

it being used, for example, to process

11:30

tricky legal documents and get to conclusions

11:32

faster or to file paperwork to smooth

11:35

out cross-border trade or when the teams

11:37

I'm working with uses it for pricing

11:39

analytics. And it turns out that when

11:42

you get a pricing right, you can

11:44

improve profitability by 20-30% very rapidly. So

11:46

I'm seeing more stories of concrete use

11:49

cases with. business ROI and this is

11:51

encouraging because this is a very welcome

11:53

shift. from some of the hype that

11:56

had been around AI for the past

11:58

couple years. So how can smaller businesses

12:00

leverage AI to become more competitive? Some

12:03

of the hype around AI has been

12:05

on how expensive AI is, and it

12:07

turns out that if you want to

12:10

build a cutting-edge large AI model, that

12:12

is expensive. Plan to spend, you know,

12:14

budget a billion dollars for that or

12:16

something. But it turns out that for

12:19

a lot of applications, because someone else

12:21

has spent billions of dollars training these

12:23

AI models, that you can now access

12:26

for dollars or cents is now actually

12:28

very capital efficient to innovate and try

12:30

new ideas and build these new things.

12:33

My team at AI Fund budgets is

12:35

55,000 US dollars to get a working

12:37

prototype to test out if an idea

12:40

is something we should double down on.

12:42

And so for small businesses and large

12:44

businesses, I think figuring out the right

12:47

corporate innovation process to build productizing experiment

12:49

and go for the upside will be

12:51

important for this next period. You've long

12:54

been an advocate for democratizing AI. What

12:56

are the biggest barriers to achieving this

12:58

and how can they be overcome? One

13:01

of the most important skills for anyone

13:03

in the future is the ability to

13:05

get a computer. to do exactly what

13:07

you wanted to do, because computers become

13:10

more powerful, and not just software engineers,

13:12

but really all knowledge workers, maybe even

13:14

non-knowledge workers, will be much more powerful

13:17

if you can bend computers to our

13:19

will. But it turns out that to

13:21

get that deeper understanding, to have the

13:24

language or the ways to manage your

13:26

computers, takes new training. So Kossera, Deep

13:28

Ground. AI, other organizations are working hard

13:31

to provide this type of upskilling. Anyone

13:33

with kids, I'm often asked, should I

13:35

still get my kids to learn to

13:38

go? My answer is yes, get them

13:40

to learn to go, because I think

13:42

that deeper understanding of how computers work

13:45

will be an important foundation for not

13:47

just the software engineers, but really all

13:49

job roles in the future. Today, I

13:52

can't imagine hiring a marketer, a recruiter,

13:54

a finance person. that doesn't know how

13:56

to search the web. Because this is

13:59

a new capability that came up, makes

14:01

everyone more productive. I think we are

14:03

soon approaching the point. I think we're

14:05

actually already at the point. I can't

14:08

imagine today hiring a market that doesn't

14:10

know how to use AI deeply. And

14:12

I think in the future, I won't

14:15

hire any markets that doesn't know at

14:17

a deeper level almost how to code

14:19

level, how they make use of AI.

14:22

One thing I've been curious about. in

14:24

tech with this concept of the 10x

14:26

programmer, where some programmers are maybe 10

14:29

times better at getting computers to do

14:31

things than other programmers. As more job

14:33

functions become more technical through the use

14:36

of AI, I actually wonder if AI

14:38

creates more opportunities for there to be

14:40

a lot more 10x marketers and 10

14:43

as recruiters and 10x finance people because

14:45

there will be people that will master

14:47

AI and be able to use it

14:50

the way that now software engineers do

14:52

and just get a lot more done

14:54

than others in the same job function

14:56

that don't know how to leverage and

14:59

manage and manage this amazing tool. So

15:01

where is the line then just between

15:03

because obviously most people are even people

15:06

who are interested in training up in

15:08

AI. They tend to just use it

15:10

as a kind of user so they're

15:13

using the large language models or that

15:15

somebody else is built. So is built.

