0:02

Welcome, Welcome, Welcome to Smart

0:05

Talks with IBM.

0:10

Hello, Hello, Welcome to Smart Talks

0:12

with IBM, a podcast from Pushkin

0:14

Industries, iHeartRadio and

0:16

IBM. I'm Malcolm Gladwell. This

0:19

season, we're diving back into the world

0:21

of artificial intelligence, but with a

0:23

focus on the powerful concept

0:25

of open its possibilities,

0:27

implications, and misconceptions.

0:30

We'll look at openness from a variety of angles

0:33

and explore how the concept is already

0:35

reshaping industries, ways of

0:37

doing business and our very notion of

0:40

what's possible. And for the first episode

0:42

of this season, we're bringing you a special

0:44

conversation. I recently sat

0:46

down with Rob Thomas. Rob is

0:48

the senior vice president of Software

0:51

and chief Commercial Officer of IBM.

0:53

I spoke to him in front of a live audience

0:56

as part of New York Tech Week. We

0:58

discussed how business is can harness

1:00

the immense productivity benefits of AI

1:03

while implementing it in a responsible

1:05

and ethical manner. We also

1:08

broke down a fascinating concept that

1:10

Rob believes about AI, known

1:12

as the productivity paradox. Okay,

1:16

let's get to the conversation. How

1:24

are we doing good?

1:26

Rob?

1:26

This is our our second

1:29

time. We did one of these

1:31

in the middle of the pandemic. But now it's

1:33

all such a blur now that us can figure out when it was.

1:35

I know it's hard to those are like a blurry

1:37

years. You don't know what happened, right.

1:39

But well, it's good to see

1:41

you, to meet you again. I

1:44

wanted to start by going back. You've been at

1:46

IBM twenty years, is that

1:48

right?

1:48

Twenty five in July, believe it or not.

1:51

So you were a kid when you joined.

1:52

I was four.

1:53

Yeah, So

1:56

I want to contrast present

1:58

day Rob and twenty

2:00

five years ago. Rob.

2:03

When you arrive at IBM, what

2:05

do you think your job is going to be? It, your

2:07

career is going. Where do you think the kind of problems

2:10

you're going to be addressing are?

2:13

Well, it's kind of surreal because I joined IBM

2:15

Consulting and I'm coming out

2:17

of school and you

2:20

quickly realize what the job of a consultant

2:22

is to tell other companies what to do. And

2:25

I was like, I literally know nothing, and

2:28

so you're immediately trying to figure out, so how am I going

2:30

to be relevant given that I know absolutely nothing to

2:33

advise other companies on what they should be doing. And

2:36

I remember it well, like we were sitting

2:38

in a room. When you're a consultant,

2:40

you're waiting for somebody else to find work for you. A

2:43

bunch of us sitting in a room, and somebody

2:45

walks in and says, we

2:48

need somebody that knows visio.

2:49

Does anybody know Visio? I'd never

2:51

heard of viseo.

2:52

I don't know if anybody in the room has. So

2:55

everybody's like sitting around looking at their

2:57

shoes. So finally I was like, I

2:59

know it. So I raised my hand. They're

3:01

like, great, we got a project for you next week.

3:04

So I was like, all right, I have like three

3:06

days to figure out what visio is, and

3:10

I hope I can actually figure out how to use it now.

3:12

Luckily, it wasn't like.

3:14

A programming language. I mean, it's pretty much

3:16

a drag and drop capability.

3:19

And so I literally left the office,

3:21

went to a bookstore, bought

3:23

the first three books on Visio I could find, spent

3:26

the whole week in reading the books, and showed

3:29

up and got to work on the project.

3:31

And so it was a bit of a risky

3:33

move, but I

3:35

think that's kind of you against

3:38

this. Well, but if you don't take risk, You'll never

3:40

you'll never achieve, and so does

3:43

some extent. Everybody's making everything

3:45

up all the time. It's like, can you

3:47

learn faster than somebody else? Is

3:49

what the difference is in almost every

3:52

part of life. And so it

3:54

was not planned, but it was an accident, but it

3:56

kind of forced me to figure out that you're gonna

3:58

have to figure things out.

4:00

You know, we're here to talk about AI.

4:02

And I'm curious about the evolution

4:04

of your

4:07

understanding or IBM's understanding of my AI.

4:09

At what point in the last twenty five years

4:11

do you begin to think, oh, this is

4:14

really going to be at the core of what we think

4:16

about and work on at this company.

4:20

The computer scientist John

4:22

McCarthy, he was he's the person that's

4:24

credited with coining the phrase artificial

4:27

intelligence.

4:27

It's like in the fifties.

4:30

And he made an

4:32

interesting comedy said he said, once it works,

4:34

it's no longer called AI, and

4:38

that then became it's called like the AI

4:40

effect, which is it seems very

4:43

difficult, very mysterious, but once it becomes

4:45

commonplace, it's just no

4:47

longer what it is. And so if

4:50

you put that frame on it, I think We've

4:52

always been doing AI at some level, and I

4:54

even think back to when.

4:55

I joined IBM in ninety nine.

