0:02

Digital. Twins Gen Vi for engineering

0:05

on today's episode funded a One

0:07

Petrochemical company up Skills it's

0:09

workforce to benefit from new tech

0:12

like generate of a I.

0:15

I'm Ella Nielsen from Chevron.

0:17

And you listening to me

0:19

myself? And A I welcome

0:21

to me myself an Ai, a

0:23

podcast on Artificial Intelligence and business.

0:25

Each episode we hundred issue to

0:27

someone innovating with a I I'm

0:29

Sam Ramsbotham Professor of Analytics the

0:32

Boston College. I'm also the Ai

0:34

and Business Tragic guest editor at

0:36

Mit saw Management Review and I'm

0:38

sure of and co two band

0:40

a senior partner would be C

0:42

G and one of the leaders

0:44

of our Ai business together M

0:46

I S M R and B

0:48

C D. Have been researching and

0:51

publishing on a I since Twenty

0:53

seventeen, interviewing hundreds of practitioners and

0:55

surveying thousands of companies on what

0:58

it takes to build and to

1:00

deploy and scale Ai capabilities and

1:02

really transform the way organizations operate.

1:06

Hi everyone! Today Salmon I are speaking

1:09

with Ellen Nielsen chief Date Officer at

1:11

Chevron Allen. Thanks for taking the time

1:13

to talk to us! Welcome to the

1:15

show! Thank. You for having

1:17

me and really excited to have a

1:19

very cool conversation today. Let's.

1:22

Get started! I would imagine

1:24

most of our listeners in fact, all

1:26

of them have heard about Chevron. But

1:29

what they may not know is the extent

1:31

to which a eyes prevalent across all of

1:33

Chevron's value chain. So maybe tell us a

1:35

little about your role and how Ai is

1:37

being used. That Chevron. May. Talking

1:39

about my role. It started three

1:41

years ago I was the first

1:43

data officer with in Chevron. That doesn't

1:46

mean that we deal with data since

1:48

a long time but to need to

1:50

put more focus on the data was

1:53

starting to emerge and to with that

1:55

I was task in evangelizing data driven

1:57

decisions and at of course include.

2:00

any kind of data science analytics along

2:03

the way. And that was

2:05

very, very interesting to see it growing over

2:07

the time. We use AI in

2:09

many places, some areas

2:11

where we use robots, for example,

2:14

in tank inspection today. You

2:16

can imagine that was very cumbersome, having

2:18

the human involved. Now we do this

2:20

with robots. And we

2:22

take basically the human beings out of

2:24

these confined spaces. And that's a

2:27

combination of computer vision,

2:30

taking images, comparing

2:32

the images, and take predictions on

2:35

what's the status of this tank and this

2:37

equipment. Is it rusting? Does it need

2:39

maintenance? Do we need to tackle

2:41

it in a very predictive way? So that's

2:43

operating in a much more reliable and safe

2:45

way in the future. The

2:47

other example is when

2:49

we talk about sensors in

2:52

compressors or any kind of equipment. In

2:54

the past, we were, of

2:57

course, installing them. But the prices

2:59

dropped dramatically for those sensors and the data

3:01

collection. And I just

3:03

saw recently, actually, it was a

3:05

citizen development application, which has been

3:07

created because these sensors have to

3:09

be installed. And when you install

3:11

them, you basically take a QR

3:13

code. And with one click, you

3:16

can add the geospatial location to

3:18

the sensor. And then you can

3:20

see all these sensors you have installed in

3:22

your facility in a map. So

3:25

you see actually really actively happening

3:27

what's going on and where are

3:30

the things actually working and which

3:32

sensors have been inventoried there. So

3:34

we have a combination here of

3:36

computer vision, of using

3:39

citizen development, and then, of course,

3:41

using the sensor in a machine

3:43

learning AI-based way to

3:45

come to predictions and how they work. So

3:48

one of the things I know that

3:50

you do quite well is digital twin.

