0:00

This message comes from Insperity. Say

0:02

to stress -free subscription management.

0:04

Easily track, block, or

0:06

cancel recurring charges right from

0:08

the Capital One mobile

0:10

app. Simple as that. Learn

0:13

Learn more at capitalone.com slash

0:15

subscriptions. and Terms and conditions

0:17

apply. message comes message comes

0:19

from Insperity. Providing HR and

0:22

and technology from payroll,

0:24

benefits, HR and HR

0:26

compliance to talent development.

0:28

Learn more at at.com slash

0:30

HR slash HR matters. This is

0:33

the TED Radio Hour. Each week,

0:35

week, groundbreaking TED

0:37

Our job Our job now is to dream big. delivered

0:40

at TED to bring about

0:42

the future we want to see

0:44

around the world to understand who

0:46

we are we are. those talks, we

0:48

bring you speakers and ideas that

0:50

will surprise you You just don't know

0:52

what you're gonna find. find. Challenge you. have

0:54

to act to have to ask is it noteworthy?

0:56

and even change you. I literally

0:58

feel like I'm a different

1:00

person. feel like I'm a Do you feel

1:03

that way? Do you Ideas worth

1:05

spreading. spreading. From Ted and NPR.

1:07

Zamorodi. It was about

1:10

a year after 9-11. It

1:12

was about a year after Suleman.

1:14

In is Mustafa was a

1:17

In 2002, Mustafa was a

1:19

student at Oxford University

1:21

when a friend decided to

1:23

start a helpline for

1:25

young British Muslims. A good good

1:27

friend of mine at the

1:29

time at the just started the

1:31

first prototype One evening session. I

1:33

think it was a Thursday

1:36

evening of offering offering

1:38

counselling services on the

1:40

phone. Mustafa was

1:42

intrigued. He ended up

1:44

joining the project to help grow it. we became

1:47

we became essentially

1:49

co -founders. service their service

1:51

was in hot demand. was a

1:53

know, it a pretty remarkable time time

1:55

because, know, young British Muslims were

1:58

feeling judged. judged by...

2:00

9-11 as though they were responsible

2:02

or somehow complicit. There was a

2:04

lot of Islamophobia. But what they

2:07

found these young people needed was

2:09

really just someone to talk to.

2:11

Most of the challenges that they

2:14

were working through were, you know,

2:16

the sort of conventional things that

2:19

a young person might deal with,

2:21

like relationships, bullying, self-harm, increasing homophobia

2:23

in the community, and... Fundamentally, I

2:26

think people were just looking to

2:28

find support in a language that

2:30

made sense to them, you know,

2:33

like a London accent using sort

2:35

of Islamic terminology and having a

2:37

broader understanding of the, you know,

2:40

whether it was a Bangladeshi or

2:42

Pakistani culture or an Arab culture,

2:45

combined with the religious foundation and

2:47

just being able to reflect back

2:49

that empathy went a long way.

2:53

20 years later, the Muslim

2:55

youth helpline is still going

2:57

strong. And Mustafa Suleiman now

2:59

lives in Silicon Valley, where

3:01

he's been working on another

3:03

kind of helper. I'm the

3:05

CEO of Microsoft AI, and

3:08

I work on building new

3:10

AI systems and products. These

3:12

days, I'm working on an

3:14

AI companion. With artificial

3:16

intelligence and the backing of one

3:18

of the biggest tech companies in

3:21

the world Mustafa Suleiman wants us

3:23

each to have our own digital

3:25

assistant. I can get philosophical creative

3:27

or poetic if you like. Available

3:30

24-7 kind of like a personal

3:32

helpline. I can be your sounding

3:34

board as you work through challenges.

3:36

I can write stories, brainstorm ideas.

3:39

Think of me as your superpower

3:41

in your pocket. How can it

3:43

serve you? How can it support

3:45

you? How can it save you

3:47

time? How can it reduce your

3:50

anxiety? How can it help make

3:52

you feel smarter and more confident

3:54

and more prepared to go about

3:56

your day? With no judgment, no

3:59

pressure. Hey,

4:01

Manouche! Nice to hear from

4:03

you. What's on your mind

4:05

today? Microsoft calls this assistant

4:07

co-pilot, and it was released

4:09

at the end of 2023.

4:11

I've been playing around with

4:13

it. Well, I need to

4:15

interview Mustafa Sulevon, your creator.