15:17

So is that enough. in this day

15:20

and age just to sort of know

15:22

how to use all the available apps,

15:24

be clever about that, or do you

15:27

have to go deeper and really know

15:29

how to construct them? It is a

15:31

great first step to learn to be

15:34

a user of AI, to know how

15:36

to prompt AI models on the internet

15:38

well. But what I'm seeing is for

15:41

the most sophisticated tasks, people that have

15:43

a deeper understanding of computers would really

15:45

go much further. And at this moment

15:48

in time, learning just a little bit

15:50

about how to code. gives you that

15:52

deeper knowledge to be much more effective.

15:54

It turns out that with AI-assisted coding,

15:57

a little bit of learning goes a

15:59

long way. And I know that some

16:01

people that are saying, yeah, we'll do

16:04

all the coding, you don't need to

16:06

code anymore, I think that's wrong. I

16:08

think the point is not coding, it's

16:11

learning that deeper understanding. how to get

16:13

a computer to do what you want.

16:15

Three years ago there's an eye-opening moment

16:18

for me when I was working on

16:20

a collaborator on using AI to generate

16:22

pictures for background images for a course

16:25

I was teaching and I was writing

16:27

problems and I got pictures that looked

16:29

so-so but one of my collaborators had

16:32

studied art history. She had the right

16:34

language to tell a computer what images

16:36

he wanted to generate and we ended

16:39

up using all of his images in

16:41

none of mine. I think that in

16:43

getting computers to do our business tasks

16:45

as well and different job functions, I'm

16:48

seeing the same thing where people that

16:50

deeper understanding with the right language and

16:52

the right mindset are much better able

16:55

to direct computers to help us with

16:57

various functions, various knowledge word functions. So

16:59

I think this is a good time

17:02

to learn to code. If you could

17:04

change one misconception the public has about

17:06

AI, what would it be? Over the

17:09

last couple years, there's been a lot

17:11

of hype about some parts of AI

17:13

that... Frankly, we're driven, I think, by

17:16

the PR goals of, you know, sometimes

17:18

specific AI companies, trying to make a

17:20

big deal or something, or fundraise or

17:23

whatever. And this has been remarkably effective

17:25

at steering society's attention. But I think

17:27

it is exciting, AI technology, and AI

17:30

companies, but where I think we pay

17:32

not enough attention is at the application

17:34

layer, because the real value of AI,

17:37

mostly value, isn't the technology, is what

17:39

applications we can build on top of

17:41

that. So to most people and businesses,

17:43

I would say, you know, pay attention

17:46

to the hyphes entertaining. This is great.

17:48

This is exciting. But focus also on

17:50

thinking through what the applications you can

17:53

build. And for most businesses is not

17:55

as simple as dumping everything into a

17:57

large language model on the internet. You

18:00

have to identify the unique use case

18:02

and then think through the workflow. Sometimes

18:04

we call these agentic workflows needed to

18:07

realize the specific valuable business application. That

18:09

was Andrew Un talking to us in

18:11

Davos in January still got Kathy Lee

18:14

with us Kathy. I wanted to also

18:16

add that the forum now is seeking

18:18

the most groundbreaking AI enabled business transformation.

18:21

case studies and solutions across industries and

18:23

functions. We just opened up the Minds

18:25

program last week. Minds stands for meaningful,

18:28

intelligent, novel, deployable AI solutions. So please

18:30

do apply and be part of the

18:32

next generation AI transformation business cases. So

18:34

what's that? So if I'm a company

18:37

and I'm doing something really interesting AI.

18:39

I can get involved in this. Absolutely.

18:41

You please, you know, be involved and

18:44

you might get recognized and most importantly,

18:46

there will be a lot you can

18:48

learn from your peers. That's the first

18:51

half of this. In the second half

18:53

of this episode, we'll be talking about

18:55

advanced manufacturing, high-tech factories. But for now,

18:58

Kathy Lee, thanks very much for joining

19:00

us on radio doubles. Thank you, Robin.

19:02

To continue our exploration of how artificial

19:05

intelligence is being adopted. applied in industry.

19:07

I'm joined by my colleague Kiva Orgood.