4:57

At that point there was work on rules

5:01

based engines, analytics.

5:04

All of this was happening.

5:05

So it all depends on

5:08

how you really define that term. You could

5:10

argue that elements of statistics,

5:14

probability, it's not exactly

5:16

AI, but it certainly feeds into it.

5:18

And so I feel like we've been working

5:21

on this topic of how do we deliver better

5:24

insights better automation

5:27

since IBM was formed. If you read about

5:29

what Thomas Watson Junior did, that was all

5:31

about automating tasks that

5:34

AI well, probably certainly not by today's

5:36

definition, but it's

5:39

in the same zip code.

5:40

So from your perspective, it feels a lot more

5:42

like an evolution than a revolution. Is that a fair

5:44

statement?

5:45

Yes, which I think most

5:47

great things in technology tend

5:50

to happen that way. Many of the revolutions,

5:53

if you will, tend to fizzle out.

5:55

But even given that is there, I guess what I'm

5:57

asking is, I'm curious about whether there was a

6:00

a moment in that evolution when

6:03

you had to readjust your expectations about

6:05

what AI was

6:07

going to be capable of. I mean, was there, you

6:09

know, was there a particular

6:12

innovation or a particular problem

6:15

that was solved that made you think, oh,

6:17

this is different than what I thought.

6:22

I would say the moments that caught

6:24

our attention certainly casper

6:27

Off winning the chess tournament Nobody

6:29

or Deep Blue beating casper

6:31

Off, I should say, nobody really thought

6:33

that was possible before that, and

6:36

then it was Watson

6:39

winning Jeopardy. These were moments that said,

6:41

maybe there's more here than we even thought was possible.

6:45

And so I do think there's points

6:48

in time where we realized

6:50

maybe way

6:52

more could.

6:53

Be done than we had even imagined.

6:56

But I do think it's consistent

6:59

progress every month and every year versus

7:02

some seminal moment.

7:04

Now.

7:04

Certainly large language models

7:06

as of recent have caught everybody's attention because

7:08

it has a direct consumer application.

7:11

But I would almost think of that as

7:15

what Netscape was for the

7:17

for the web browser. Yeah, it brought

7:19

the Internet to everybody, but that

7:22

didn't become the Internet per se.

7:25

Yeah.

7:25

I have a cousin who worked for IBM

7:28

for forty one years. I saw him this weekend.

7:30

He's in Toronto, by the way, I said,

7:32

do you work for Rob Thomas? He

7:35

went like this, He goes, he

7:39

said, I'm five layers down. But

7:43

so I always whenever I see my cousin, I ask him,

7:45

can you tell me again what you do? Because it's always changing,

7:47

right, I guess this is a function of working at IBM.

7:50

So eventually he just gives up and says,

7:53

you know, we're just solving problems. So what we're doing, which

7:55

I sort of loved as a kind of frame,

7:58

And I was curious, what's what's the coolest

8:00

problem you ever worked on? Not biggest, not

8:02

most important, but

8:05

the coolest, the one that's like that

8:07

sort of makes you smile when you think back on it.

8:09

Probably when I was in microelectronics,

8:12

because it was a world

8:14

I had no exposure to. I hadn't studied

8:16

computer science, and

8:19

we were building a lot of high

8:22

performance semiconductor technology,

8:24

so just chips that do a really great

8:27

job of processing something or

8:29

other. And we

8:31

figured out that there was a market in consumer

8:34

gaming that was starting to happen, and

8:37

we got to the point where we became

8:39

the chip inside the Nintendo. We

8:43

the Microsoft

8:45

Xbox Sony PlayStation, so

8:47

we basically had the entire gaming market running

8:50

on ib and chips and.

8:52

To use every parent basically

8:55

is pointing at you and saying.

8:57

You're the Probably

9:00

well, they would have found it from anybody. But it

9:03

was the first time I could explain

9:06

my job to my kids, who were quite young at that time,

9:09

like what I did, Like it was more

9:11

tangible for them than saying we solve

9:13

problems or douce you know, build solutions like

9:15

it became very tangible for them,

9:18

and I think that's, you know,

9:20

a rewarding part of the job is when you can help

9:23

your family actually understand what you do. Most people can't

9:25

do that. It's probably easier for you. They can, they can see the

9:27

books, but for

9:30

for some of us in the business the business world,

9:32

it's not always as obvious. So that was like one example

9:35

where the dots really connected.

9:38

There were a couple

9:40

there's a couple of stuck about a little bit of this in the context

9:42

of of AI. I love because I love

9:44

the frame of problem solving

9:47

as a way of understanding what the function

9:49

of the technology is. So I know that you

9:51

guys did something, did some work with

9:55

I never know how to pronounce it

9:57

is it Sevilla Sevilla with

9:59

the football club Severe in Spain. Tell

10:01

me about tell me a little

10:03

bit about that. What problem were they trying to

10:05

solve and why did they call you?

10:07

In Every

10:11

sports franchise is

10:14

trying to get an advantage, right, Let's just be that clear.

