0:02

All right. Welcome everybody. You

0:05

guys excited? Here we go. Hello, hello.

0:08

Welcome to Smart

0:11

Talks with IBM, a podcast from Pushkin

0:13

Industries, iHeartRadio, and

0:15

IBM. I'm Malcolm Gladwell. This

0:18

season, we're continuing our conversations with new

0:20

creators, visionaries who are creatively

0:23

applying technology and business to drive change,

0:26

but with a focus on the transformative

0:28

power of artificial intelligence and

0:30

what it means to leverage AI as

0:32

a game-changing multiplier for

0:35

your business. Today's

0:37

episode is a bit different. I was recently

0:39

joined on stage by Dario

0:41

Gil for a conversation in front of a

0:43

live audience at the iHeartMedia headquarters

0:46

in Manhattan. Dario is the senior

0:48

vice president and director of IBM

0:50

Research, one of the world's largest

0:53

and most influential corporate research

0:55

labs. We discussed the rise

0:57

of generative AI, what it means

0:59

for business and society. He also

1:01

explained how organizations that leverage

1:03

AI to create value will dominate

1:06

in the near future. Okay, let's

1:09

get on to the conversation.

1:10

Hello,

1:12

everyone. Welcome. And

1:15

I'm here with Dr. Dario

1:17

Gil. And I wanted to say before

1:19

we get started, this is something I said backstage that I

1:22

feel very guilty

1:24

today because you're the, you

1:28

know, arguably one of the most important

1:30

figures in AI research in

1:32

the world. And we have taken you away from your job

1:36

for a morning. It's like if, you

1:38

know, Oppenheimer's wife in 1944 said,

1:42

let's go and have a little

1:44

getaway in the Bahamas. It's

1:46

that kind of thing. You know, what do you say to your wife?

1:49

I can't. We have got

1:51

to work on this thing. I can't tell you

1:53

about. She's like getting me out of Los Alamos. No, so

1:56

I do feel guilty. We've set

1:58

back AI research by.

1:59

I have about four hours here.

2:04

But I wanted to, you've been

2:06

with IBM for 20 years? 20 years,

2:08

yeah, this summer. So, and how old were

2:10

you when you, not to give away your age, but you were, how

2:12

old when you started? I was 28, okay,

2:15

yeah. So I wanna go back to your 28 year old self. If

2:18

I asked you about artificial intelligence,

2:21

I asked 28 year old Dario, what

2:24

does the future hold for AI? How quickly

2:26

will this new technology transform

2:29

our world, et cetera, et cetera? What

2:31

would 28 year old Dario say? Well,

2:33

I think the first thing is that even though

2:36

AI, as a field has been with us

2:38

for a long time, since the mid 1950s,

2:40

at that time, AI was not

2:42

a very polite word to say. Meaning

2:45

within the scientific community, people

2:47

didn't use sort of that term. They would have said things

2:50

like, maybe I do things related to machine

2:52

learning, right? Or statistical

2:54

techniques in terms of classifiers and so

2:56

on. But AI had a mixed

2:59

reputation, right? It had gone through different cycles

3:01

of hype, and it's

3:03

also moments of a

3:06

lot of negativity towards it because

3:08

of lack of success. And

3:11

so I think that that would be the first thing. We probably say like

3:13

AI is like, what is that? Like

3:16

respectable scientists are not working on AI defined

3:19

as such. And that really changed over

3:21

the last 15 years only, right? I would say

3:23

with the advent of deep learning over

3:25

the last decade is when that reenter

3:28

again, the lexicon of saying AI,

3:30

and that that was a legitimate thing to

3:32

work on. So I would say that that's the first thing I think

3:34

we would have noticed at contrast 20 years ago. Yeah.

3:37

So at what point in your 20 year tenure

3:40

at IBM, would you say you

3:42

kind of snapped into present kind

3:45

of wow mode? I

3:48

would say in, late

3:52

2000s when IBM was working

3:55

on the Jeopardy project. and

4:01

just seeing the demonstrations of what

4:03

could be done in question answering. It's

4:06

literally jeopardy is this crucial

4:08

moment in the history of AI. You

4:10

know there had been a long and wonderful history

4:13

in inside IBM on AI. So

4:16

for example like you know in terms of like these grand

4:18

challenges at the very beginning

4:20

of the field founding which is this famous Dartmouth

4:23

conference that actually IBM sponsored

4:26

to create there was an IBM

4:28

out there called Nathaniel Rochester and

4:31

there were a few others who right after

4:34

that they started thinking about demonstrations

4:36

of this field and for example they created

4:38

the first you know game to

4:40

play checkers and to demonstrate that you could

4:42

do machine learning on that.

4:45

Obviously we saw later in the 90s like

4:47

chess that was very famous example of that.

4:50

Deep blue. With deep blue. Yeah. Right and

4:52

playing with Kasparov and then but I think

4:54

the moment that was really those other

4:56

ones felt like you know kind of like brute force anticipating

4:59

sort of like moves ahead. But this aspect

5:01

of dealing with language

5:02

and question answering felt different

5:05

and I think for us

5:07

internally and many others was when a moment of saying

5:09

like wow you know what are the possibilities

5:11

here and then soon after that connected

5:14

to the sort of advancements in computing

5:16

and with deep learning the last decade

5:18

has just been an all out you know sort of like

5:20

front of advancements and that and I just continue

5:23

to be more and more impressed and the last few

5:25

years have been remarkable too. Yeah. So

5:27

we'll ask you three quick

5:29

conceptual questions before we dig into it. Just

5:32

so I sort of get a we all get a feel

5:34

for

5:35

the shape of AI.

5:38

Question number one is where are we in

5:41

the evolution of this? So you

5:44

know

5:44

the obvious quite we're all suddenly aware

5:47

of it we're talking about it. What can you give

5:49

us an analogy about where we are in the kind

5:51

of likely evolution of this as a technology?