19:09

Hi Kiva, how are you? Hi, Roman,

19:12

I'm doing well today. Remind us, what

19:14

is it you do at the World

19:16

Economic Forum? Yes, I lead the Center

19:19

for Advanced Manufacturing and we are entirely

19:21

focused on really unifying the global manufacturing

19:23

and supply chain leaders across all industries.

19:25

So we've got 22 different industries that

19:28

we work with to really share best

19:30

practices, spark innovation and scale impact. So

19:32

it's advanced manufacturing and we're looking at

19:35

ways. companies are making stuff, factories of

19:37

the future, this kind of thing. So

19:39

AI must be an important thing for

19:42

them right now. Oh, it is. I

19:44

mean, I'd say, you know, we partner

19:46

with anybody who makes or moves things.

19:49

So the laptop or that you're working

19:51

on, the speakers we're speaking into, these

19:53

are all manufacturers and they have a

19:56

full supply chain and value chain. They

19:58

have been learning and leveraging AI for

20:00

decades. in different forms. So if you

20:03

think about machine learning, the beginning parts

20:05

of using technology to create automation, manufacturers

20:07

are really at the forefront of that,

20:10

and AI is no different. And

20:12

you created this report, one of

20:14

this series of reports that we're

20:16

looking at in this podcast, frontier

20:19

technologies in industrial operations. Maybe

20:21

you can you pick out some

20:23

highlights, maybe that people... might be

20:25

interested in how AI is being

20:27

applied now and in the near

20:29

term. You bet. I mean, part

20:31

of my role, we also run

20:33

a program called the Global Lighthouse

20:35

Network, which is a peer-based awards

20:38

program. Think of it like a

20:40

Michelin Star. We, you know, cite

20:42

supply and they have to supply

20:44

at least 18 months' worth of

20:46

performance improvement data across five different

20:48

dimensions. And really, it's the best

20:50

of the best. And this is

20:52

a peer group award. And so

20:54

we use a lot of that

20:56

insight and data. This program started

20:58

over six years ago. When we

21:00

first launched the Global Lighthouse Network, we

21:02

had a very small number of use

21:04

cases and applications of AI. Today, we

21:07

have almost 80. So 80% of the

21:09

people who are now considered the best

21:11

of the best as far as manufacturing

21:13

and supply chain practices are using AI

21:16

in some way. A lot of them

21:18

are using a generative AI. And so

21:20

I think the... Board of the Day is

21:22

AI, but really what it comes down

21:25

to is process improvement. How are you

21:27

helping the factory or the logistics, save

21:29

money, make money, or come up with

21:32

a new business model? AI plays a

21:34

really interesting role in that. But they

21:36

also have learned over the last bit

21:39

that you need to bring the... the

21:41

floor, the people who work there on that

21:43

journey as well. So a lot of the

21:45

use cases we're highlighting around AI and AI

21:48

agents and a different type is based off

21:50

the fact that you've got to collect the

21:52

insights and the intelligence from the people on

21:54

the factory floor and embed those into these

21:57

different types of agents in order to see

21:59

the biggest... impact and process improvement. Are

22:01

there any examples you could give us

22:03

kind of real world of how AI

22:06

has been deployed in a way that

22:08

has really transformed the way things were?

22:10

There's so many examples actually that the

22:12

the white paper does an amazing job

22:14

of kind of outlining those and giving

22:17

a lot of contextualization to it's not

22:19

really just about AI. I had the

22:21

you know I travel I get to

22:23

travel the world since I've been at

22:25

the form I think I've been around

22:28

the the globe twice and I get

22:30

to walk into a lot of these

22:32

factories and I get to see how

22:34

they're really demonstrating best practices. There's one

22:37

factory I had the opportunity and they're

22:39

using AI in their aluminum stamping process

22:41

to really really get the right strength

22:43

out of the aluminum, you have to

22:45

manage the temperature of the water. So

22:48

they use these different AI agents and

22:50

tools to understand how is that aluminum

22:52

being processed by every millisecond really to

22:54

be able to control the water so

22:57

that they get the best quality product

22:59

at the end. We've seen where you've

23:01

got logistics and supply chain managers, where

23:03

it would have taken them probably a

23:05

month. right, to run a whole bunch

23:08

of different scenarios, say nine different ways

23:10

of thinking about procurement and supply chain.