3:53

Maybe you can comment a little bit about that

3:56

example. Digital twin is

3:58

one of many examples. where we use

4:00

that, what triggers to

4:02

do digital twin? One is

4:05

you can imagine that we have people out

4:07

in the field so we want to make

4:09

their life easier and safer. That

4:11

means the more data and the more

4:14

information we can gather about our field

4:16

assets and how to operate them will

4:19

serve the purpose for being more safe,

4:21

more reliable in the operations and that

4:23

was one trigger. The second

4:25

trigger is that you collect a lot

4:27

of information based on

4:30

let's say internet of

4:32

things IIoT devices, censoring

4:35

and that feeds into another pool of

4:38

information where you can drive even predictive

4:40

decisions in these assets. So with

4:42

the digital twin we want to basically serve both.

4:45

We want to be safer, reliable but

4:48

also more predictive on what we do that

4:50

speaks to efficiency and do the right thing

4:52

at the right time. Can

4:54

you give us a specific example of a

4:56

place you're using a digital twin? How does

4:58

that help with safety? How does that help with efficiency? If

5:02

you take a digital twin and you'd

5:04

say you digital twin basically a facility,

5:06

a refinery. So in a

5:08

refinery you can imagine there are lots

5:10

of pipes, there are lots of equipment,

5:13

there are compressors, there are

5:15

generators, there are things

5:17

very mechanically working and

5:19

people have to maintain those to get

5:21

the product out. So when you see

5:23

the value chain from product coming in or

5:25

materials coming in and product comes out everything

5:27

between this goes for this

5:29

refinery and if you have everything

5:32

digital twin you can plan better, you

5:34

can operate better, you know when

5:36

things are coming in you can predict better

5:38

on how to get a better output and

5:41

that's basically how we do it in

5:43

refineries or facilities where we operate is

5:46

really looking at the flow of information

5:48

and the data driven decisions. So we

5:50

were always driving decisions with information, you

5:52

know in the past information was more

5:54

in the in the heads of the

5:56

people who were very experienced and sometimes

5:58

augmented of course. i

6:42

think that data

8:00

at hand in a digital way, this is

8:02

quite cumbersome. I cannot imagine how the people did

8:04

it in the past. They may were printing out

8:06

things and laying it on top of it and

8:09

coming with assumptions based on their

8:11

experience. And of course they gained

8:13

a lot of experience. Now we

8:16

do this with machine learning algorithms.

8:18

We understand how the rock composition

8:20

is. We even created

8:22

actually a rockerpedia to know what

8:25

are the different rock conditions and compositions

8:27

so that we can tap into this

8:30

data every day when we need it. Yeah,

8:32

and I think it's a bigger theme

8:34

that with the advent of these technologies,

8:37

the sky's the limit. And so the

8:39

question is, how else can

8:41

you apply it? And what else can you do with

8:43

it? And I think this brings me

8:45

to a question around the mission

8:47

and the purpose. Because there's obviously a

8:49

ton of data. There is obviously a

8:51

lot of tools. And the

8:53

use cases are driven by the mission and what

8:55

are some of the things we want to do

8:57

with that? Yeah, I would

9:00

link it actually in Chevron back

9:02

to our strategy. We do higher

9:04

returns and lower carbon safely. And

9:07

this is our guiding principle. So everything what

9:09

we do should of course benefit the

9:12

success of the company, the

9:14

impact of the company, but also

9:16

doing it in a low carbon environment.