10:16

Everybody's how can I use data,

10:19

analytics, insights, anything

10:22

that will make us one percent better on

10:24

the field at

10:26

some point in the future. And

10:30

Seville reached out to us because

10:32

they had seen some of the We've done some work

10:34

with the Toronto Raptors in the past and others,

10:37

and their thought

10:39

was maybe there's something we could do. They'd heard all about

10:43

generative AI, they heard about large language

10:45

models.

10:46

And the problem, back to.

10:47

Your point on solving

10:49

problems, was we want to do a way

10:51

better job of assessing

10:53

talent, because really

10:56

the lifeblood of a sports franchise

10:58

is can you continue to cult a talent?

11:01

Can you find talent that others don't

11:03

find? Can you see something in somebody

11:05

that they don't see in themselves or maybe no other.

11:08

Team season them?

11:09

And we ended up building somed

11:12

with them called Scout Advisor, which

11:14

is built on Watson X, which

11:17

basically just ingests tons

11:20

and tons of data, and we

11:23

like to think of it as finding, you know, the needle

11:25

in the haystack of you know, here's

11:28

three players that aren't being considered.

11:30

They're not on the top teams

11:32

today, and I think

11:35

working with them together we found some pretty good insights

11:37

that's helped them out.

11:38

How What was interesting to me was we're

11:40

not just talking about quantitative

11:43

data. We're also talking about qualitative

11:45

data. But that's the puzzle

11:47

part of the thing that fastens me. How does one

11:49

incorporate qualitative analysis into

11:51

that sort of so you just feeding

11:54

in scouting reports and things like

11:56

that.

11:58

I got to realize think about how much I can act actually

12:00

disclosed it.

12:03

But if you think about so, quantitative

12:06

is relatively easy.

12:08

Every team collects that, you

12:11

know, what's their forty yard

12:13

dash? They use that term, certainly not in Spain.

12:16

That's all quantitative. Qualitative is

12:19

what's happening off the field. It

12:22

could be diet, it could be habits, it

12:24

could be behavior. You

12:26

can imagine a range of things that would all

12:28

feed into an athlete's

12:31

performance and so relationships.

12:35

There's many different aspects, and.

12:37

So it's trying to figure out the

12:39

right blend of quantitative and qualitative

12:42

that gives you a unique insight.

12:44

How transparent is that kind of system? I

12:46

mean, is it telling you it's saying

12:49

pick this guy not this guy, But is it telling

12:51

you why it prefers this guy to this guy?

12:53

Is that?

12:54

I think for anything in the realm of AI, you

12:57

have to answer the why question, otherwise

12:59

you fall into the trap of the

13:03

you know, the proverbial black box, and

13:05

then wait, I made this decision, I'd never

13:07

understood why it didn't work out.

13:09

So you always have to answer why without

13:11

a doubt?

13:12

And how is why? Answered?

13:16

Sources of data, the reasoning

13:19

that went into it, and so it's

13:21

basically just tracing back the

13:23

chain of how you got to the answer. And

13:26

in the case of what we do in Watson X is

13:28

we have IBM models. We also

13:30

use some other open source models, So it

13:32

would be which model was used, what

13:35

was the data set that was fed into that model, How

13:37

is it making decisions?

13:38

How is it performing? Is

13:40

it robust?

13:42

Meaning is it reliable in terms of if you

13:44

feed it two of the same data set, do you get

13:46

the same answer. These are all the

13:48

you know, the technical aspects of understanding

13:50

the why.

13:52

How quickly do you expect all

13:54

professional sports franchises to adopt

13:56

some kind of are they already there? If I went out

13:58

and pulled the general managers of

14:01

the one hundred most valuable sports

14:03

franchises in the world, how many of them would be using

14:05

some kind of AI system to assist

14:08

in their efforts.

14:10

One hundred and twenty percent would, meaning

14:13

that everybody's doing it, and some think

14:15

they're doing way more than they probably actually are. So

14:18

everybody's doing it. I think what's weird

14:20

about sports is everybody's

14:23

so convinced that what they're doing is

14:25

unique that they

14:28

generally speaking, don't want to work with a third party

14:30

to do it because they're afraid that that

14:32

would expose them. But in reality,

14:35

I think most are doing eighty to ninety

14:37

percent of the same things.

14:39

So but without a doubt, everybody's doing

14:41

it. Yeah.

14:43

Yeah. The other

14:45

I say that I loved was there was one but a

14:48

shipping line tricon on the Mississippi

14:50

River. Tell me a little bit about

14:53

that project. What problem were they trying to solve?

14:56

Think about the problem that I

14:59

would say every Boddy noticed if you go back

15:01

to twenty twenty, was things

15:04

are getting hold held up in ports. It

15:06

was actually an article in the paper this morning kind of

15:08

tracing the history of what happened twenty

15:10

twenty twenty one and

15:12

why ships were basically sitting at seas

15:14

for months at a time. And

15:17

at that stage we just we had a massive

15:19

throughput issue. But moving

15:23

even beyond the pandemic, you can see it now

15:26

with ships getting through like

15:28

Panama Canal, there's like a narrow

15:31

window where you can get through, and if

15:33

you don't have your paperwork done,

15:36

you don't have the right approvals, you're not going through

15:38

and it may cost you a day or two and that's a lot of money.