23:12

They can do that in minutes today.

23:14

So I think there's no end. I

23:16

mean, if you walk in a factory

23:19

that's truly advanced today, you'll have autonomously

23:21

guided vehicles everywhere. They'll be delivering different

23:23

pieces parts just in time for when

23:25

that specific part is needed. Those are

23:28

all great examples of how they're using

23:30

machine learning AI and AI agents on

23:32

the factory floor. Is it easy for

23:34

companies to adopt these technologies? Are there

23:36

some companies you speak to like, I

23:39

feel we should be doing more with

23:41

AI, but we're not really sure where

23:43

to go, you know, which parts of

23:45

our operations we should be developing with

23:47

AI? What are the barriers that... some

23:50

companies may be finding difficult. Typically a

23:52

barrier's data, right? Do you have the

23:54

data in order to create a model?

23:56

And sometimes that's in someone's brain. They

23:59

know how the machine, when it starts

24:01

to vibrate like this, they need to

24:03

do. this. That is also critical. So

24:05

I'd say data tends to be one

24:07

of the areas. That's where you see

24:10

a lot of advancement in spaces like

24:12

financial services or insurance or taxes because

24:14

you've collected that data for decades, right?

24:16

When you think about logistics on a

24:19

factory floor, you have lots of process

24:21

inputs. You have lots of data often.

24:23

It's not clean. It is from lots

24:25

of different types of systems. It has

24:27

different types of taxonomy, so you have

24:30

to kind of collect it and clean

24:32

it before you can take action from

24:34

it. So I interviewed in Davos, Cedric

24:36

Nyker, CEO of Digital Industries at Siemens,

24:38

big company, involved in making big machines,

24:41

often involved in manufacturing, and he's in

24:43

the digital side of that, and he

24:45

says at the start of the interview,

24:47

we make the brains that will run.

24:50

factories in some cases, can you tell

24:52

us something about him? Yeah, I'd say

24:54

he's a pioneer in the industry. I

24:56

think he's, and Siemens in general, they've

24:58

really leaned into the concept of how

25:01

do we ensure that we as an

25:03

industry can collectively move forward, especially when

25:05

you think about sustainability. You can improve

25:07

saving water and electricity and doing things

25:10

with less natural resources if you don't

25:12

have that brain and you don't have

25:14

that insight. And that's something Siemens is

25:16

firmly, firmly leaned into Cedric as well.

25:18

I think they've challenged their, peer group

25:21

to elevate and think more as an

25:23

open ecosystem and to evolve to get

25:25

to that next level of really understanding

25:27

human machines and how they interact on

25:29

a factory floor but 100% that kind

25:32

of intelligence manufacturing or say a software-defined

25:34

factory that's here today. They exist and

25:36

a lot of that is due to

25:38

the fact that Cedric and companies like

25:41

Siemens have really thought about what is

25:43

the industry going to be like 10-15

25:45

years from now and they've really aimed

25:47

for that and leaned forward. Well let's

25:49

hear from Cedric now. My name is

25:52

Cedric Nyke. I'm a board member at

25:54

Siemens and I'm the CEO of the

25:56

digital industry part of Siemens. What is

25:58

the digital industry? So we basically built

26:00

the software and also the automation for

26:03

most industries out there. And this starts

26:05

really on the software side, imagine you

26:07

dream up a new idea so you

26:09

can sort of build it in software,

26:12

you can simulate it, and then you

26:14

make sure that you can build it,

26:16

and then once you put it into

26:18

automation, it actually gets done. These are

26:20

the two things which we do the

26:23

actual, if you want the PLC, is

26:25

the brain of the machines. Siemens built

26:27

a third of them in the world.

26:29

So let me just try and understand

26:32

that. These are other companies who want

26:34

a brain, as you call it, and

26:36

they'll come to you to get that.

26:38

Is that right? So you want to

26:40

build a car. If you build a

26:43

car, we will have the software to

26:45

design the car, to test it, if

26:47

you crush it against the wall, to

26:49

actually make sure how it's getting produced.

26:51

So we do that for a car.

26:54

And then we would build the factory,

26:56

actually building the brains behind that factory.