9:18

We know the world looks different in

9:20

a few decades. We look

9:22

after methane. We look after

9:25

greenhouse emissions. We look

9:27

after our carbon footprint overall. So

9:29

this is something what we always

9:31

tackle. And data and AI

9:33

plays their role, but also plays a

9:35

role in how we operate and how

9:38

we operate safely. Safety is

9:40

a big component of Chevron's value

9:42

system. And when you think

9:44

about the future and think about AI and robots

9:47

and digital twin and all of that, there

9:49

is technology out there where we can

9:51

help our people to do their work

9:54

safer and much more

9:56

reliable and in better ways and in new

9:58

ways in the future. What's

10:00

interesting to me about Chevron

10:03

or a company that's predominantly

10:05

an engineering and

10:07

science company is when

10:10

AI is being put

10:12

in production to

10:14

augment some of

10:16

the decisions and some of the

10:18

insights that workers and engineers and

10:20

scientists are making. But

10:23

as an engineer, as an operator of

10:26

these plants, I may not

10:28

quite agree with it. I don't

10:30

know whether this resonates. How

10:32

do you get scientists and

10:34

engineers comfortable to use these

10:36

tools? I

10:38

think it's actually helping because engineers

10:41

have a very logical mindset and

10:43

they know the science and we

10:45

have a lot of science people

10:47

in the company. So when you

10:49

talk about data science and the things behind

10:51

it, we have many people very interested in

10:54

learning data science. And

10:57

we also would say

10:59

we have started to provide

11:02

education. So I think where

11:04

do I start with? So you

11:06

start with learning. I

11:08

don't understand this. That's a typical engineering mindset.

11:11

I don't understand it. I want to understand

11:13

it. I'm looking for what does it tell

11:15

me? How can it influence my solution?

11:17

We have a digital scholarship program

11:20

since a while. And actually we do

11:22

this with MIT where we have cohorts going

11:24

for a year and they're not coming out

11:26

of one department. They're really

11:29

coming out of the whole company going

11:31

through a design engineering master

11:33

in one year, which is really a tough thing

11:35

to do. But they're coming back

11:37

and understanding the new technology, understanding

11:40

the things how we can use it

11:42

differently. And they are the first

11:44

going back into their normal environment

11:46

and influence and basically have

11:48

other people participating from their knowledge and to

11:50

venture out different things maybe they have not

11:53

tried before. So this is one

11:55

thing to influence culture. The

11:57

second thing in the data science space,

11:59

we... started to work with the Rice

12:01

University. We have a six, seven months

12:04

program also going across the company that's

12:06

not only for IT people to

12:08

learn what data science means and they bring

12:10

it back to their environment. So they are

12:12

not leaving their role completely, they go into

12:14

six months, seven months and then they return

12:17

back in the best way possible

12:19

to influence the company. Hey, what is possible?

12:21

The last piece is maybe the broadest

12:24

way because we call it citizen development.

12:27

We believe that many, many people in

12:29

the company get things in their

12:31

hand now with the evolution of AI

12:34

and we just saw gen AI is now

12:36

in the hands of any of everybody who

12:38

wants it. And with this

12:40

kind of citizen development overall, we want

12:42

to bring the technology which

12:45

is becoming much easier to many people

12:47

so that they can use it. And

12:49

of course, they need data for this.

12:51

And that's why we provide the data

12:53

in these systems to be more self

12:55

efficient. So I would say that's a

12:57

three prong kind of approach to influence

12:59

the culture, leadership and we have really

13:02

nice cases over in AI, citizen

13:04

development, we are also publicly talking

13:06

about it with certain use cases

13:08

we do. I think that's the

13:10

culture piece. It takes a while,

13:13

you know, to get into every artery of

13:15

the company, but I feel there's really excitement

13:17

in the company right now to go down

13:19

that road. What I like

13:21

about what you're saying is that actually

13:24

doubling down on the predominantly engineering

13:26

and scientific culture of the company

13:28

and making this a cross

13:31

disciplinary collaboration between science and

13:33

engineering and AI versus any

13:36

of these replacing each other.

13:38

It's an end, not an

13:40

or. Is there a specific

13:42

example you have where you where someone has

13:44

gone to one of these seven month programs

13:47

or the digital scholar program and

13:49

brought back something that's made some change made a

13:51

difference? Yeah, definitely. So we

13:53

have many because we are I think two or

13:55

three years into this. And of

13:57

course, they bring it back and solve several

13:59

issues. We even have this sometimes with

14:01

internships. After two, three

14:04

weeks they recognized they could solve a

14:06

planning issue where they were chewing on

14:08

since some time and it was

14:10

pretty complex but with the new, let's

14:12

say, views and data and artificial

14:15

intelligence, the outcomes were really

14:17

stunning. We

14:19

have actually somebody also

14:21

influencing really the planning of our

14:23

field, field development, creating

14:26

a low-code environment and really

14:28

just breaks in and really

14:31

changes the way how we work. In

14:34

terms of making the company more

14:37

productive, more efficient, ensuring

14:40

it's safe, ensuring that it

14:42

does good for people and communities

14:44

and environment and species in

14:47

all different forms. What has

14:49

been challenging? What's hard? I

14:51

would say there are definitely some challenging parts.