15:41

In the shipping industry and the tricon

15:43

example, it's really just about

15:46

when you're pulling into a port,

15:49

if you have the right paperwork done,

15:52

you can get goods off the ship very

15:54

quickly. They ship a

15:57

lot of food, which by definition,

16:00

since it's not packaged food, it's fresh food,

16:02

there is an expiration period and

16:05

so if it takes them an extra two

16:07

hours, certainly multiple

16:10

hours or a day, they have a massive

16:12

problem because then you're going to deal with spoilage and

16:15

so it's going to set you back. And what

16:17

we've worked with them on is using

16:20

an assistant that we've built in Watson

16:22

x called orchestrate, which basically

16:25

is just AI doing digital

16:28

labor, so we can replicate

16:31

nearly any.

16:32

Repetitive task and

16:34

do that with software instead of humans.

16:37

So, as you may imagine, shipping

16:39

industry still has a lot of paperwork that

16:42

goes on, and so being able to

16:44

take forms that normally would be multiple

16:46

hours of filling it out, Oh this isn't right, send

16:48

it back. We've basically built

16:50

that as a digital skill inside

16:53

of watsonex orchestrate, and

16:55

so now it's done in minutes.

16:58

Did they really did Did they realize

17:00

that they could have that kind of efficiency

17:02

by teaming up with you? Or is that something you came to them

17:05

and said, guys,

17:08

we can do this way better than you think.

17:09

What's the.

17:11

I'd say, it's always, it's always

17:14

both sides coming together at a moment

17:16

that for some reason makes sense because

17:19

you could say, why didn't this happen like five years ago, like

17:22

seems so obvious. Well, technology wasn't

17:24

quite ready then, I would say,

17:27

But they knew they had a need because

17:29

I forget what the precise number is, but you

17:32

know, reduction of spoilage has massive

17:35

impact on their bottom line, and

17:38

so they knew they had a need, we.

17:41

Thought we could solve it, and the

17:43

two together.

17:44

Who did you guys go to them thought?

17:47

Or did they come to you?

17:48

I recall that this one was an inbound

17:51

meaning they had reached out to IBM

17:54

and that we'd like to solve this problem. I think

17:56

it went into one of our digital centers, if

17:58

I if I recall so literary.

18:00

Call, yeah, but the other

18:02

the reverse is more

18:04

interesting to me because there seems to be a

18:07

very very large universe of people who have

18:09

problems that could be solved this way and

18:11

they don't realize it.

18:13

What's your.

18:15

Is there a shining example of this of

18:17

someone you just can't you just think could

18:19

benefit so much and isn't benefiting right

18:21

now?

18:24

Maybe I'll answer it slightly differently.

18:26

I'm I'm surprised by

18:29

how many people can benefit that you wouldn't even

18:32

logically think of.

18:33

First, let me give you an example.

18:35

There's a

18:38

franchiser of hair salons,

18:41

sport Clips is the name. My

18:44

sons used to go there for haircuts because they have like TVs

18:46

and you can watch sports, so they loved

18:49

that. They got entertained while they would get their haircut. I

18:52

think the last place that you would think is using

18:54

AI today would be a franchiser

18:57

of hair salons. Yeah,

18:59

but just follow it through. The

19:02

biggest part of how they run

19:04

their business is can I get people to cut hair?

19:08

And this is the high turnover industry because

19:10

there's a lot of different places you can work if you want to cut

19:12

hair. People actually get injured cutting hair

19:14

because you're on your feet all day, that type of thing. And

19:18

they're using same technology orchestrate

19:21

as part of their recruiting process.

19:24

How can they automate a lot of people submitting

19:26

resumes, who they speak

19:28

to, how they qualify.

19:30

Them for the position.

19:32

And so the reason I give that example

19:34

is the opportunity for AI, which

19:37

is unlike other technologies,

19:39

is truly unlimited. It

19:42

will touch every single business.

19:45

It's not the realm of the fortune five hundred

19:47

or the fortune one thousand. This

19:50

is the fortune any size.

19:52

And I think that may be one thing that people underestimate

19:55

about AI.

19:56

Yeah, what about I mean I was thinking

19:58

about education as as a kind of I

20:01

mean, education is a perennial whipping

20:06

boy for you guys that are living

20:08

in the nineteenth century, right. I'm just curious

20:10

about if a superintendent

20:14

of a public school system or the president of the

20:16

university sat down and had lunch

20:18

with you and said, do

20:21

the university first. My cost are out of control,

20:24

my enrollment

20:26

is down, my students hate

20:28

me, and my board is revolting.

20:31

Help.

20:33

How would you think about

20:36

helping someone in that situation.

20:39

I spend some time with universities. I

20:41

like to go back and there's.