26:58

We would do this also for pharmaceuticals.

27:00

We would do this for pizzas. We

27:03

would do this. I mean, imagine anything

27:05

which needs to get built, we would

27:07

be in that process. That sounds like

27:09

an area where you'd have been using

27:11

artificial intelligence for a long time. This

27:14

is nothing new to you. So yes,

27:16

we've been looking at AI for a

27:18

long, long time. I think we started

27:20

with some of the base concept in

27:23

the 70s, and I think Siem has

27:25

more than 3,700 patterns in AI, and

27:27

a lot of them are applied to

27:29

industry. What's changed recently then with this

27:31

kind of generative AI explosion? It's a

27:34

bit like the internet, right? I mean

27:36

the internet has... for a long time

27:38

and then boom the browser came up

27:40

and the internet sort of really sort

27:42

of got on its feet and the

27:45

same thing was genitive AI. AI has

27:47

been existing for a long long time

27:49

but Gen AI made it much easier

27:51

to train it where we use specialists

27:54

and months to train AI onto the

27:56

shop floor we can now do it

27:58

in days or even hours and that's

28:00

a huge difference. It's the ease of

28:02

adoption. has increased dramatically and of course

28:05

therefore also the impact. It means that

28:07

people with ideas but not necessarily training

28:09

in software engineering can start working on

28:11

that thing straight away for example. Yeah,

28:13

and I think that's what the generative

28:16

AI has done. It's put AI into

28:18

the hands of everybody. Everybody could become

28:20

a prompt engineer and come up with

28:22

some new ideas. AI was the domain

28:25

of the nerds in the basements, which

28:27

would come up and look at a

28:29

problem and go back and calculate it.

28:31

Now everybody can. I mean, that makes

28:33

a huge difference if you put it

28:36

in the hand of everybody. In what

28:38

way is AI already changing the world

28:40

apart from what we've just mentioned? We

28:42

talk about revolutionary ideas, people even talk

28:45

about, and I'm going to get on

28:47

to your TED Talk in a moment,

28:49

which I watched yesterday, about it even

28:51

saving the world potentially, you know, what

28:53

are we seeing already that isn't just

28:56

efficiency gains or maybe some kind of

28:58

easier interface, things that are really revolutionary.

29:00

Everybody wants the big thing, but I

29:02

think it's the small steps which actually

29:04

make the difference, because if you look

29:07

at people like me, which are sort

29:09

of industry... people. We optimize the world

29:11

in little sort of increments and make

29:13

sure that it continuously improves. So that

29:16

we put sort of products in everybody's

29:18

hands. Imagine a car, right, and when

29:20

Ford came in he made it abatable

29:22

for everything. So we're looking at not

29:24

only the big ideas but also the

29:27

ideas which enable to do that. And

29:29

I'll give you a couple of example

29:31

to make it real, right? First example

29:33

is if you have an idea of

29:36

a new car or if you have

29:38

a new... or Shaver, or whatever product

29:40

you imagine, you probably have built thousands

29:42

of designs before. If you say my

29:44

next design should learn from all the

29:47

designs I had before, AI is wonderful

29:49

for it because it could say, okay,

29:51

I want to optimize for cost or

29:53

for sustainability or for reliability, you could

29:55

optimize it and it would, based on

29:58

all the data you had before, come

30:00

up with some of the designs. So

30:02

in the software part, it's super interesting

30:04

to use it to come up way

30:07

faster with new designs. It could be

30:09

even new molecules. The second thing where

30:11

it's also, so it's on the design

30:13

phase, is you then need to take

30:15

this product and put it into production.

30:18

In the past, you would sort of

30:20

plan the production process, you would test

30:22

it out and optimize, optimize it. You

30:24

can all do it now virtually and

30:26

use AI to run sort of optimization

30:29

cycles. So by the time you put

30:31

it in the production, you're very fast

30:33

in doing it. So with one big...