5:55

So I think we're in a significant

5:58

inflection point that each feels

6:00

the equivalent of the first browsers

6:03

when they appear and people imagine

6:06

the possibilities of the internet or more imagine

6:09

Experience the internet the internet

6:11

had been around right for quite a few decades

6:13

AI has been around for many decades I

6:16

think the moment we find ourselves is that people

6:18

can touch it and they can

6:20

before they were AI systems that were like behind the

6:22

scenes like your search results or or Translation

6:25

systems, but they didn't have the experience

6:28

of like this is what it feels like to interact with this

6:30

thing So so that's what I mean

6:32

I think maybe that analogy of the browser is appropriate

6:34

because it's all of a sudden is like whoa You

6:36

know this this network of machines

6:38

and content can be distributed and everybody

6:40

can self publish And there was a moment

6:42

that we all remember that and I think that that

6:45

is what the world has experienced over the last

6:47

nine months or so on so and

6:49

But fundamentally also what is important is

6:51

that this moment is where the ease of

6:53

the number of people that can build and

6:56

use AI Has skyrocketed

6:59

so over the last decade,

7:01

you know technology firms that

7:03

had

7:03

large research teams Could

7:05

build AI that worked really well, honestly

7:08

But when you went out into say hey

7:10

can everybody use it kind of data science

7:13

team in a bank You know go and develop these

7:15

applications

7:15

It was like more complicated some

7:17

could do it, but it was more the barrier of entry was

7:20

high now It's very different because

7:22

of foundation models and the implications

7:24

that that has for the moment when the technology

7:27

is being democratized It's been democratized

7:30

Frankly works better For

7:32

classes of problems like programming and other things

7:34

is really incredibly impressive what he can do So

7:37

the accuracy

7:37

and the performance of it is much better and

7:40

the

7:40

ease of use and the number of use cases

7:42

We can pursue is much bigger so that democratization

7:45

is a big difference But when you say when you make an

7:47

analogy to the first browsers If

7:49

you if we do another one

7:51

of these time travel questions back

7:53

at the beginning of the first browsers It's

7:56

safe to say many of the potential

7:58

uses of the Internet such we

8:01

hadn't even begun, we couldn't even anticipate. Right.

8:04

So we're at the point where the future direction

8:06

is largely unpredictable. Yeah,

8:08

I think that that is right because it's such

8:10

a horizontal technology that

8:13

the intersection of the horizontal capability,

8:15

which is about expanding our productivity

8:17

and tasks that we wouldn't

8:19

be able to do efficiently without it, has

8:22

to marry now the use cases that

8:24

reflect the diversity of human experience, our institutional

8:26

diversity. So as more and more institutions

8:28

say, you know, I focus on agriculture, you know,

8:31

to be able to improve seeds, you know,

8:34

in these kinds of environments, they'll find their

8:36

own context in which that matters that the creators

8:38

of AI did not anticipate at the beginning. So

8:41

I think that that is then the fruit of surprises

8:43

will be like, why, we didn't even think that it could

8:45

be used for that. And also clever people

8:47

will create new business models associated

8:50

with that, like it happened with the internet, of course,

8:52

as well. And that will be its own

8:55

source of transformation and change in its own right.

8:57

So I think all of that is yet to unfold.

8:59

Right. What we're seeing is this catalyst moment of

9:02

technology that works well enough and it can be democratized.

9:05

Yeah. What next sort of conceptual

9:07

question, you know, we can loosely

9:09

understand or categorize

9:12

innovations in

9:14

terms of their impact on the kind of

9:18

balance of power between have and have

9:20

nots. Some innovations, you

9:23

know, obviously favor those who

9:25

already have a make the rich richer. Some

9:29

it's arising to either the left or the left of both. And some

9:33

are biased in the other direction. They close the gap

9:35

between is it possible

9:37

to say, to predict which

9:39

of those three categories AI might fall

9:42

into? It's a great question.

9:44

You know, a first observation

9:46

I would make on your first two categories

9:50

is that it will be both likely

9:52

be true that the use of

9:54

AI will be highly democratized, meaning the number of

9:57

people that have access to its power to

9:59

make. improvements in terms of efficiency and so

10:01

on will be fairly universal. And

10:04

that the ones who are able

10:06

to create AI may

10:09

be quite concentrated. So if you

10:11

look at it from the lens of who

10:13

creates wealth and value over

10:15

sustained periods of time, particularly,

10:18

say, in a context like business, I

10:20

think just being a user of AI

10:23

technology is an insufficient strategy. And

10:26

the reason for that is, yes, you will get the immediate

10:29

productivity boost of just making API

10:31

calls and that will be a new baseline

10:33

for everybody. But you're

10:35

not accruing value in terms

10:37

of representing your data inside the AI

10:40

in a way that gives you a sustainable competitive advantage.

10:43

So I always try to tell people, don't

10:45

just be an AI user, be an AI

10:47

value creator. And I think that

10:49

that will have a lot of consequences

10:52

in terms of the haves and have-nots as an example

10:55

and that will apply both to institutions and

10:57

regions and countries, etc. So

11:00

I think it would be kind of a mistake to

11:02

just develop strategies that are just

11:04

about usage. Yeah. But

11:08

to come back to that question for a moment, to give you a specific,

11:10

suppose I'm an industrial

11:13

farmer in Iowa with $10 million

11:17

of equipment and I'm comparing

11:19

it to a subsistence farmer somewhere

11:22

in the developing world who's got a cell phone. Over

11:26

the next five years, whose

11:29

well-being rises by a greater amount?

11:32

Yeah, I think,

11:34

I mean, it's a good question, but it might be

11:36

hard to do a one-to-one sort of like attribution

11:39

to just one variable in this case, which is AI.

11:43

But again, provided that you have access

11:45

to a phone and some kind

11:47

to be able to be connected, I

11:49

do think, so for example, in that context, we've

11:52

developed, we've done work with NASA as an

11:54

example to build geospatial models

11:57

using some of these new techniques. And I

11:59

think, for example, or ability to do

12:01

flood prediction. I'll tell you an advantage of why it would

12:03

be a democratization force in that context.

12:06

Before, to build a flood model based

12:08

on satellite imagery was

12:11

actually so on earth and so complicated and difficult

12:13

that you would just target to very specific regions.

12:16

And then obviously, countries prioritize their own, right?

12:18

But what we've demonstrated is actually you

12:20

can extend that technique to have like global coverage

12:23

around that. So in that context, I would say

12:25

it's a force towards democratization that everybody

12:27

sort of would have access if you have some kind of connectivity.

12:30

That Iowa farmer might have a flood model.

12:33

The guy in the developing world definitely

12:35

didn't. And now he's got a shot at getting one. Yeah, but now

12:37

he has a shot at getting one. So there's aspects of

12:39

it that so long as we provide connectivity and

12:42

access to it, that there can be democratization

12:44

forces. But I'll give you another example that can

12:47

be quite concerning, which is language, right?

12:49

So there's so much language in

12:53

English. And there

12:55

is sort of like this reinforcement loop

12:57

that happens that the more you concentrate because

12:59

it has obvious benefits for global communication and

13:01

standardization, the more you can enrich

13:04

base AI models based on that capability.