14:54

This is an early

14:56

stage technology, especially in the

14:58

Gen. AI. Things

15:00

are moving very fast. What is challenging,

15:02

whatever you do today might be different

15:05

in three months. The

15:07

challenging part is you cannot work in the

15:09

same way you worked maybe in the past.

15:11

You have to maybe pivot

15:13

faster. It's not that you

15:15

build a solution. I think a company told me they

15:18

built a solution and six

15:20

months later if they would build it now

15:22

again they would do it totally different. You

15:25

have to watch when you, I call

15:27

it maybe put the X in a

15:29

basket, you have to think about

15:31

what's the right timing for what kind of

15:33

use case and figuring this out because you

15:35

don't want to lock yourself in when

15:38

the technology is still in that kind of

15:40

an evolution stage. This is

15:42

something what we watch. The second

15:44

thing is not everything

15:46

in terms of security

15:50

or handling data in the right

15:52

way is solved yet in Gen. AI.

15:56

The technology is not ready. There

15:58

are no solutions yet. and

16:00

you can build a kind of a sandbox

16:03

or a kind of fenced environment,

16:06

but you have to fence it by yourself.

16:08

And I think the hyperscalers like Microsoft and

16:10

so on, I think they're working on also

16:14

adapting those use cases in

16:16

their normal, let's say, landscape

16:18

where you can have

16:20

an authorization process, where you have

16:22

an access process, how you're administering

16:25

and governing this the right way.

16:27

So this is, I would say, still missing.

16:30

I'm very hopeful that this will be

16:32

closed very fast, but today you have

16:35

to pull different technologies if it's a vector database

16:37

to talk a little bit to tech language here.

16:41

It's not already to be used

16:43

on a really wide scale very

16:45

safely. And you have to imagine

16:47

if you have a corporation, there

16:50

are rights in terms of what information

16:53

can be shared, what should be not

16:55

shared, and so on. And that's something

16:57

that we think is a challenge. The

16:59

third challenge I want to mention is

17:02

the policymakers. So we follow

17:04

this very closely with Responsible

17:06

AI. We are a member of the Responsible

17:09

AI Institute and

17:11

watching very carefully what's happening there, what

17:13

kind of policy are coming around the

17:15

corner, how do we

17:17

incorporate that responsibly into our

17:19

operations, into our

17:22

productization of AI models. And

17:25

that's, of course, evolution. It's not something

17:27

you can buy and run

17:29

it. And yeah, we'll see how

17:31

companies are filling these gaps. Helen,

17:34

can you comment on Generative AI and

17:36

if and how it's being used or

17:38

planned to be used? Yeah, yeah, absolutely.

17:40

We are following Generative AI already since

17:42

two years or so, maybe a little

17:44

longer. We were not

17:47

totally surprised by the development. Maybe you

17:49

can say, okay, when was chat GPT

17:52

coming? That was maybe a surprise for everybody

17:54

that it was coming so fast. But

17:56

we were watching this and already did some

17:59

use cases kind of innovative

18:01

sandbox environment to see what will that

18:03

be. And when it came

18:05

out, we said, okay, this is new

18:07

technology. We want to understand it. We

18:10

give it into the hands of the

18:12

people and use it and

18:14

then understand the telemetry of what do we

18:16

use it for and how does it resonate.

18:18

And in May-June, we

18:20

decided to put a more

18:22

dedicated team on those activities.

18:25

And, yeah, we have hundreds

18:27

of use cases now in

18:29

the pipeline, which we down-select

18:31

to the most prominent ones and

18:33

approach them. But technology-wise, we

18:35

are really, I would say, very much

18:37

on top of what's going on and

18:40

have really super smart people working on

18:42

it. I can tell you my own

18:44

use case. I use it for

18:47

writing things down. You can

18:49

talk about maybe writing your

18:51

performance agreement with your

18:53

supervisor or with your team.