30:35

food and beverage company, we have the

30:38

idea that you have one recipe idea

30:40

in one day and the next day

30:42

you have the product in production. It

30:44

takes at the moment six months, 12

30:46

months. So things become much faster. And

30:49

what are the barriers to a company

30:51

to implement that? Because if it's as

30:53

great as you say, let's all do

30:55

it, what is stopping this being scaled

30:58

up and really becoming the thing everyone's

31:00

doing now? If you have kids and

31:02

they do their homework, sometimes they come

31:04

up with crazy things. So if you

31:06

would sort of ask an open AI

31:09

who's Cedric Nike and where was he

31:11

born, it would say Paris. I'm actually

31:13

born in Berlin. So there's mistakes which

31:15

happen. Now it's funny if it's a

31:17

sort of a submission of your kids

31:20

in school, but if you get something

31:22

wrong on the shop floor, it has

31:24

potentially huge impact, life-threatening impacts. So we

31:26

need to make, we need to trust

31:29

the problem and we need to see

31:31

that it's safe. is the first thing.

31:33

The second thing is a shop flow

31:35

worker or specialists are often not trained

31:37

IT specialists. That's not the world they're

31:40

coming from. So you need to make

31:42

that technology easy to use and approve

31:44

and you need to show the benefits.

31:46

I give you the example. Normally the

31:49

biggest problem is when you have the

31:51

so-called graveyard shift and factory is that

31:53

the specialist is at home and sleeps.

31:55

So often the problem is if a

31:57

machine has a problem you wait till

32:00

the next day it comes forward. Now

32:02

with Gen AI, the machine can have

32:04

read all its manual. knows that if

32:06

it sees a problem, says, look, on

32:08

page 263, it's probably that issue. So

32:11

when the operator, which is in the

32:13

graveyard shift, goes to it, the machine

32:15

could talk to the operator and say,

32:17

look, I have a problem, this is

32:20

the solution, you should order this piece,

32:22

and this is how you do it.

32:24

I think it solves a lot of

32:26

problems you actually have on the shop

32:28

floor. It's interesting because you said these

32:31

incremental steps, but I was pushing for

32:33

a revolutionary one, which is called... Time

32:35

is running out on climate change, the

32:37

metamorph could help. Within 10 minutes there,

32:39

you tell us, because I've been wondering

32:42

this for the last two years, how

32:44

is AI going to solve climate change?

32:46

Because so far, it's just put more

32:48

emissions in the air. But you, convincingly,

32:51

in 10 minutes. So we don't have

32:53

10 minutes to do that here. But

32:55

I'll try to expect. The biggest problem

32:57

we have in industry is we dig

32:59

stuff up. We then sort of ship

33:02

it around the world. We have an

33:04

idea of what to do with it.

33:06

So I don't know, we build a

33:08

car, we crush it a couple of

33:11

times against the wall, and by the

33:13

end, we're happy. So it's a very

33:15

physical environment. The advantage you have on

33:17

going digital and AI is that you

33:19

can learn from your mistakes way faster.

33:22

So once you have a so-called digital

33:24

twin of a product, you can actually

33:26

say, What material should I use? How

33:28

would it behave? How can I produce

33:30

it? How can I recycle it? So

33:33

you can optimize it in the digital

33:35

world, thanks to AI, because you run

33:37

scenarios. Not one, nor ten. or 100

33:39

times, but thousands or millions of times

33:42

fastest. So you come up with solutions

33:44

way, way faster, which means you can

33:46

build products, which are 80% of riskability

33:48

of a product is defining the design

33:50

stage. So if you can fix that

33:53

up front, you have a huge impact

33:55

there. And I think it's important, and

33:57

I talked about it in the TED

33:59

Talk, is, look, we're going to be

34:02

10 billion people on earth, right? less

34:04

resources than somebody in Zurich or Davos.

34:06

That's unfair. So if everybody wants the

34:08

same level of standard, we cannot use

34:10

that amount of more resources. So we

34:13

have to use the digital space, NAI,

34:15

to see on how efficiently you can

34:17

actually use a resource in the most

34:19

precious and fastest way possible. Yeah, the

34:21

subtitle users, how do we cheat time

34:24

to do more with less? That's, it

34:26

seems like kind of philosophically, it's all

34:28

about time for you that we can

34:30

do. All the things we want need

34:33

to do if we have the time.