13:07

If you have very resource-carched languages,

13:10

you tend to develop less powerful

13:12

AI with those languages and so on. So

13:15

one has to actually worry and

13:17

focus on the ability to

13:19

actually represent, in that case, is

13:22

language as a piece of culture, also in

13:25

the AI such that everybody can benefit

13:27

from it too. So there's a lot

13:29

of considerations in terms of equity about

13:32

the data, the data sets that we accrue,

13:35

and what problems are we trying to solve. I mean, you

13:37

mentioned agriculture or health care and so on. If

13:39

we only solve problems that are related

13:41

to marketing, as an example, there will be a less

13:43

rich world in terms of opportunity that

13:46

if we incorporate many, many other broader set

13:48

of problems. Yeah. Who do you

13:50

think, what do you think are the biggest impediments

13:53

to the adoption of AI

13:57

as you think AI ought to be adopted? But

14:00

look, what are the sticking points that you would... Look,

14:03

in the end, I'm gonna give a non-technological

14:05

answer as a first one, has to do with workflow,

14:07

right? So even if the technology is very

14:11

capable, the organizational change

14:13

inside a company to incorporate into the natural

14:15

workflow of people on how we work is,

14:19

it's a lesson we have learned over the last decade, it's hugely

14:22

important. So there's a lot

14:24

of design considerations, there's

14:26

a lot of how do people want to work,

14:29

right? How did they work today, and what is the

14:31

natural entry point for AI? So that's like number

14:33

one. And then the second one is,

14:36

for the broad value creation

14:38

aspect of it, is the understanding inside

14:41

the companies of how

14:43

you have to curate and create data

14:46

to combine it with external data such

14:48

that you can have powerful AI models

14:51

that actually fit your need. And that

14:53

aspect of what it takes to actually

14:56

create and curate the data for these modern

14:58

AI, it's still a work

15:01

in progress, right? I think part

15:03

of the problem that happens very often when I talk

15:05

to institutions is that they say, AI, yeah, yeah,

15:07

yeah, I'm doing it, I've been doing it for

15:10

a long time. And the reality

15:12

is that that answer can sometimes be a little of a

15:14

cop-out, right? It's like, I know you were

15:16

doing machine learning, you were doing some

15:18

of these things, but actually the latest

15:20

version of AI what's happened with foundation

15:22

models, not only is it very new,

15:24

it's very hard to do.

15:26

And honestly, if you haven't been assembling

15:29

very large teams and spending hundreds of millions

15:31

of dollars of compute, and you're probably not

15:33

doing it, right? You're doing something else

15:36

that is in the broad category. And

15:38

I think the lessons about what it means

15:40

to make this transition to this new wave is

15:42

still in early phases of understanding. So

15:44

what would you say, I wanna give you a couple of examples

15:47

of people with kind of

15:49

real world, in real world positions of responsibility.

15:52

Imagine I'm sitting right here. So imagine that

15:54

I am the president of a small liberal

15:56

arts college. And I come to you and I say,

15:58

Dario, I keep hearing about AI. AI, my

16:01

college has, you know, I don't make

16:03

it, you know, I'm, my, my, I'm not,

16:05

I'm making this much money if that every year, my enrollment's

16:08

declining, I

16:10

feel like this maybe is an opportunity. What is the

16:12

opportunity for me? What would

16:14

you say? Um, so

16:17

it's probably in a couple of segments around that, right?

16:20

One has to do is, well, what

16:22

is the implications of this technology inside

16:25

the institution itself instead of the college

16:28

and how we operate and can

16:30

we improve, for example, efficiency, like if

16:32

you're having very low levels of,

16:35

of sort of margin to be able to reinvest is,

16:37

you know, you run IT, you

16:39

run, you know, infrastructure,

16:42

you run many things inside the college. What are the

16:44

opportunities to increase the productivity or

16:46

automate and drive savings such that you

16:49

can reinvest that money into the mission

16:51

of education, right? As an example. Number one is

16:53

operational efficiency. Operational efficiency

16:56

is a big one. I think the second one is within

16:58

the context of the college is implications for the educational

17:01

mission on its own, right? How will, you

17:03

know, how does the curriculum need to evolve

17:05

or not? What are acceptable use policies

17:08

or some of these AI? I don't think we've all read

17:10

a lot about like what can happen in terms of exams

17:12

and, and so on and cheating and not cheating

17:14

or what are they actually positive elements of it in

17:17

terms of how curriculum should be developed and professions

17:20

sustain around that. And then there's another

17:22

third dimension, which is the outward oriented element

17:24

of it, which is like prospect students, right? So so

17:27

which is frankly speaking, a big use

17:29

case that has happened right now, which in the broader

17:31

industry is called customer care or client care

17:33

or citizen care. So in this question will be education

17:36

like, you know, hey, are you reaching the right

17:38

students around that that may

17:40

apply to the college? How can you

17:42

create them, for example, an environment to interact

17:44

with the college and answering questions that could be a chatbot

17:47

or something like that to learn about it and

17:49

personalization? So I would say there's

17:51

like at least three lenses with which I would

17:53

give advice. The second

17:55

part of the second one, because it's really interesting. So

17:58

I really

17:59

You can't assign an

18:01

essay anymore, can I?

18:03

Can I assign an essay? Yeah. Can

18:06

I say, write me a research paper and come

18:08

back to me in three weeks? Can I do that anymore? I think

18:10

you can. How do I do that? And then

18:12

you can then. Look, there's

18:14

two questions around that. I think

18:17

that if one goes and explains

18:19

in the context like, why are we here? Why in

18:21

this class? What is the purpose of this? And

18:25

one starts with assuming like an element

18:27

of like decency on people or people are there or like

18:29

to learn and so on. And you just give a disclaimer,

18:32

look, I know that one option you have

18:34

is like just put the essay question and

18:36

click go and give an answer. But

18:39

that is not why we're here. And that is not

18:41

the intent of what we're trying to do. So first I would start

18:43

with the norms

18:46

of intent and decency and appeal

18:49

to those as step number one. Then

18:51

we all know that there will be a distribution of use cases

18:54

that people like that will come in one ear and come

18:56

out of the other and do that. So

18:58

for a subset of that, I think the

19:01

technology is going to evolve in such a way that we

19:03

will have more and more of the ability to discern

19:07

when that has been AI generated and created. It

19:10

won't be perfect. But there's some

19:12

elements that you can imagine inputting the essay

19:15

and you say, hey, this is likely to be generated

19:17

around that. And for example, one way you

19:19

can do that just to give you an intuition, you could just have an

19:22

essay that you write with pencil and

19:24

paper at the beginning. You get a

19:26

baseline of what your writing is like. And

19:28

then later when you generate

19:31

it, there will be obvious differences around

19:33

what kind of writing has been generated on the other.