18:56

You check on

18:58

presentations or documentations you

19:00

have to do to really optimize

19:02

the writing. I know that

19:05

my team is using it because we

19:07

are thinking in product development and product

19:10

management and portfolio management. So

19:12

in the past, they took much

19:14

longer to write down their thinking

19:16

and I talked with one of my

19:18

team members and she said, you know, in the past it

19:20

took me maybe one or two weeks. Now it

19:22

takes me one hour to get this done. So

19:24

there are lots of efficiency in using,

19:27

let's say, GPT in

19:29

this space. When we look

19:31

into other examples, you can imagine

19:33

we have knowledge databases. We have

19:36

knowledge around system engineering

19:38

and other information we have available within

19:40

the company on a very broad scale.

19:42

And in the past if

19:44

you wanted to know how this generator works,

19:47

you had to basically type and

19:49

search criteria and then finally you found

19:52

the document and you had to read the

19:54

document. Oh, this document was not enough. You

19:56

need another document. Okay, you find the second

19:58

document. Then you complete basically your

20:00

answer and then you go back

20:02

basically execute on it. We have created

20:04

a chat system where you can collaborate

20:07

with this kind of information and figuring

20:09

this out much faster. So

20:11

these are maybe two, maybe more one

20:13

on a daily thing and one maybe

20:16

more related to kind of how

20:18

we work in a systems approach. If

20:21

I combine some of your ideas, I see

20:23

some difficulties. So earlier on you were talking about

20:25

citizen developers and the idea of putting a lot

20:27

of these tools in the hands of people. And

20:30

then later you're talking about problems

20:33

of security and policy that are not

20:35

part of the infrastructure yet. Historically,

20:38

security always follows features. We care

20:40

about features first and then we

20:43

care about security. So we have

20:45

the combination of a widespread proliferation

20:47

of tools amongst citizen developers and

20:51

low infrastructural guardrails or

20:53

policies and then concern

20:56

about inability to fast follow.

20:59

Those seem like they could smash together

21:01

and create a lot of tension.

21:03

How do you navigate that? Yeah,

21:06

I would say maybe we have to

21:08

talk about AI in general and then

21:10

generative AI. So when I talked about the

21:12

policy makers, this was more in the

21:14

generative AI perspective. When you

21:16

think about citizen development, we have

21:19

models or algorithms in

21:21

the box. We have proven, we have secured.

21:24

They have followed a review process. We

21:26

checked on them in terms of responsible

21:28

AI. So they're ready to use for

21:31

any citizen developer who wants to use that.