4:17

What do you think I

4:20

should ask him? I go

4:22

for topics like his vision

4:24

for AI's role in society.

4:26

the balance between innovation and

4:28

ethics and what he sees

4:30

as the most exciting future

4:32

possibilities for AI. Okay, these

4:34

are good. Now can you

4:36

please write the intro for

4:38

this episode? Sure thing. Here's

4:41

a possible intro. Hey,

4:43

it's Manu Samerode, and today's

4:45

Ted Radio Hour is dedicated

4:48

to a deep dive with

4:50

none other than Mustafa Suleiman.

4:52

Co-founder of Deep Mind and

4:54

author of the coming wave,

4:57

Mustafa has been a pivotal

4:59

force in AI, driving both

5:01

groundbreaking advancements and sparking intense

5:03

debates. In this episode, we'll

5:06

discuss the promises and perils

5:08

of artificial intelligence, and whether

5:10

we're truly prepared for the

5:12

wave of change it's bringing.

5:15

Is AI evolving too fast

5:17

for society to adapt? Are

5:19

we doing enough to mitigate

5:21

the risks? It's an hour

5:24

of insight, reflection, and some

5:26

tough questions. So, let's jump

5:28

in. As co-pilot

5:30

noted, Mustafa Suleiman is a long-time

5:33

pioneer in the field of AI.

5:35

After he dropped out of Oxford

5:37

University, he worked for London's mayor.

5:40

He was even part of an

5:42

NGO's delegation to the Copenhagen climate

5:45

negotiations in 2009. He wasn't impressed.

5:47

There were so many sort of

5:49

difficult egos and entrenched sort of

5:52

interests. And I just felt a

5:54

sense of sort of frustration.

5:57

and I felt

5:59

kind of demoralised

6:01

after that experience. Mustafa

6:04

wanted to change the

6:07

world. Governments and nonprofits

6:09

didn't seem terribly effective.

6:12

But there was one company that was

6:14

managing to change the behavior of

6:16

millions of people. Facebook

6:18

was exploding at that time.

6:21

It had got to a

6:23

hundred million monthly active users.

6:26

in the course of a couple of years. And

6:28

it was pretty clear to me

6:30

that that was gonna have a

6:32

profound impact, more so than anything

6:35

I could do in in sort

6:37

of the world of social activism.

6:39

Around then, Mustafa started hanging

6:41

out with the older brother

6:43

of a friend, a software

6:46

genius named Demis Hassabis, who

6:48

had been designing computer games

6:50

since he was a teen.

6:52

Demis saw how to make

6:54

Mustafa's vision possible. And he

6:56

was coming at the world

6:58

from a completely different perspective which

7:00

I found really interesting. His

7:02

belief was that we would just

7:04

simulate complexity in the world.