19:37

Everything you're describing

19:39

makes sense, but in this

19:42

respect at least, it seems to greatly complicate

19:44

the life of the teacher. Whereas the other two use

19:46

cases seem to kind of clarify

19:50

and simplify the role, suddenly

19:53

teaching prospective

19:55

students sounds like they can do that much more

19:58

kind of efficient a lot. administration

20:00

costs, but the teaching thing is

20:02

tricky.

20:03

Well, until we develop

20:05

the new norms, right? I mean, again,

20:07

I know it's an abuse analogy, but calculators,

20:10

we deal with that too, right? And

20:12

it says, well, calculator, what is the purpose of math,

20:14

how are we going to do this, and so on. And we

20:17

have. Can I tell you my dad's calculator story? Yes,

20:19

please. My father taught mathematics at

20:21

the University of Waterloo

20:23

in Canada. In the 70s,

20:25

when people started to get OCA calculators,

20:28

his students demanded that they be able to

20:30

use them. And he said no, and they took

20:32

him to the administration, and he lost. So

20:35

he then changed completely

20:38

throughout all of his old exams, introduced new

20:40

exams where there was

20:42

no calculation. It was all

20:46

like, figure out the problem on a conceptual

20:48

level and describe it to me. And they

20:51

were all students, deeply unhappy that he

20:53

had made their lives better complicated. But to

20:56

your point, I mean, he the

20:59

result was probably a better education. He

21:02

just removed the element that

21:04

they could gain with their pocket calculators.

21:06

I suppose it's a version of. I think it's a version

21:08

of that. And I think they will develop the equivalent

21:10

of what your father did. And I think people say, you know what, it's

21:12

like these kinds of things, everybody's doing it generically,

21:15

and none of us have any meaning. Because all

21:17

you're doing is pressing buttons. And the intent of

21:19

this was something, which was to teach you how to write or

21:21

to think or something. There may be a variant of

21:23

how we do all of this. I mean, obviously, some version

21:26

of that that has happened is like, OK, we're all going

21:28

to sit down and do it with pencil and paper. And my computer's

21:30

in the classroom. But there'll be other variants of creativity

21:33

that people will put forth to say, you know what, that's

21:36

a way to solve that problem, too. But this is interesting

21:38

because, to stay on this analogy,

21:41

we're really talking about a profound

21:44

rethinking, just using a college

21:46

as an example, a real profound rethinking

21:49

of the way. There's no part

21:51

of this college that's unaffected by AIA.

21:55

In one case, I've made everyone's job

21:58

easier. In one case, I've made I'm

22:00

asking us to really rethink from the ground up what

22:03

teaching means. In another

22:05

case, I've automated systems that I didn't think of.

22:07

I mean, it's like, that's right. That's

22:10

a lot to ask someone who got

22:12

a PhD in medieval language literature 40 years

22:15

ago. Yeah, but

22:17

I'll tell you a positive development that I'm seeing

22:19

in the sciences around this, which is you're

22:22

seeing, as you see more and more examples

22:25

of applying AI technology within

22:27

the context of historians to as an example.

22:31

You have archival and you have all

22:33

these books and being able to actually help

22:35

you as an assistant around that,

22:37

but not only with text now, but with diagrams.

22:41

And I've seen it in anthropology too, and

22:44

archeology with examples of engravings

22:46

and translations and things that can happen. So

22:49

as you see in diverse fields, people

22:52

applying these techniques to advance on how

22:54

to do physics or how to do chemistry. They

22:56

inspire each other, right? And they say,

22:58

how does it apply actually to my area? So

23:01

once, as that happens, it becomes less of

23:03

a chore of like, my God, how do I have to deal

23:06

with this? But actually it's triggered by curiosity.

23:09

It's triggered by, there'll be like

23:11

faculty that will be like, you know what, let me explore

23:13

what this means for my area. And

23:16

they will adapt it to the local context, to

23:18

the local language and

23:20

the profession itself. So I see

23:22

that as a positive vector that is

23:24

not all going to feel like homework. It's

23:26

not going to feel like, oh my God, this is so overwhelming.

23:29

But rather to be very practical to see what works,

23:31

what have I seen others to do that is inspiring,

23:34

and what am I inspired to do? You know, what, what

23:36

is, how is this going to help my career? I think

23:38

that that's going to be an interesting question for, you

23:41

know, those faculty members, for the students and professionals.

23:43

Yeah. Sorry, I'm going to stick with this

23:45

example a lot because it's really interesting. I'm curious

23:47

following up on what you just said, that one

23:50

of the most persistent critiques of

23:53

academia, but also of many, of many

23:55

corporate institutions in

23:57

recent years, has been siloing.

24:00

Right? Different parts of the

24:02

organization are going off on their

24:04

own and not speaking to each other. Is

24:08

a real potential benefit

24:11

to AI the kind of breaking

24:13

down, a simple tool for breaking

24:15

down those kinds of barriers? Is that a

24:17

very elegant way of sort of saying

24:20

what we are going to do? I really think that I was actually

24:22

just having a conversation with Provost,

24:24

very much on this topic very recently, exactly

24:27

on that, which is all

24:30

this appetite to collaborate across disciplines.

24:32

There's a lot of attempts towards

24:34

our goal, like creating interdisciplinary centers,

24:37

creating dual degree programs or dual appointment

24:39

programs. But actually, in a

24:42

lot of progress in academia,

24:44

happens by methodology too.

24:46

When some methodology gets adopted,

24:49

I mean the most famous example

24:51

of that is a scientific method, as an example of

24:53

that. But when you have a methodology that

24:56

gets adopted, it also provides a way

24:58

to speak to your colleagues across

25:00

different disciplines. And I think what's happening

25:03

in AI is linked to that. That within

25:05

the context of the scientific method, as an example,

25:08

the methodology about

25:11

which we do discovery,

25:13

the role of data, the role of these neural

25:15

networks of how we actually find proximity

25:17

to concepts to one another, is actually

25:20

fundamentally different than

25:22

how we traditionally applied it. So

25:25

as we see across more professions, people

25:27

applying this methodology is also

25:29

going to give some element of common language

25:32

to each other. And in fact,

25:35

in this very high dimensional representation

25:37

of information that is present to neural networks,

25:40

we may find amazing adjacencies

25:42

or connections of themes and topics in

25:45

ways that the individual practitioners cannot

25:47

describe, but yet will be latent

25:49

in these large common neural networks. We

25:52

are going to suffer a little bit from causality,

25:54

from the problem of like, hey, what's the root cause

25:56

of that? Because I think one

25:58

of the unsatisfied

25:59

aspects that this methodology

26:02

will provide is they may give you answers from

26:04

which they don't give you good reasons for

26:06

where the answers came from and

26:09

then there will be the traditional process of discovery

26:11

of saying if that is the answer what are the

26:13

reasons so we're gonna have to

26:16

do this sort of hybrid way of

26:18

understanding the world but I do think

26:20

that common layer of AI is a powerful

26:22

new thing. Yeah, what a

26:24

couple of random questions that kind of mind as you talk

26:27

in the in the writers strike that

26:29

just ended in Hollywood one of the sticking points

26:31

was how the studios and writers

26:33

would treat AI generated content

26:36

right would writers get credit

26:38

if their material was somehow the source

26:41

for a but more broadly

26:44

did the writers need protections against the use

26:46

of I could go on you know what yeah I'll be worth familiar

26:48

with all of this had you been I don't know

26:50

whether you were but had either

26:53

side called you in for advice

26:55

during that

26:56

the writers had the writers called you and said Dario

26:59

what should we do about AI and

27:01

how should we that should be reflected

27:04

in our contract negotiations what would you

27:06

have told them?