20:42

Alma maters

20:44

where I went to school, and so

20:46

I do that every year. The challenge

20:49

I have hall of Seeming University is there has to be

20:51

a will. Yeah, and I'm

20:53

not sure the incentives are quite right today because

20:58

bringing in new technology, say we want

21:00

to go after we can help you figure out student

21:02

recruiting or

21:05

how you automate more of your education,

21:09

everybody suddenly feels threatened that university.

21:11

Hold on, that's my job.

21:13

I'm the one that decides that, or I'm

21:15

the one that wants to dictate the course. So

21:18

there has to be a will. So

21:20

I think it's very possible, and

21:23

I do think over the next decade you

21:25

will see some universities that jump all over

21:27

this and they will move ahead, and you see

21:30

others that do not.

21:31

Because it's very possible.

21:35

Where how does when you say there

21:37

has to be a will. Is that

21:39

the kind? Is that a kind of thing that

21:41

that people that IBM to think about, Like

21:45

when in this conversation you hypothetical conversation

21:47

you might have with the university president, would

21:49

you give advice on where

21:52

the will comes from?

21:55

I don't do that as much in a university context.

21:57

I do that every day in a business context, because

22:02

if you can find the right person in a business

22:04

that wants to focus on growth

22:07

or the bottom line or how do you create

22:09

more productivity. Yes, it's going to create

22:11

a lot of organizational resistance

22:14

potentially, but you can find somebody that will

22:16

figure out how to push that through. I

22:19

think for universities, I

22:21

think that's also possible. I'm not sure

22:23

there's there's there's a return on investment

22:26

for us to do that.

22:27

Yeah, yeah, yeah, God,

22:30

let's let's find some terms AI

22:34

years I told you'd

22:36

like to use what does that mean?

22:39

We just started using this term literally

22:41

in the last three months, and

22:45

it was it was what we observed internally,

22:48

which is most technology

22:50

you build, you say, all right, what's going to happen in year

22:52

one, year two, year three, and

22:55

it's you know, largely by by

22:57

a calendar AI years are the idea

23:00

that what used to be a year is

23:02

now like a week. And

23:04

that is how fast the technology is moving.

23:07

And do you give you an example. We had one

23:09

client we're working with.

23:11

They're using one of our granite

23:13

models, and the results they were getting we're not

23:15

very good. Accuracy was not there, their

23:18

performance was not there. So I

23:20

was like scratching my head. I was like, what is going on? They

23:23

were financial services, the

23:25

bank, So I'm scratching my head, like what is going

23:27

on? Everybody else is getting this and like these

23:30

results are horrible. And I

23:32

said to the team, which version of the model

23:35

are you using? This was in

23:37

February, Like we're using the one from October.

23:41

I was like, all right, now we know precisely the problem

23:44

because the model from October is

23:46

effectively useless now since we're here in February.

23:49

Serious, actually useless,

23:52

completely useless.

23:53

Yeah, that is how fast this is

23:55

changing. And so the minute, same

23:58

use case, same day, you

24:00

give them the model from late

24:03

January instead of October,

24:05

the results are off the charts.

24:07

Yeah.

24:07

Wait, so what exactly happened between October

24:10

and January?

24:10

The model got way better?

24:12

Could dig into that, like what do you mean by the way.

24:14

We are constant.

24:15

We have built large

24:17

compute infrastructure where we're doing model

24:19

training. And to be clear, model

24:22

training is the realm of probably in

24:25

the world my guess is five to ten companies.

24:28

And so.

24:30

You build a model, you're constantly training

24:33

it, you're doing fine tuning, you're

24:35

doing more training, you're adding data every

24:37

day, every hour it gets better. And

24:40

so how does it do that. You're feeding

24:42

it more data, you're feeding it more

24:45

live examples. We're

24:47

using things like synthetic data at this point,

24:49

which is we're basically creating data to do the training

24:52

as well. All of this feeds into

24:54

how useful the model is. And

24:56

so using the October

24:59

model, those were the results in October, just

25:01

a fact, that's how good it was then. But

25:04

back to the concept of AI years, two

25:07

weeks is a long time.

25:10

Is that are we in a steep

25:12

part of the model learning carve or do you expect

25:14

this to continue along this at

25:16

this pace?

25:19

I think that is the big question and

25:23

don't have an answer yet.

25:24

By definition, at some point you would think it would

25:26

have to slow down a bit, but it's not obvious

25:29

that that is on the horizon.

25:31

Still speeding up. Yes, how

25:33

fast can it get?

25:37

We've debated, can you actually have

25:39

better results in the afternoon than you did in the morning.

25:42

Really it's nuts.

25:44

Yeah, I know, but that's why

25:47

we came up with this term, because I think you also

25:49

have to think of like concepts that.

25:53

Gets people's attention.

25:54

So you're basically turning into a bakery.

25:56

You're like the bread from yesterday.

25:59

You know you can have it for twenty five cents. But

26:02

I mean you do proferential pricing. You could

26:04

say, we'll judge you

26:06

x for yesterday's model, two

26:09

x for today's model.

26:12

I think that's dangerous as a

26:14

merchandising strategy, but I guess your point.

26:17

Yeah, but that's crazy.