21:34

So they are secured and safe and they're

21:36

actually in our safe environment. So

21:38

you can already start there and make it

21:40

safe. But the new technology which is coming

21:42

on the Gen AI with these large language

21:44

models and the data behind

21:46

it, where the large language models

21:48

learn from, that's maybe not

21:51

ready yet to put into a citizen

21:53

development perspective. So to make this very

21:55

clear, when I talk about citizen development,

21:57

everything what is secured, kind of

21:59

the telemetry. is there, the space is

22:01

there, we have ensured that we do the right

22:03

thing, this is made available

22:06

for everyone in the company and

22:08

the other things which are maybe not

22:10

secure yet, we are not putting

22:12

that into the system, we are

22:14

waiting. So we cannot just afford

22:16

to have unsecured things into our

22:18

citizen development program. Yeah,

22:20

that brings out a nice sort

22:23

of differentiation between the ideas that

22:25

citizen data ship, data scientist can't

22:28

just be able, there's

22:30

a curation process that goes on and

22:32

it sounds like you're pretty active in

22:34

that curation process and deciding what tools

22:36

go to citizen developers and

22:39

which tools are still

22:41

investigating and you're protecting that makes

22:43

sense. Yeah, that's

22:45

exactly. Chevron is

22:47

obviously a giant petrochemical company out

22:50

there worldwide, everyone knows it and

22:52

you're the chief data officer. How did you get

22:54

there? Tell us a little bit about your history

22:56

of how did you get to this role. Yeah,

22:59

I'm happy to be in this role,

23:01

it's a super exciting area I'm always

23:03

passionate about. When you

23:05

follow my start of my career,

23:07

I'm from Germany, I did a

23:10

system engineering degree and then ventured

23:12

out into digital data later

23:14

on to procurement and supply chain and

23:16

I think the big red thread throughout

23:18

my whole career is the data part

23:21

but of course in different ways. So one can

23:24

say when I ventured out into

23:26

supply chain, you deal with a lot of

23:28

money from the company bought by third

23:30

parties, how do you organize that and

23:32

there's a lot of data

23:34

and thinking and strategic thinking

23:37

about how you do that and I

23:39

would say I'm a learner, I'm a

23:41

humble learner, I like to embrace

23:44

new things and very diverse perspectives

23:46

for the best of the company and

23:49

it's just by coincidence maybe that I got

23:51

into this role because when I joined Chevron

23:53

five years ago, I started

23:56

in the procurement space because I have a

23:58

procurement and data digital lab. I would

24:00

call it, we tackled on

24:02

data right away because the data was

24:04

not as sufficient to drive these decisions

24:06

and maybe the first two years proved

24:09

me right in terms of that's possible. I'm

24:11

also a big believer that data and AI

24:13

will be all around

24:15

us. So this is an

24:18

exciting space to be in and to

24:20

learn and to see what's

24:22

coming next there. So I'm just happy

24:24

to be there. Actually a former executive

24:26

said when I said to him and

24:28

not in Chevron, I'm so lucky all

24:31

the opportunities I had in my career and he said,

24:33

Ellen, you are not lucky. So he

24:35

sent me a book home. You basically

24:37

condition your path, you know, so you're

24:39

open to things even when

24:41

you think it's not on your direct

24:44

trajectory but it's really enhancing your skills

24:46

and how you connect the dots. So

24:49

I like connecting the dots and that's why I'm enjoying this

24:51

role. That's a great story.

24:54

Okay, so these are a series of rapid fire

24:56

questions we ask. Just tell us the first thing

24:58

that comes to your mind. It's

25:00

kind of a speed dating question maybe. What

25:04

do you see is the biggest opportunity for AI

25:06

right now? Healthcare.

25:10

What is the biggest misconception about AI?

25:15

Replacing human beings. What

25:18

was the first career you wanted? What did you want to be

25:20

when you grew up? I didn't want

25:22

to sit on a desk. I

25:24

failed. AI

25:27

is being used like in our daily lives a lot.

25:29

When is there too much AI? I

25:32

would say too much AI would mean

25:34

if it guides me in

25:36

the wrong direction and influences me

25:41

in a way which is not based on the real

25:43

facts. I already

25:45

have too much AI in my car

25:47

because I cannot open the garage because

25:50

it recognizes where I am and

25:52

which thing it has to open

25:54

and if it doesn't work, I can't get in. I

25:57

enjoy this. We have a pretty smart home

25:59

here. with all the

26:01

kind of voice recognition, electronics,

26:04

garage door opener, sprinklers, starters and

26:06

whatsoever. But I would say it

26:09

helps to be more efficient and

26:11

if the network is down, that's

26:13

really hard now. That's right.

26:15

That's right. So last question, what

26:18

is the one thing you wish AI could do

26:20

right now that it can't? Cure

26:24

cancer. Very good. It

26:26

seems like there's a headline every week that this

26:28

new AI thing is going to solve cancer and

26:30

then you look back and none

26:33

of these seem to pan out. I'm not saying we

26:35

should quit trying but it's always the example and it

26:38

seems like it never really gets there. Well,

26:40

it's a little bit of a stochastic process too,

26:42

right? I mean, if you have enough trials

26:45

at it, right? I mean,

26:47

we could for sure try a lot

26:49

more things because of AI and our

26:52

ability to experiment.

26:55

In my answer, it may be slightly different. So I

26:57

think the other thing would be what

26:59

AI maybe cannot do, which would be

27:01

great, really help us with the climate

27:03

transition, the climate questions we have on

27:05

this planet. I think it helps

27:07

here and there. But that would be

27:10

like fantastic if it can help more. Yeah.