27:08

The way I think about that is that I divided

27:11

I would divide it into two pieces first is what's

27:13

technically possible right and

27:15

anticipate scenarios like you

27:17

know what can you do with voice cloning for example

27:20

you know now for example it is possible there's

27:23

been dubbing right

27:25

like let's just take that topic right around the world

27:27

there was all these folks that would dub

27:29

people in other languages well now

27:32

you can do this incredible renderings

27:34

I mean I don't know if you've seen them where you

27:36

know you match the lips is your original

27:38

voice but speaking any language that you want as

27:40

an example so busy that has a set of

27:42

implications around that I mean just to give an example so

27:44

I would say create a taxonomy that

27:47

describes technical capability that we know

27:49

of today and applications

27:52

to the industry and two examples of

27:54

like hey you know I could film you for five minutes and I

27:56

could generate two hours of content of you and I don't

27:58

have to you know that And if you get paid by

28:00

the hour, obviously I'm not paying you for that other thing.

28:03

So I would say technological capability

28:05

and then map with their expertise consequences

28:08

of how it changes the way they work or

28:10

the way they interact or the way they negotiate

28:12

and so on. So that would be one element of

28:14

it. And then the other one is like a non-technology

28:17

related matter, which is an element of almost

28:19

of distributed justice. It's like who deserves what,

28:21

right? And who has the power to get what? And

28:25

then that's a completely different discussion. That

28:27

is to say, well, if this is the scenario of what's possible,

28:31

what do we want and what are we able

28:33

to get? And I think that that's a different

28:35

discussion, which is all that's life. Which

28:37

one do you do first? I

28:40

think it is very helpful to have

28:42

an understanding of what's possible

28:44

and how it changes the landscape as

28:47

part of a broader discussion,

28:50

right, and a broader negotiation. Because

28:53

you also have to see the opportunities because there

28:55

will be a lot of ground to say, actually,

28:58

you know, if we can do it in this way

29:01

and we can all be that much more efficient in

29:03

getting this piece of work done or this filming done,

29:06

but we have a reasonable agreement about

29:08

how we both sides benefit from it,

29:11

right? Then that's a win-win for

29:13

everybody. Yeah. Right? So

29:16

that's a, I think that would be a golden triangle, right? Here's my

29:18

reading and I would like you to correct me if I'm wrong

29:20

and I'm likely to be wrong. When

29:23

I looked at that strike, I said, if they're worried

29:25

about AI, the writers

29:27

are worried about AI. That seems silly.

29:30

It should be the studios who are worried about

29:32

the economic impact of AI. Doesn't in

29:34

the long run AI put the studios

29:36

out of business long before it puts the writers out of business?

29:39

I only need the studio because the cost of

29:41

production are as high as the sky

29:44

and the cost of production are overwhelming. And

29:47

whereas if I don't,

29:48

if I have a tool which brings, introduces

29:51

massive technological efficiencies to the production

29:54

of movies, then I don't need, why don't we need a studio?

29:57

Why would they be the scared ones? Maybe

29:59

you need a. like a different kind of studio. Or a different kind

30:01

of studio. A different kind of studio. But I mean,

30:04

in the strike, the frightened

30:08

ones were the writers and the, you

30:10

know, were the studios. Wasn't

30:12

that backwards? I haven't

30:15

thought about it.

30:16

It can be, but the implications of it, it goes

30:18

back to what we were talking before. The implications, because

30:20

they are so horizontal, it is right to

30:22

think about it like what does it do to the studios as well,

30:24

right? Yeah.

30:26

And you know, the reason why that happens

30:28

is that

30:29

it's the order of either negotiations

30:32

or who first got concerned about

30:35

it and did something about it, right? Which is

30:37

in the context of the strike. You

30:39

know, I don't know what the equivalent conversations are

30:41

going inside the studio and whether they have a war room

30:43

saying what this is going to mean to us, right? But

30:46

it doesn't get exercised through a strike, but

30:48

maybe through a task force inside, you know, the

30:50

companies about what are they going to do, right? And

30:53

to go back to your thing, you said the first thing you do is you

30:55

make a list of what technological capabilities are, but

30:58

don't technological capabilities change every,

31:00

I mean, you're

31:02

racing ahead so fast. So you can't, can

31:05

you have a contract? Sorry

31:07

for getting a little weeds here, but this is interesting. Can

31:09

you have a, you can't have a five year contract

31:12

if the contract is based on an assessment of technological

31:15

capabilities in 2023, because by the time it gets to 2028, it's

31:18

totally

31:22

different, right?

31:24

Yeah. So, you know, I mean, where

31:26

I was going is like there are some

31:28

abstractions around that is like, you

31:31

know, what can we do with my image,

31:33

right? Like if I generally get the category

31:35

that my image can be reproduced, generated

31:37

content and so on, it's like, let's talk about

31:40

the abstract notion about who has rights to that

31:42

or do we both get to benefit from that? If

31:45

you get that straight, yes, the nature

31:47

of how the image gets altered, created at

31:49

something will change underneath, but the

31:51

concept will stay the same. And so

31:53

I think is what's important is to get the categories right.

31:56

Yeah. Yeah. If

31:58

you had to, if you just think. about the biggest technological

32:03

revolutions of the post-war

32:05

era, the last 75 years. We can

32:08

all come up with a list. Actually, it's really fun

32:10

to come up with a list. I was thinking about this when we were,

32:13

you

32:13

know,

32:14

containerized shipping is my favorite. The

32:18

Green Revolution, the internet.