26:19

And this, by the way, so this model is the same

26:21

true for almost You're talking specifically about

26:23

a model that was created to help

26:26

some aspect of a financial services.

26:29

So is that kind of model accelerating

26:31

faster and learning faster than other models for other

26:34

kinds of problems?

26:35

So this domain was code,

26:38

Yeah, and so by

26:40

definition, if you're feeling feeding in more data,

26:43

some more code, you get those kind of results.

26:46

It does depend on the model type. There's

26:49

a lot of code in the world and so we

26:51

can find that we can create it. Like I said,

26:55

there's other aspects where there's probably

26:57

less inputs available, which

26:59

means you probably won't get the same level of iteration.

27:02

But for code, that's certainly the cycle times that we're

27:04

seeing.

27:05

Yeah, and how do you know that Let's

27:07

stick with this one example of this model you have.

27:10

How do you know that your model is better

27:12

than big company

27:14

b down the street? Client

27:16

asks you, why would I go with IBM as opposed to

27:20

some the some firm in the valley that says,

27:22

let's they have a model on this, what's your how

27:24

do you frame your advantage?

27:28

Well, we benchmark all of this, and

27:31

I think the most important is metric

27:33

is price performance, not

27:35

price, not performance, but the combination of

27:37

the two.

27:38

And we're super competitive there.

27:41

Well, for what we just released, with

27:43

what we've done in open source, we know that nobody's

27:46

close to us right now on code.

27:47

Now.

27:48

To be clear, that will probably change because

27:50

it's like leapfrog.

27:51

People will jump ahead, then we jump back

27:53

ahead.

27:54

But we're very confident

27:56

that with everything we've done

27:59

in the last few months taken a huge lead

28:01

forward here.

28:01

Yeah, it's I mean this

28:04

goes back to the point I was making in the beginning, so

28:06

about the difference between your twenty

28:09

something self in ninety nine and yourself

28:11

today. But this time compression

28:15

has to be a crazy adjustment. So

28:18

the concept of what you're working on and

28:20

how you make decisions internally and things has

28:23

to undergo this kind of revolution.

28:25

If you're switching from I mean back

28:27

in the day, a model might be useful for how

28:31

long.

28:31

Years years I think about you

28:34

know, statistical models that set inside

28:36

things like SPSS,

28:38

which is a product that a lot of.

28:40

Students use around the world.

28:41

I mean, those have been the same models for twenty years

28:44

and they're still very good at what they do. And

28:46

so yes, it's a completely it's

28:49

a completely different moment

28:51

for how fast this is moving. And I think

28:54

it just raises the bar for everybody,

28:56

whether you're a technology

28:58

provider like us, or you're

29:01

a bank or an insurance company or a

29:03

shipping company, to say, how

29:05

do you really change your

29:07

culture to be way more aggressive

29:11

than you normally would be?

29:14

Does this means it's a weird question, But does

29:17

this mean a different set of kind

29:19

of personality or character traits

29:21

are necessary for a decision maker in

29:24

tech now than twenty five years ago.

29:29

There's a book I saw recently,

29:32

it's called The Geek Way, which talked

29:34

about how technology companies

29:36

have started to operate in

29:38

different ways, maybe than many

29:41

traditional companies, and

29:45

more about being data driven, more

29:48

about delegation. Are

29:51

you willing to have the

29:53

smartest person in the room make decisions opposed

29:55

to the highest paid.

29:56

Person in the room.

29:57

I think these are all different aspects that

29:59

ever company.

30:00

Is going to face.

30:01

Yeah, yeah, next

30:04

term, talk about open. When you

30:06

use that word open, what do you mean.

30:10

I think there's really only one definition of

30:12

open, which is for technology

30:14

is open source. An

30:17

open source means the code

30:19

is freely available. Anybody

30:22

can see it, access it, contribute

30:26

to it.

30:26

And what is Tell me about why that's an important

30:28

principle.

30:32

When you take a topic like AI. I

30:35

think it would be really bad for the world

30:39

if this was in the hands of one or two companies,

30:43

or three or four, doesn't matter the number, some small

30:46

number. Think about like in

30:48

history sometimes early nineteen hundreds,

30:51

the Interstate Commerce Commission

30:53

was created, and the whole idea

30:55

was to protect farmers

30:57

from railroads, meaning they

31:00

wanted to allow free trade. But they

31:02

knew that well, there's only so many railroad tracks,

31:04

so we need to protect farmers from

31:06

the shipping costs that railroads could impose.

31:09

So good idea, but over time

31:12

that got completely overtaken by the railroad lobby

31:15

and then they use that to basically

31:17

just increase prices, and it

31:19

made the lives of farmers way more

31:21

difficult. I think you

31:23

could play the same analogy through with AI.

31:27

If you allow a handful of companies

31:29

to have the technology, you

31:31

regulate around the principles of those

31:33

one or two companies, then you've trapped the entire

31:35

world.

31:36

Think that would be very bad. So

31:39

the danger of that app for

31:42

sure.

31:42

I mean there's companies in Watson in

31:44

Washington every week trying to

31:47

achieve that outcome.

31:49

And so the.