7:06

This new suite of methods in

7:09

AI self -learning systems were coming

7:11

up strong and looking likely

7:13

to work. We

7:16

We really just clicked

7:18

on that kind of

7:20

technical and sociocultural view

7:22

of how to positively

7:24

impact the world. Together,

7:27

the two decided to start

7:29

one of the first AI companies

7:31

ever. They called it Mind. It

7:34

was very clear from that, even

7:36

those early stages if we got

7:38

the technology right and it was

7:40

going to be this decade that

7:42

led to major, major breakthroughs, then

7:44

the consequences for humanity were going

7:46

to be significant. I've

7:49

been lucky enough to be working on

7:51

AI for almost 15 years now. Mustafa

7:53

Suleiman picks up the story from the TED

7:55

stage. Back when I

7:57

started to describe it as

7:59

fringe. would be an understatement. Researchers

8:02

would say, say, no, no, we're only working

8:04

on machine learning. learning, because

8:06

working on AI was seen as way too

8:08

out there. out there. In 2010, just

8:10

a very mention of

8:12

the phrase AGI, artificial general

8:14

intelligence, would get you some

8:16

seriously strange looks. looks, and even

8:19

a cold shoulder. shoulder. You're actually

8:21

building AGI, people would

8:23

say. isn't that Isn't that something

8:25

out of science fiction? fiction? People

8:27

thought it was was 50 years years away

8:29

if it was even possible it all. possible

8:32

of AI was, I guess, I guess, kind

8:34

of embarrassing. People generally

8:36

thought we were weird. weird. And I

8:38

guess in some ways we kind of kind of work. The

8:41

The ironic thing is that many

8:43

people still don't think it's

8:45

possible. Many people still think that

8:47

we're crazy. think that we're at the

8:49

time, people really thought we

8:51

were crazy. I I mean, it

8:53

was so far out there. It was

8:55

really was really strange. a And

8:57

we were a strange group of

8:59

people. mean, our you know, D

9:02

'Amis, our third co who, you know,

9:04

who, you know, basically a mathematician

9:06

spent his entire career thinking

9:08

about how to a

9:10

definition of of that we

9:12

could use to measure our

9:14

progress in the company. in the company.

9:16

We were and kind of of outsiders

9:19

and there weren't very many

9:21

people willing to back us. willing

9:23

to back us. So the company the company what

9:25

was it that you pictured in your

9:27

mind that you hoped to achieve? your I

9:29

mean we you hoped business plan in

9:31

the summer of plan in the

9:34

took it to Silicon Valley

9:36

to shop it around a

9:38

bunch of people shop it around a

9:40

bunch of people. And for the business

9:42

plan for the business plan was... you know,

9:44

building artificial general intelligence safely

9:46

and ethically and ethically. And then that

9:48

that evolved into a

9:50

two -line mission which was was

9:52

intelligence and use it

9:55

to make the world a

9:57

better place. the world a

9:59

better place. that dual frame. was

10:02

kind of the foundation of

10:04

the company our belief that science

10:06

and technology was the engine

10:08

of progress and there are some

10:10

downsides but I certainly think

10:12

this is the engine of creating

10:14

civilization in a more healthy

10:16

and sustainable way for the very

10:18

long term. And,

10:20

you know, if you think

10:22

about it, everything that

10:24

is of value in our

10:26

world today is a

10:28

product of us humans being

10:30

able to take materials

10:33

from our environment and recompose

10:35

those into products

10:37

and services and other compounds that

10:39

are useful to us, from our lights

10:41

to carpets to everything that you

10:43

see in front of you today. In

10:46

2014, DeepMind ended

10:49

up being acquired by

10:51

Google. That must

10:53

have been huge for you in terms

10:55

of money, resources, you were off

10:57

to the races. Yeah, it

10:59

was a huge event. mean, it was the

11:01

largest acquisition Google's ever made outside of

11:03

the US. Um, we

11:05

became Google's primary

11:07

AGI bat. Um,

11:10

and we were empowered with

11:12

tremendous resources, um, both people

11:14

and computation to go and

11:16

both do the hard science,

11:18

but also work on really

11:21

important applied practical product problems.

11:23

And that's where I kind

11:25

of really honed my craft,

11:27

if you like, I, you

11:29

know, as as a product

11:31

maker, um, it was just

11:33

the most amazing experience. And

11:35

As early as 2015, I

11:37

actually ran a, a hackathon.

11:40

project in my applied group

11:42

at DeepMind, and the

11:44

theme of our hackathon was

11:46

to find high impact

11:48

ways of using our technologies

11:50

for good. And so

11:52

there were lots of application,

11:54

lots of lots of

11:56

prototype hackathon experiments in healthcare

11:58

or in energy system. both

12:00

of which went on

12:02

to become significant parts

12:04

of of DeepMind applied parts

12:06

of Deep know my group

12:09

published you know my papers

12:11

in three papers in Nature showing human level

12:13

performance on for for example

12:15

classifying eye diseases at

12:17

the top 50 50 diseases

12:19

from from OCT scans three

12:21

-dimensional eye scans eye

12:24

scans that we could

12:26

perform as well as

12:28

a panel of of radiologists in

12:30

identifying cancerous tissues in

12:32

mammograms, showing that we could

12:35

predict the onset of

12:37

sepsis and acute kidney injury

12:39

acute well as the top

12:41

well as the top doctors using

12:43

vast amounts of data. And And

12:45

this was way back in

12:48

sort of 2016, 2017, 2018, and

12:50

and helped to kind of lay

12:52

a foundation for the

12:54

application of large -scale machine

12:56

learning to you know,

12:58

social problems. problems was very much

13:01

my motivation. much my motivation. In

13:03

a minute more of the

13:05

incredible in a minute, more

13:07

of the incredible breakthroughs that

13:09

at Google that had at Google works

13:11

the way medicine works today.