32:22

Where is AI in that list? So

32:26

I would put it first. In that context

32:28

that you put forth over since World

32:30

War II, undoubtedly

32:33

computing as a category is one

32:35

of those trajectories that

32:37

has reshaped our world. And

32:40

I think within computing, I

32:42

would say the role that

32:44

semiconductors have had has been

32:46

incredibly defined. I would say AI

32:49

is the second example of

32:51

that as a core architecture that

32:54

is going to have an equivalent level of impact.

32:57

And then the third leg I would put to that equation

32:59

would be quantum and quantum information. And

33:01

that's sort of like I like to summarize that the future

33:03

of computing is bits, neurons, and qubits. And

33:06

it's that idea of high precision computation,

33:08

the world of neural networks and artificial

33:10

intelligence and the world of quantum. And

33:13

the combination of those things is going to

33:15

be the defining force of the next hundred years in

33:18

that category of computing. But it makes the list

33:20

for sure. If it's that high up

33:22

on the list, this is a total hypothetical,

33:25

if you were starting over, if you're

33:28

starting IBM right now, would

33:30

you say, oh, our AI operations actually should be

33:34

way bigger? Like how many thousands

33:36

of people working for you? So within

33:38

the research division, it's about

33:40

like 3,500 scientists. In a perfect

33:42

world, would you, if it's that big, isn't that

33:45

too small? I think blue?

33:47

Yeah. Well, that's like in the research division. I

33:49

mean, IBM overall. I know, I know. There's thousands

33:52

of people working on that. But I mean, like, so

33:54

starting from first, so we have a, we've

33:57

got a technology that you're ranking

33:59

with. compute and you

34:01

know up there with as a well

34:08

what I'm basically asking is are we under invested

34:10

in this you know

34:12

but so yeah it's a good

34:14

question so like what I would say is that I

34:16

think we should segment how many

34:18

people do you need on the creation

34:21

of the technology itself and what is the

34:23

right size of research and engineers and compute

34:26

to do that and how many people do you

34:28

need in the sort of application

34:31

of the technology to create better products

34:34

to deliver services and consulting and

34:36

then ultimately to diffuse it through you know

34:38

sort of all spheres of society and

34:41

the numbers are very different and that is not different

34:43

than anywhere else I mean I mean if you give

34:45

examples of since you were talking about

34:47

in the context of World War two how many people does

34:49

it take to create you know an atomic

34:52

weapon as an example it's a large number

34:54

I mean it wasn't just Los Alamos there's a lot of

34:56

people in okay it's a large number but it

34:58

wasn't a million people right yeah

35:01

so so you could have highly concentrated

35:03

teams of people that with

35:05

enough resources can do extraordinary scientific

35:08

and technological achievements and that's

35:10

always by definition is going to be a fraction of

35:12

like 1% compared to the total

35:14

volume that is going to require to then deal with it yeah

35:17

but the application side is infinite

35:19

almost that's exactly so that is where

35:21

like in the end the bottleneck really is so

35:24

with you know thousands of

35:26

scientists and engineers you can create world-class

35:29

AI right and so

35:31

no you don't need 10,000 to be able to create

35:33

the large language model in the generative model but you need

35:36

thousands and you need you know very

35:39

significant amount of computing data you need that the

35:42

rest is okay I build

35:44

software I build databases or I build a

35:47

software product that allows you to do inventory

35:49

management or I build you know a photo

35:51

editor and so on now

35:53

that product incorporating

35:55

the AI

35:55

modifying expanding it and so

35:58

on well now you're talking about the

36:00

entire software industry. So now you're talking about

36:02

millions of people, right, who are

36:04

necessary, you know, who are required to

36:06

bring AI into their product. Then you go

36:09

a step beyond the technology creators

36:11

in terms of software and you say, well, okay,

36:13

now what? The skills to help organizations

36:15

go on deployed in the department

36:18

of, you know, the interior, right? And then

36:20

I said, okay, well, now you need like consultants

36:23

and experts and people to work. They are

36:25

to integrate into the workflow. So now you're

36:27

talking into the many tens of millions of people

36:29

around them. So I see it as these concentric

36:32

circles of it. But to some degree

36:34

in many of these core technology areas, just

36:36

saying like, well, I need a team of like a hundred thousand

36:38

people to create like AI or a, or

36:40

a new transistor or a new quantum computer. It's

36:43

actually a diminishing return, right? In the end,

36:45

like many people connecting with each other is very

36:47

difficult. But on the application side,

36:49

I was just thinking about to go back to our, our, our

36:53

example of that college, just

36:55

the task of sitting down with a

36:58

faculty and working with them

37:00

to reimagine what they do with

37:03

this, with these new set of tools in mind,

37:05

with the understanding that the students coming in are probably

37:07

going to know more about it than they do. That

37:10

alone, I mean, that's a, that is a Herculean

37:13

people problem. It's a people problem.

37:16

Yeah. That's why I started in terms of the barriers of

37:18

adoption of that. I mean, the context of IBM, an

37:20

example, that's why we have

37:22

a consulting organization, IBM consulting

37:24

that complements IBM technology. And

37:27

the IBM consulting organization has over 150,000

37:29

employees because of this question, right?

37:33

Because you have to sit down and you say, okay, what

37:35

problem are you trying to solve? What is

37:37

the methodology we're going to do? And here's the technology

37:40

options that we have to be able to bring into the table.

37:42

In the end, the adoption across

37:46

our society will be limited by

37:48

this part. The technology is going

37:50

to make it easier, more cost-effective

37:52

to implement those solutions.

37:55

But you first have to think about what you want to do,

37:58

how you're going to do it, and how you're going to not

38:00

bring it into the life of this in this context faculty

38:02

member or you know the administrator

38:05

and so on in this college. Was that Hollywood that

38:08

notion I thought which was absolutely I

38:12

thought really interesting that in a Hollywood

38:14

strike you have to have this conversation about a distributive

38:17

justice conversation about how do we that

38:20

it's a really hard conversation right to

38:22

have in a boy so this brings me to

38:24

my next point which is that you we were talking backstage you have

38:28

you have two daughters one

38:30

in college one about to go to college that's right so

38:33

they're both science-minded so

38:35

tell me about the conversations you you

38:38

have with your daughter you you have a unique conversation

38:40

with your daughters because your conversation your

38:43

advice to them is is

38:45

influenced by what you do for a living yes

38:48

it's true so did

38:50

you warn your daughters away from certain fields

38:53

did you say whatever you do don't

38:55

be you know no

38:57

no that's not my style I mean for

38:59

me not I try not to be like you know preachy

39:02

about that so for me

39:04

was just about showing by example of things

39:06

I love right and yes I care about

39:09

and then you know bringing them to the lab and seeing

39:11

things and then the natural conversations of things

39:14

working on or interesting people I meet

39:16

so so to the extent that they have chosen

39:18

that and obviously this has an influence on them

39:21

it has been through seeing it you know

39:24

perhaps through my eyes right I'm going to see me

39:26

do and that I like my profession right but one of your

39:28

daughters you said is thinking

39:30

that she wants to be a doctor but

39:33

being a doctor in a post AI world

39:35

is surely a very different proposition than

39:37

being a doctor in a pre AI world do

39:39

you think have you have you tried to prepare

39:42

her for that difference have you

39:44

explained to her what you think will happen to this profession

39:46

she might enter yeah I mean not

39:49

in like you know incredible amount of detail

39:51

but but but yes at the level

39:53

of understanding what is changing like

39:56

this lens of the information lens with

39:58

which you can look at the world what is possible

40:02

and what it can do. Like what is our role

40:04

and what is the role of the technology and how that shapes

40:06

out that level of abstraction for sure.