31:50

Opposite of that is to say it's going to be an

31:52

open source because

31:54

nobody could dispute open source because it's

31:57

right there, everybody can see it. So

32:00

I'm a strong believer that open source will win for

32:02

AI. It has to win. It's not

32:05

just important for business, but it's important

32:07

for humans.

32:10

On the I'm curious about

32:12

on the list of things you worry about, Actually,

32:16

let me before I ask, let me ask this question very

32:18

generally, what is the list of things you worry

32:20

about? What's your top five business

32:22

related worries right now?

32:25

Tops from those are the first question. We

32:27

could be here for hours for me to answer.

32:30

I did say business related. We could leave. You know, your

32:34

kids' haircuts got it out of.

32:36

The Number

32:38

one is always It's the

32:40

thing that's probably always been true, which

32:42

is just people. Do

32:45

we have the right skills? Are we doing a good

32:48

job of training our people? Are

32:50

our people doing a good job of working with clients

32:53

like That's number one? Number

32:55

two is innovation? Are

32:59

we pushing the envelope enough? Are

33:02

are we staying ahead? Number

33:05

three is which kind of feeds into

33:07

the innovation one is risk taking? Are

33:09

we taking enough risk? Without

33:11

risk, there is no growth. And

33:13

I think the trap that every larger

33:15

company inevitably

33:18

falls into is conservatism.

33:21

Things are good enough, and

33:23

so it's are we pushing the envelope?

33:25

Are we taking enough risk to

33:27

really have an impact? I'd say those are probably

33:29

the top three that I spend talk

33:32

about.

33:32

The vast trend to define productivity paradox

33:35

something I know you've thought a lot about what does

33:37

that mean?

33:39

So I started thinking hard about this because all

33:41

I saw and read every day was

33:44

fear about AI, and

33:48

I studied economics, and

33:51

so I kind of went back to like basic

33:54

economics, and there's been like a macro investing

33:58

formula I guess I would say it's

34:00

been around forever that says growth

34:02

comes from

34:05

productivity growth plus

34:08

population growth plus

34:10

debt growth. So

34:13

if those three things are working, you'll

34:15

get GDP growth. And

34:17

so then you think about that and you say, well, debt

34:20

growth, we're probably not going

34:22

back to zero percent interest rates, so

34:25

to some extent there's going to be a ceiling on that.

34:28

And then you.

34:29

Look at population growth. There

34:31

are shockingly few countries

34:33

or places in the world that will see population growth

34:36

over the next thirty to fifty years. In

34:38

fact, most places are not even at

34:40

replacement rates. And

34:43

so I'm like, all right, so population growth is not going to be there.

34:46

So that would mean if you just take.

34:48

It to the extreme, the

34:50

only chance of continued

34:53

GDP growth is

34:55

productivity.

34:57

And the best way to

35:01

solve productivity as AI.

35:03

That's why I say it's a paradox.

35:05

On one hand, everybody's scared after

35:07

death it's going to

35:09

take over the world, take all of our

35:11

jobs, ruin us, But

35:14

in reality, maybe it's the other way, which is it's

35:16

the only thing that can save us.

35:18

Yeah, and if you believe.

35:20

That economic equation, which I think has proven

35:23

quite true over hundreds of years, I

35:25

do think it's probably the only thing that can save us.

35:28

Actually looked at the numbers yesterday for

35:30

total random reason on population growth

35:33

in Europe and receive this is

35:35

a special bonus question. See how smart you are?

35:37

Which country in Europe continentally

35:40

Europe has the highest population growth?

35:43

It's small continental Europe,

35:48

probably one of the Nordics, I would guess.

35:50

Close Luxembourg.

35:53

Okay, something that's going on in Luxembourg. I

35:57

feel like, well, all of this need to investigate.

36:00

There're at one point four nine, which in the day, by

36:02

the way, would be a relatively that's

36:04

the best performing country. I mean in the

36:06

day, you'd countries had routinely had two

36:09

points something, you know, percent

36:11

growth in a given year. Last

36:14

question, you're writing a book. Now, we were talking

36:16

chatting about it backstage, and now

36:18

I appreciate the paradox

36:20

of this book, which is universe

36:23

with a model is better in the afternoon than it is in

36:25

the morning. How do you write a book that's like

36:27

printed on paper? I expected

36:29

to reuse Aul.

36:34

This is the challenge. And I am

36:37

an incredible author of useless books.

36:39

I mean most of what I've spent time

36:41

on in the last decade of stuff that's completely

36:44

useless, like a year after it's written. And

36:47

so when we

36:49

were talking about it, I was like, I would like to do something around

36:51

AI that's timeless. Yeah,

36:54

that would be useful ten or twenty

36:56

years from now. But

36:58

then to your so, how

37:01

is that even remotely possible if

37:04

the model is better in the afternoon and in the morning.

37:07

So that's the challenge in front of us.

37:09

But the book is around AI value creation, so

37:12

kind of links to this productivity paradox,

37:14

and how do you actually get

37:17

sustained value out

37:19

of AI, out

37:22

of automation, out of data

37:24

science? And so the biggest

37:26

challenge in front of us is can we make this relevant?