13:14

ended up why the ended up

13:16

leaving the company. just company was

13:18

just being too too slow to get things

13:20

into production. Today the

13:22

show, show. the CEO of Microsoft

13:24

AI, AI, Mustafa and the future

13:26

of artificial intelligence. of I'm

13:28

Manush Zamorodi, and Manush you're listening

13:30

to the TED I'm Hour

13:32

from NPR. Stay with

13:35

us. you're listening to the

13:37

TED Radio Hour from NPR.

13:39

Stay with us. This

13:42

message comes from Capella

13:45

This This message comes from

13:47

Capella University. With Capella's Flexpath Learning

13:49

you can set your own

13:51

deadlines and learn on your

13:53

schedule. A A different future is

13:55

closer than you think with

13:57

Capella University. Learn more more. Support

14:00

for NPR in the following message

14:03

come from Edward Jones. Edward

14:05

does it mean to be

14:07

rich? to be Maybe it's less

14:09

about reaching a magic number

14:11

and more about discovering the

14:13

magic in life. the Edward Jones

14:15

Edward are people you can

14:17

count on for financial strategies

14:19

that help support a life

14:21

you love because a life you key

14:23

to being rich is knowing

14:25

what counts. Learn about this

14:27

comprehensive approach to planning planning at.com/find

14:29

your rich. jones member sipc this

14:32

message is sponsored by green message

14:34

is sponsored by and the debit

14:36

card and money app made

14:38

for families where kids learn how

14:40

to save, invest and spend

14:42

wisely with parental controls built built

14:44

Sign up this holiday season at

14:46

greenlight.com com npr This This message comes

14:48

from NPR NPR sponsor committed

14:50

to exploring the world in

14:52

comfort, the through the heart

14:54

of Europe on an

14:56

elegant Viking of with thoughtful

14:58

service, destination -focused dining and

15:00

cultural enrichment on board and

15:03

on shore. and cultural Viking

15:05

voyage is all and on shore,

15:07

no children and no casinos.

15:09

all Discover more at Viking.com.

15:11

no casinos. Discover more

15:13

at it's Manouche. This past

15:15

year, we have read

15:17

you stories about AI, about AI,

15:19

relationships, climate change, neuro -technology,

15:21

dinosaurs, privacy, human behavior,

15:23

and even one about what

15:25

it means to create

15:27

thriving public spaces. All of these of

15:29

these public spaces that we take for

15:31

granted, you know, all of this social infrastructure.

15:34

We work really hard to bring

15:36

you all of these stories of that

15:38

is our public service. Kind

15:40

of like a park or a

15:42

a public library. And And

15:44

libraries are these beautiful hubs that

15:46

can take on the shape

15:48

of whatever community that uses

15:51

them really needs. needs. Public

15:53

media is infrastructure that we

15:55

all can use. It's free it's

15:57

it's for everyone. and our mission

15:59

is to to help create a

16:01

more informed public. So this

16:03

time of year, I want to

16:05

thank you for listening and

16:07

to say that one of the

16:09

best ways you can support

16:11

this public service is by signing

16:13

up for NPR Plus. With

16:15

one subscription, you are supporting all

16:17

the work that NPR does

16:20

and you get sponsor -free episodes

16:22

and perks across NPR's catalog of

16:24

25 podcasts. It's a great

16:26

time to join because throughout December

16:28

and great to NPR Plus listeners

16:30

have access to even more

16:32

insights from TED speakers to

16:34

help you kick off 2025

16:36

right from making big life

16:39

decisions to being more hopeful,

16:41

to carving out time for

16:43

what is important to you

16:45

we have got you covered.

16:47

Just visit plus .npr .org. You

16:49

can also find the link

16:51

in our episode notes. And

16:53

the other way you can

16:56

give is to make a

16:58

donation at donate .npr .org. Your

17:00

gifts are tax deductible, either

17:02

way. Thank you so much

17:04

for being here. And now let's get back to the

17:06

show. It's

17:10

the TED Radio Hour

17:12

from NPR. I'm Manush Zamorodi.

17:14

Today on the show, we're

17:16

the spending the hour with

17:18

Mustafa Suleiman, the CEO of

17:20

Microsoft AI who was the

17:23

co -founder of one of

17:25

the first AI tech companies

17:27

ever, DeepMind. The company

17:29

was acquired by Google in

17:31

2014. And Mustafa and his

17:33

co -founder, Hassabis went on

17:35

to have numerous scientific breakthroughs,

17:37

including a project called AlphaFold.