40:09

But not at the level of like don't be a radiologist,

40:12

you know, because this is what happens. I was gonna

40:14

say, if you're unhappy with your current job, you

40:16

could do a podcast called Parenting Tips with

40:18

Dario, which is just an

40:20

AI person, gives you advice

40:22

of what your kids should do based on exactly

40:24

this, like should I be a radiologist? Dario,

40:27

tell me. I'm sorry guys, it seems to me like

40:29

a really important question. Yeah. Let

40:31

me ask this question in a more, I'm joking, but in a more

40:34

serious way. Surely

40:36

it would, I don't mean to use your

40:38

daughter as an example, but let's imagine we're giving

40:40

advice to someone who wants to enter medicine. A

40:43

really useful conversation to have is, what

40:46

are the skills that will

40:48

be most prized in

40:51

that profession 15 years

40:53

from now? And are they different from the skills that

40:55

are prized now? How would you answer that question?

40:58

Yeah, I think for example,

41:01

this goes back to how is the scientific

41:04

method in this context like the practice

41:06

of medicine gonna change. I think

41:08

we will see more changes on how we practice the

41:10

scientific method and so on as a consequence

41:13

of what is happening with the world

41:15

of computing and information, how we represent

41:18

information, how we represent knowledge, how

41:20

we extract meaning from knowledge as a

41:22

method than we have

41:24

seen in the last 200 years. So

41:26

therefore, what I would strongly encourage

41:29

is not about like, hey, use these tools for doing

41:31

this or doing that, but in the curriculum

41:33

itself, in understanding how we do

41:35

problem solving in the age

41:38

of data and data representation and so

41:40

on, that needs to be embedded in the curriculum

41:43

of everybody that is, I

41:45

would say, actually quite horizontally, but certainly

41:47

in the context of medicine and scientists and so

41:49

on, for sure. And to

41:51

the extent that that gets ingrained, that

41:54

will give us a lens that no matter what

41:56

specialty they go with in medicine,

41:58

they will say, actually.

41:59

The way I want to be able to tackle improving

42:02

the quality of care, the way to do that,

42:04

in addition to all the elements that we

42:06

have practiced in the field of medicine,

42:09

is this new lens. And are we representing

42:11

the data the right way? Do we have the right tools

42:13

to be able to represent that knowledge? Am

42:16

I incorporating that in my own, sort

42:18

of with my own knowledge in a way that gives me better

42:20

outcomes, right? Do I have the rigor of benchmarking

42:24

too and quality of the results?

42:26

So that is what needs to be incorporated. How?

42:29

Well, in a perfect world, if

42:32

I asked you to, your team, to

42:35

rewrite curriculum for American Medical Schools, how

42:38

dramatic a revision is

42:40

that? Are we tinkering with 10% of the curriculum

42:42

or are we tinkering with 50% of it? I

42:46

think there would be a subset

42:49

of classes that is about the method, the

42:51

methodology, what has changed, like have these

42:53

lens of it to understand. And

42:56

then within each class, that

42:59

methodology will represent something that

43:01

is embedded in it, right?

43:03

So it will be substantive,

43:06

but doesn't mean

43:07

replacing the specialization and

43:10

the context and the knowledge of each domain. But

43:12

I do think everybody should have sort

43:14

of a basic knowledge of the horizontal, right?

43:17

What is it? How does it work? What tools

43:19

do you have? What is the technology? And like,

43:21

you know, what are the do's and don'ts around

43:23

that? And then every area you say,

43:25

and you know, that thing that you learn, this is how it applies

43:27

to anatomy.

43:28

And this is how you know, how it applies

43:30

to, you know, radiology if you're studying that or

43:33

this is how you apply, you know, in the context of discovery,

43:36

right, of cell structure. And this is how we can use

43:38

it or protein folding. And this is

43:40

how it does. So that way you'll

43:42

see a connecting tissue throughout

43:44

the whole thing. Yeah. I mean, I

43:46

would add to that, because I was thinking

43:48

about this, that it's

43:52

also this incredible opportunity to do what

43:54

doctors are supposed to do but don't

43:56

have time to do now, which is

43:58

they're so consumed with

43:59

figuring out what's

44:02

wrong with you, that they have little

44:04

time to talk about the implications of

44:06

the diagnosis. Well, we really wondered

44:09

if we can

44:10

free them of some of the burden of

44:12

what is actually quite a prosaic question of what's wrong

44:15

with you, and leave the hard human

44:17

thing of, let me, should you be

44:19

scared or hopeful, should

44:21

you, what do you need to do?

44:24

Let me put this in the context of all the patients I've seen,

44:26

that conversation, which is the most important one, the

44:28

one that seems to me, so

44:31

like if I had to, I would add, if we're

44:33

reimagining the curriculum of

44:35

med school, I'd like, with

44:37

whatever, by the way, very little time,

44:40

maybe we have to add two more years to med school. But

44:43

like a whole. That's not gonna be popular. That's not

44:45

gonna be popular. But the whole thing about bringing

44:48

back the human side of, Yeah.

44:51

you know, now, if I can give you 10 more

44:54

minutes, how do you use that 10 more minutes? But

44:56

in that, in that reconceptualization

44:59

that you just did, is what we should be doing

45:01

around that, because I think the debate as

45:03

to like, well, am I gonna need doctors

45:05

or not, is actually not a very useful debate. But

45:08

rather, this other question is, how is your

45:10

time being spent? What problems are you getting

45:12

stuck? I mean, I generalize this by

45:14

like the obvious observation that if you look

45:16

around in our professions, in our daily lives,

45:19

we have not run out of problems to solve. So

45:21

as an example of that is, hey, if I'm spending

45:23

all my time trying to do diagnosis, and I could do

45:25

that 10 times faster, and it allow me actually

45:28

to go and, you know, and take

45:30

care of the patients and all the next steps of what

45:32

we have to do about it, that's probably a trade

45:34

off that a lot of doctors would take, right?