37:30

How's the day that it's published?

37:31

How are you setting out to do that?

37:35

I think you have to to some extent level

37:38

it up to bigger concepts, which

37:40

is kind of why I go to things like macroeconomics,

37:43

population geography

37:45

as opposed to going into the weeds

37:48

of the technology itself. If you write

37:50

about this is how you get better performance

37:52

out of a model we can

37:55

agree that will be completely useless

37:57

two years from now, but maybe even two months

37:59

from now, and so it will

38:01

be less in the technical

38:03

detail and more of what

38:06

is sustained value creation for AI, which

38:09

if you think on what is hopefully a

38:11

ten or twenty year period, it's probably

38:14

we're kind of substituting AI for technology.

38:16

Now I've realized, because I think this has always

38:18

been true for technology. It's just now

38:21

AI is I think that everybody wants to talk about.

38:25

But let's see if we can do it. Time will

38:27

tell.

38:28

Did you get any inkling that the pace

38:30

that this AI year's phenomenon

38:32

was gonna that things with the pace

38:34

of change was going to accelerate so much because

38:37

you had More's law, right, You had a model in

38:40

the technology world for this

38:42

kind of exponential increase in so

38:45

were you were you

38:47

thinking about that kind of accelerate

38:50

similar kind of acceleration in

38:52

the.

38:55

I think anybody that said they expected

38:57

what we're seeing today is probably exactly.

39:01

I think it's way faster than

39:03

anybody expected.

39:05

Yeah, but technologies,

39:08

back to your point at More's law has always

39:10

accelerated through the years,

39:12

So I wouldn't say it's a shock, but

39:15

it is surprising.

39:16

Yeah, You've had a kind of extraordinary

39:20

privileged position to watch

39:23

and participate in this revolution, right, I

39:25

mean, how many other people have been in that

39:27

have ridden this

39:30

wave like you have?

39:32

I do wonder is this really

39:34

that much different or does it feel different just

39:36

because we're here? I mean

39:39

I do think on one level, yes, So

39:41

in the time I've been an IBM, Internet happened,

39:45

mobile happened, social

39:48

network happened, blockchain

39:50

happened.

39:51

AI, So a lot has happened.

39:53

But then you go back and say, well, but if I'd been here between

39:57

nineteen seventy and ninety five, there

40:00

were a lot of things that are pretty fundamental

40:03

then too, say, I wondered, almost,

40:04

do we do we always exaggerate

40:06

the timeframe that we're in. I

40:10

don't know.

40:11

Yeah, but it's

40:13

a good idea though.

40:16

I think the ending with the phrase

40:18

I don't know it's a good idea

40:20

though. Great

40:23

way to wrap this up.

40:24

Thank you so much, Thank you, Malcolm.

40:29

In a field that is evolving as quickly as artificial

40:32

intelligence, it was inspiring

40:34

to see how adaptable Rob has been over

40:36

his career. The takeaways from

40:38

my conversation with Rob had been echoing

40:41

in my head ever since. He

40:43

emphasized how open source models

40:45

allow AI technology to be developed

40:47

by many players. Openness

40:50

also allows for transparency.

40:52

Rob told me about AI use cases

40:55

like IBM's collaboration with

40:57

Sevilla's football club. That exam

41:00

really brought home for me how AI

41:02

technology will touch every industry.

41:05

Despite the potential benefits of AI,

41:08

challenges exist in its widespread

41:10

adoption. Rob discussed how

41:12

resistance to change, concerns

41:15

about job security and organizational

41:17

inertia can slow down implementation

41:20

of AI solutions. The

41:22

paradox, though, according to Rob, is

41:24

that rather than being afraid of a world with

41:27

AI, people should actually be more

41:29

afraid of a world without it. AI,

41:32

he believes, has the potential to make

41:34

the world a better place in a

41:36

way that no other technology can. Rob

41:39

painted an optimistic version of

41:41

the future, one in which AI technology

41:44

will continue to improve at

41:46

an exponential rate. This

41:48

will free up workers to dedicate their

41:50

energy to more creative tasks.

41:53

I for one am on board

41:57

Smart Talks with IBM is produced by Matt

41:59

Romano, Joey Fishground, and

42:01

Jacob Goldstein. We're edited

42:03

by Lydia Gene kott Our engineers

42:06

are Sarah Bruguer and Ben Tolliday.

42:08

Theme song by Gramscow. Special

42:11

thanks to the eight Bar and IBM teams,

42:13

as well as the Pushkin marketing team.

42:15

Smart Talks with IBM is a production

42:18

of Pushkin Industries and Ruby Studio

42:20

at iHeartMedia. To find more

42:23

Pushkin podcasts, listen on the

42:25

iHeartRadio app, Apple Podcasts,

42:28

or wherever you listen to podcasts.

42:31

I'm Malcolm Gladwell. This is a paid

42:33

advertisement from IBM. The

42:35

conversations on this podcast don't

42:38

necessarily represent IBM's

42:40

positions, strategies, or

42:42

opinions.