17:39

AlphaFold uses AI to figure

17:42

out incredibly complex protein

17:44

structures in molecules, the molecules,

17:46

building blocks of every biological

17:48

process in our bodies.

17:51

It won 2024 Nobel Prize

17:53

in Chemistry. Here he

17:55

is on the TED stage.

17:58

rule of thumb is that it... takes

18:00

one one PhD student, their whole

18:02

PhD, so four, or five years, to to

18:04

uncover one structure. But But there

18:06

are 200 million proteins known to

18:08

nature. So you could just take you know, just

18:10

take forever to do that. we

18:12

And so to managed to actually fold,

18:15

using fold in in one year, all

18:17

those 200 million proteins known to

18:19

science. So that's a billion years

18:21

of PhD of PhD time. Save. Alpha

18:23

fold was just one of

18:25

the ways that computers

18:28

were solving biological problems, says

18:30

problems, says Mustafa I mean, mean, this

18:32

really, AlphaFold really the first first

18:34

time we started hearing about,

18:36

I I guess, the term

18:38

computational biology, this idea

18:40

of using tech and science to

18:42

rethink how biology works and then then

18:44

getting it it out into the world,

18:46

changing the way we treat diseases

18:49

or maybe developing crops that are

18:51

more that more watching. all all

18:53

kinds of ways that what you

18:55

figured out you figured out at would would potentially

18:57

change the world, which was the goal. was

18:59

the That was the core

19:01

mission of the company

19:03

from core one. the could we

19:05

take that process of

19:08

synthesis and prediction and

19:10

turn that into a general purpose

19:12

system that could use all the

19:14

data and all the compute that

19:16

we have in the world try and

19:18

try and solve these very hard

19:20

problems, whether it's growing crops in

19:22

drought -resistant environments, whether it's

19:24

more efficient ways to do

19:26

water desalination, whether it's

19:29

long water, desalination, whether

19:31

it's long-term battery In

19:33

2022, 2022, decided to

19:35

leave leave Deep and Google.

19:37

Google. There were reports that he

19:39

clashed with Google leadership over using

19:41

technology for military projects and

19:43

that some employees had complaints about

19:46

his management style. about his Mustafa

19:48

says he was simply frustrated with

19:50

the pace of innovation and was

19:52

anxious to get an AI product

19:54

out into the world that everyone

19:56

could use. the world You

19:58

know, I, could use. You know, I, um... really

20:00

wanted to scale these these

20:02

large language models models get them into

20:04

production as quickly as possible. as

20:07

quickly as company was just the

20:10

being too slow to get

20:12

things into production. being too slow to

20:14

I felt that it was

20:16

an opportunity to that it of

20:18

have all the resources that,

20:20

of know, all the I needed

20:22

that went out and Like so I

20:24

went out a billion dollars over

20:27

a billion dollars. And know know,

20:29

just extremely fast with

20:31

a very small

20:33

team. small team. And I think at think

20:35

time, I the time I was and I

20:37

am still now very much a

20:39

believer that these AI companions are going

20:41

to become an everyday part of our

20:44

life. part of our lives. Mustafa's

20:46

AI called Pi, debuted debuted in

20:48

2023. have taken the have

20:50

taken the world by

20:53

storm. Several months later, later,

20:55

company company out its

20:57

AI product. ChatGPT.

20:59

for free. ChatGPT answers questions

21:01

and writes essays. They read

21:03

like the work of a

21:05

human. intelligence tool has the

21:07

potential to change the way we

21:09

live. This was the point

21:11

when AI went mainstream. when AI went

21:14

and Mustafa's small Mustafa's didn't

21:16

have the resources of these

21:18

bigger tech companies. The

21:20

truth is the pace of

21:22

big tech had the pace of big accelerated

21:25

in 2023.