45:37

And then you say, well, you know, to what degree does

45:39

it allow me to do that? And I can do these other

45:41

things. And these other things are critically important

45:44

for my profession around that. So

45:46

when you actually become less abstract,

45:48

and like we get past the futile

45:50

conversation of like, oh, there's no more jobs,

45:53

and AI is gonna take it all of it, which is kind of nonsense,

45:55

is you go back to to say in practice,

45:58

in your context, right?

45:59

you. What

46:01

does it mean? How do you work? What

46:03

can you do differently around that? Actually that's

46:05

a much richer conversation and very often we would

46:07

find ourselves that there's a portion of the work we

46:09

do that we say I would rather do less of

46:11

that. This other part I like a lot

46:14

and if it is possible that technology

46:16

could help us make that trade-off I'll take it

46:18

in a heartbeat. Now poorly

46:21

implemented technology can also create another

46:23

problem you say hey this was supposed to solve me things

46:26

but the way it's being implemented is

46:28

not helping me right is making my life much

46:30

more miserable or so on or I've lost connection

46:33

in how I used to work etc. So

46:36

that is why design is

46:38

so important that is why also workflow

46:41

is so important in being able to solve these

46:43

problems but it begins

46:45

by you know going from the intergalactic

46:48

to the reality of it of that faculty

46:50

member in the liberal arts college or you know

46:52

or a you know a practitioner in medicine

46:55

in a hospital and what it means for them

46:57

right. Yeah what struck

46:59

me Dario throughout our conversation is how

47:02

much of this revolution

47:05

is non-technical. As I

47:07

say you guys are doing a technical thing here

47:10

but the real the revolution is going to require

47:12

a whole range of people doing things that

47:14

have nothing to do with software

47:17

that have to do with working out new new

47:19

human arrangements. Talking about that

47:21

I mean does I keep going back to the

47:24

Hollywood strike thing that you have to have

47:26

a conversation about our values

47:29

as creators of movies

47:33

how are we going to divide up the credit

47:36

and the like that's a conversation

47:38

about philosophy and you

47:41

know. It is and it's in the grand

47:43

tradition of why you know

47:47

a liberal education is so important

47:49

in the broadest possible sense right.

47:51

There's no common conception

47:54

of the good right. That is always a

47:56

contested dialogue that

47:58

happens within our society and And technology

48:00

is going to fit in that context too. So that's

48:02

why, personally, as a philosophy, I'm not at technological

48:05

determinants. And I don't like

48:07

when colleagues in my profession start

48:10

saying, well, this is the way the technology is going to

48:12

be, and by consequence, this is how society

48:14

is going to be. I'm like, that's a highly contested

48:17

goal. And if you want to enter into a realm

48:19

of politics or a realm of other ones, go

48:21

and stand up on a stool and discuss

48:24

whether that's what society wants, you will find that

48:26

there's a huge diversity of

48:28

opinions and perspective, and that's what makes

48:31

a democracy the richness of our society.

48:34

And in the end, that is going to be the centerpiece

48:36

of the conversation. What do we want? Who

48:40

gets what? And so

48:41

on. And that is, actually, I don't think it's

48:43

anything negative. That's as it should be,

48:45

because in the end, it's anchored of who we

48:47

want as humans, as friends,

48:49

families, citizens. And we have

48:51

many overlapping

48:52

sets of responsibilities. And as a technology

48:54

creator, my only responsibility is not

48:57

just as a scientist and a technology creator. I'm

48:59

also a member of family. I'm a citizen, and I'm

49:01

many other things that I care about.

49:03

And I think that that's sometimes in the debate

49:05

of the technological

49:07

determinists. They start

49:09

now budding into what

49:11

is the realm of justice

49:15

and society and philosophy and democracy.

49:18

And that's where they get the most uncomfortable, because

49:21

it's like, I'm just telling you what's

49:23

possible. And when there's pushback,

49:26

it's like, yeah, but now we're

49:28

talking about how we live

49:29

and how we work and

49:32

how much I get paid or not paid. So

49:34

that technology is important. Technology

49:37

shapes that conversation. But we're going to have

49:39

the conversation with a different language,

49:42

as it should be. And technologies

49:44

need to get accustomed to if they want to participate

49:46

in that world with the broad consequences, hey,

49:49

get accustomed to deal with the complexity of

49:51

that world of politics, society,

49:53

institutions, unions, all that stuff.

49:56

And you can be whiny about it. It's

49:58

like, they're not adopting my technology. That's what

50:00

it takes to bring technology into the world. Yeah.

50:04

Well said.

50:05

Thank you Dario

50:07

for this wonderful conversation.

50:10

Thank you to all of you for

50:12

coming and listening. Thank

50:16

you. Thank

50:18

you. Dario Gill transformed how

50:21

I think about the future of AI.

50:23

He explained to me how huge of a leap

50:25

it was when we went from chess playing

50:27

models to language learning models.

50:30

And he talked about how we still have a

50:32

lot of room to grow. That's why it's important

50:35

that we get things right. The future

50:38

of AI is impossible to predict,

50:40

but the technology has so much potential

50:43

in every industry. Zooming

50:45

into an academic or medical setting showed

50:47

just how close we are to the widespread

50:49

adoption of AI. Even

50:52

Hollywood is being forced to figure this

50:54

out. Humans of all sorts

50:56

will have to be at the forefront of integration

50:59

in order to unlock the full power of AI

51:02

thoughtfully and responsibly. Humans

51:05

have the power and the responsibility to

51:07

shape the tech for our world. I,

51:10

for one, am

51:11

excited to see how things play out.

51:14

Smart Talks with IBM is produced by

51:17

Matt Romano, Joey Fishground, David

51:19

Jha, and Jacob Goldstein. We're

51:22

edited by Lydia Jean Cott. Our engineers

51:24

are Jason Gimbrel, Sarah Bruguire,

51:27

and Ben Tolode. A theme song

51:30

by Gramiscope. Special thanks to

51:32

Andy Kelly, Kathy Callahan, and

51:34

the 8-Bar and IBM teams, as

51:37

well as the Pushkin marketing team. Smart

51:40

Talks with IBM is a production of Pushkin

51:42

Industries and Ruby Studio at

51:44

iHeartMedia. To find more Pushkin

51:46

podcasts, listen on the iHeart

51:49

Radio app, Apple Podcasts, or

51:51

wherever you listen to podcasts.

51:54

I'm Malcolm Dabham. This

51:56

is a paid advertisement from IBM.

51:59

you