27:13

At the same time though, I don't think we can abdicate

27:15

and just hope the machine solves or some

27:18

of the problems that we have created either. I

27:20

think it's going to take both

27:22

of us out working together on that. It's

27:24

okay, that's part of the hope. Is there

27:26

anything you're excited about, artificial intelligence? What's the next thing

27:28

coming to your most excited about right now? Hmm,

27:33

good question. I

27:35

think we want to improve our

27:37

lives. And I think where I

27:39

live right now, we are very privileged. We

27:42

already have AI access in

27:44

many ways. We just talked about it in our

27:46

smart homes and the cars and etc. But that

27:48

doesn't count for everybody in the world. It

27:51

would be great if those

27:53

advances and those benefits would

27:56

be broader available. You

27:58

didn't ask me Sam. I totally agree.

28:01

I mean, I think that if you

28:03

think about just like in education, right,

28:05

and the impact that it can have

28:07

on underprivileged communities and nations that, you

28:09

know, they don't need to have a

28:11

school set up anymore. You

28:13

could do so much and help so

28:16

many people just, you know,

28:18

learn and develop and build skills

28:20

that normally would rely on infrastructure

28:23

and physical people and teachers

28:25

and all that. You

28:27

think I'd be threatened by that, but I'm not

28:29

a bit. I mean, I think that's our biggest

28:31

opportunity. We have so many people that, I mean,

28:33

we just cannot get them all through education programs.

28:36

And the education programs we have are not

28:38

particularly optimized or fast. And if we

28:41

could solve that problem and get better, better

28:43

resources out of our brains, then there'll be

28:45

a huge win. Hey, Sam,

28:47

can I ask you a question? I know I turned this

28:49

now around, but if you think

28:51

that the shelf life of knowledge is

28:54

decreasing, right, there were some recent articles

28:56

about it that maybe what you learned

28:58

today is maybe worth for five years

29:00

and then kind of obsolete.

29:03

So how do you think this will

29:05

evolve in the education system? That's

29:08

huge because I think about that. I mean, I

29:10

teach a class in machine learning and AI, and

29:12

I am acutely aware that unless they're

29:14

graduating the semester that I teach them, everything

29:17

that I'm, you know, these specifics that

29:19

we're teaching them are likely to be

29:22

quite ephemeral. I mean, we've seen how

29:24

rapidly this evolves. I think that pushes

29:26

us to step back and

29:28

be higher level. If we slip into teaching

29:30

a tool, teaching how to click file, how

29:32

to click new, how to click open, how

29:34

to click save, those are very

29:37

low level skills. And

29:39

when we think about what kinds of things we should

29:41

be teaching, I mean, my university is a liberal arts

29:43

university. And I think that's a

29:45

big deal because if we

29:47

think about teaching technical

29:49

skills within a world of

29:52

liberal arts, I think that's

29:54

a big deal. We had the sexiest job

29:56

of the 21st century being data science.

30:00

The next one is not clear to me that data

30:02

science is involved. And it's not that data science isn't

30:04

important, it's just rapidly becoming commoditized.

30:06

And so then we have things like

30:09

philosophy, which become more important, and

30:11

ethics, which as the

30:13

cost of the data science drops, these

30:15

things become more important. Linguistics.

30:17

Linguistics, yeah. There you go.

30:20

Or large language models, right? Yeah.

30:24

Wonderful. Ellen, thank you so much. This has

30:26

been so insightful, and we thank you for

30:28

making the time. Yeah, thank you.

30:30

Thanks for tuning in. On our

30:33

next episode, Shervin and I venture

30:35

into the use of AI in

30:37

outer space with Vandy Verma, Chief

30:40

Engineer of Perseverance Robotic Operations and

30:42

Deputy Manager at NASA's Jet Propulsion

30:44

Laboratory. Please join us. Thanks

30:48

for listening to me, myself, and AI. We

30:51

believe, like you, that the conversation about

30:53

AI implementation doesn't start and stop with

30:55

this podcast. That's why we've

30:57

created a group on LinkedIn specifically for

30:59

listeners like you. It's called AI for

31:01

Leaders. And if you join us, you

31:03

can chat with show creators and hosts,

31:05

ask your own questions, share your insights,

31:08

and gain access to valuable resources about

31:10

AI implementation from MIT SMR and BCG.

31:13

You can access

31:15

it by visiting

31:17

mitsmr.com/AI for Leaders. We'll put that link in

31:19

the show notes, and we hope to see you there.