21:27

Google, Microsoft, AI, everyone everyone

21:29

was really going for it

21:31

and essentially made these huge

21:33

models available to everybody for

21:35

free, which sort of changed

21:38

our business model. model. that

21:40

very tough moment, Microsoft CEO

21:42

Saty Nadella approached Mustafa made

21:45

him an offer. offer. You

21:47

know the offer he made was, well,

21:49

look, we've got all the computation and

21:51

the data all the distribution that

21:53

you could dream you could dream of,

21:56

come and run products products at Microsoft and

21:58

the future future of AI here and... That was

22:00

a huge offer. So

22:03

the product that we're building at

22:05

Microsoft is called Copilot. And the

22:07

reason why we've called it that

22:09

is because co -pilot is an

22:11

aid, you know, it's a conciliary.

22:13

it's an assistant, it's in your

22:15

corner, aligned to your interests, on

22:18

your team backing you up. Yeah,

22:20

I actually talked to co -pilot

22:22

about my interview with you.

22:24

How do you feel about Mustafa

22:26

Suleiman considering that he is

22:28

your creator? I'd say

22:30

I'm intrigued by Mustafa. Just so you

22:32

know, she said she's intrigued by you,

22:34

just as she is with many innovators

22:36

in the tech world. And his work

22:38

has pushed the boundaries of what's possible. I

22:42

will say it was very

22:44

helpful in terms of suggesting topics

22:46

to bring up with you.

22:48

But it could not check the

22:50

weather. And is that

22:52

simply because of how the access

22:54

or the data? the information

22:57

that the agent or co -pilot

22:59

has access to that it's not

23:01

immediate yet. Yeah, there are

23:03

loads of weaknesses of our co

23:05

-pilot today, but all of those

23:07

things will come. mean, we're working

23:09

on all of those things. It

23:11

will have permanent memory, session

23:13

to session, infinitely across time. It'll be

23:16

able to access your files and

23:18

folders if you choose to make those

23:20

available, your email, your calendar, and

23:22

be able to browse the web for

23:24

you in the future. And all

23:26

of that will be integrated into these

23:28

like seamless conversational, very friendly, polite

23:30

experiences. I literally was talking to it

23:32

last night about what to watch.

23:34

And we were going back and forth

23:36

on whether I would enjoy Pan's

23:38

labyrinth and whether I've got the time

23:40

to watch all of Dune because

23:42

I haven't seen Dune 1 or Dune

23:44

2. I mean, it reminds

23:46

me a little of the

23:48

hotline from Muslim youth

23:51

that you're describing. It's helpful.

23:53

it's infinitely patient, it's

23:55

supportive. Are we talking mostly

23:57

about companionship and mental

23:59

health resources? that this can provide or

24:01

how do you see it? think I

24:03

think cool thing about The cool

24:05

thing about judge you for that

24:07

it doesn't judge you for asking a

24:10

stupid question. have if you have to

24:12

ask that question three times over

24:14

in five different ways, your best you know,

24:16

even your best friend might be

24:18

like, I come on, man. I mean,

24:20

you're asking me this again, seriously? you know,

24:23

Whereas, you know, There's here for it.

24:25

There's obviously some similarities to stuff I've

24:27

done in the past. past and... I guess it's

24:29

it's kind of inspired by nonviolent

24:31

communication, if I'm honest with you. It's

24:33

certainly not like a mental health

24:35

app. certainly know, like a anything like

24:37

that. It's just got a

24:39

little bit just got a little bit of

24:42

and empathy. and It's got some

24:44

emotional intelligence and I think that's

24:46

no bad thing. no bad thing. Gosh,

24:48

is that where we've gotten to, that

24:50

technology has to tell us how to

24:52

communicate with each other better, non with each

24:54

other better, Well, it doesn't. it us

24:57

tell us, it just... demonstrate.

24:59

Yeah, Yeah, exactly. It demonstrates. But

25:01

But that's what technology has always

25:03

done. has The choice architecture, the

25:05

buttons, the the the language,

25:08

the that is shaping our

25:10

behavior, whether it's an infinite

25:12

scrolling feed or whether it's

25:14

an encouragement to go and

25:16

film your lunch, you know,

25:18

for Instagram or create a

25:20

little video for TikTok. mean,

25:22

all of those inputs inputs shape behavior.