34:35

And you're saying this gives us so

34:37

much more time. Takes 100 years to

34:39

build the perfect combustion engine. 100 years.

34:41

We need to get our electrical cars

34:44

to be way more efficient, much faster.

34:46

And the only way of cheating time

34:48

of getting faster is that you basically

34:50

simulate in the digital world on how

34:52

to build the best battery, the best

34:55

car, the best production in the possible

34:57

way. And that's what I mean with

34:59

cheating cheating time is. If you look

35:01

at the 1.5 degrees and wind up

35:04

and it's way warmer than it's normally,

35:06

yes, and every year we say look

35:08

at special policy, but the reality is,

35:10

is if the world is getting warmer,

35:12

we need to find a solution for

35:15

it to get cooler way faster. In

35:17

order to do this, we need to

35:19

cheat time and way to do it

35:21

is to use digital processes and AI

35:24

to do that. So I'll give you

35:26

two examples of factories. So we have

35:28

one which is our sustainability sustainability factory.

35:30

Our sustainability factory inferred we get a

35:32

we get an award this year this

35:35

very normal factory and things to AI

35:37

and simulating energy demand and the things

35:39

I described to you for each product

35:41

we use now 64% less energy for

35:43

the same products. 72% less CO2 which

35:46

means we were planning by 2040 to

35:48

be CO2 neutral in that factory we

35:50

will already beat in 2026 CO2 neutral

35:52

because we use AI to simulate and

35:55

optimize the use of resources. That's what

35:57

I mean by cheating time and being

35:59

much faster. In addition to those massive...

36:01

efficiencies, particularly energy efficiency or resource efficiency.

36:03

And this may be outside of your

36:06

remit and your job, but do you

36:08

see AI coming up with novel solutions

36:10

that could help us on climate change?

36:12

It's a bit like, I mean, look,

36:15

I stood in front of when the

36:17

internet came and it was supposed to

36:19

be the biggest thing and then the

36:21

internet bubble burst and the internet has

36:23

changed everything. It has changed everything. Has

36:26

made the world better. It's definitely made

36:28

it more connected, right? So that's the

36:30

one thing. I think that, I mean,

36:32

the biggest thing of generative AI is

36:34

the capability to build connections. I mean,

36:37

it learns from billions of data points

36:39

to find out new ideas, which is

36:41

if you think about, if you would

36:43

compare it, you would compare it with

36:46

renaissance. In the main idea of the

36:48

engineers was to look at problems from

36:50

four, five, six different ways and come

36:52

up with a new solution. So S-A-I

36:54

goes into that environment. It will definitely

36:57

help human engineers to not do a

36:59

lot of the boring calculation and come

37:01

up with new ideas. So it has

37:03

the potential to find way more efficient

37:05

ways of building solar panels, way more

37:08

efficient ways of heating a house, right?

37:10

I mean, at the moment, 40% of

37:12

all energy is being wasted in buildings.

37:14

It's insane. So imagine that this could

37:17

be done much more dynamically. So yes,

37:19

I believe that in its right way.

37:21

Yeah, I could have a big impact

37:23

on accelerating the energy transition. at Siemens.

37:25

You can find the white can

37:28

find the in this

37:30

discussed in this episode

37:32

the the other work

37:34

of the World

37:37

Economic Center for Advanced Manufacturing and

37:39

the Center for for the

37:41

Fourth Industrial Revolution on

37:43

our website, links

37:45

in the show notes.

37:48

follow follow Radio Davos

37:50

wherever you get

37:52

your podcasts and please please

37:54

take a moment

37:56

to leave us a

37:59

rating and a

38:01

review and join the

38:03

about about podcasts on

38:05

the World Economic

38:08

Forum Podcast Club on

38:10

Facebook. This This episode

38:12

of Radio Davos

38:14

was written and presented

38:16

by Robin Pomeroy with additional

38:19

reporting by by Anna Bruce Lockhart.

38:21

Studio was by by Tass

38:23

We'll be back

38:25

next week with more

38:28

stories about how

38:30

humanity can tackle our

38:32

biggest challenges. biggest for

38:34

now, thanks you

38:36

for listening for listening and goodbye.