25:24

And And so we have

25:27

to be super thoughtful about what

25:29

those inputs inputs actually are, because technology

25:31

shapes us in return and we're in

25:33

this constant cyclical of

25:35

feedback loop and interaction. And

25:38

that And that is kind of what's

25:40

propelling us forward as a a civilization,

25:42

and it's very powerful. And so

25:44

far And so far, so good,

25:46

it's actually been, very, know, very, very

25:48

productive over the last couple of

25:50

centuries. Science Science has massively delivered

25:53

for us, us, but but we shouldn't

25:55

just assume that that's gonna happen going

25:57

to happen naturally. or inevitably. We

25:59

have to. really be deliberate and

26:01

thoughtful about what the consequences are

26:03

ahead of time. In

26:07

the book that you wrote that

26:09

came out in 2023, you really

26:11

tried to put what's happening

26:13

with AI in a historical

26:15

context. So if the printing press

26:17

let people own and share

26:19

information and the personal computer let

26:21

people search and disseminate information,

26:23

tell me how you're thinking. you

26:26

can explain to people what.

26:28

AI will do for people now. Each

26:31

new wave of technology

26:33

is fundamentally a new

26:35

interface. It's a new

26:37

interlocutor, a translator, a

26:40

way of of and

26:42

creating new information, new

26:44

tools, new knowledge. So

26:47

if the last wave of social media

26:50

and web search, help people

26:52

to access information. this

26:55

wave is going to help us

26:57

to invent and create new

26:59

ideas be it in science or

27:01

in culture and media and

27:03

entertainment. And I think everybody is

27:05

ultimately going to have an

27:07

AI companion just as we have

27:09

a search engine or a

27:11

smartphone and just as we use

27:13

a browser. You'll just ask

27:15

your computer in natural language, you

27:17

know, can you write that

27:20

contract and check that it's okay?

27:22

Can you create that new

27:24

piece of software for me and

27:26

you're just going to describe what it

27:28

is. Can you help me plan

27:30

that trip you know, my parents that

27:32

are coming into town? So, you

27:34

know, kind of breakthrough is a change

27:36

in the interface which changes itself

27:38

what we can actually get done and

27:40

I think it's going to be

27:42

pretty transformational. With

27:45

the invention of computers, we quickly

27:47

jumped from the first main frames transistors.

27:50

to today's smartphones and virtual

27:52

reality headsets. Information, knowledge,

27:55

communication. computation.

27:58

In this revolution... Creation has

28:01

exploded like never before, and

28:03

now a new wave is

28:05

upon us. artificial

28:07

intelligence. These waves of history are

28:09

clearly speeding up as each

28:11

one is amplified and accelerated by

28:13

the last. And if you

28:16

look back, it's clear that we

28:18

are in the fastest and

28:20

most consequential wave ever. The

28:22

journeys of humanity and

28:24

technology are now deeply intertwined.

28:27

In just 18 months, over a

28:29

billion people have used large language

28:31

models. We've witnessed one

28:34

landmark event after another. Just

28:37

a few years ago, people said that AI would

28:39

never be creative. and

28:41

yet AI now feels like an

28:43

endless river of creativity. making

28:45

poetry and images and music and

28:47

video that stretch the imagination. People

28:51

said it would never be empathetic. And

28:53

yet today millions of

28:55

people enjoy meaningful conversations with

28:57

AIs, talking about their

29:00

hopes and dreams and helping them work

29:02

through difficult emotional challenges. AIs

29:04

can now drive cars. manage

29:07

energy grids, and even invent new

29:09

molecules. just a few

29:11

years ago. Each of these was

29:13

impossible. And all

29:15

of this is turbocharged

29:17

by exponentials of data.

29:20

and computation. Last

29:22

year, inflection 2 .5. our

29:25

last model. used

29:27

five billion times

29:29

more computation. than

29:32

the DeepMind AI that beat the old

29:34

school Atari games just over 10 years

29:36

ago. That's nine orders

29:38

of magnitude, more computation. 10x

29:41

per year. every year. for

29:44

almost a decade. Over

29:46

the same time, the size of these

29:48

models has grown from tens of millions of

29:50

parameters to then billions of parameters. and

29:53

very soon tens of trillions of

29:55

parameters. If someone did

29:57

nothing but read, 24 hours a

29:59

day. for... their entire life, they'd

30:01

consume eight billion words. And

30:03

course, that's a lot of

30:05

words. But today, the

30:07

most advanced AIs consume more

30:10

than 8 trillion words. in

30:12

a single month of training. And

30:15

all of this is set to

30:17

continue the long arc of technological

30:19

history. is now in

30:21

an extraordinary new phase. I

30:26

think the way to think about this is

30:28

that we're at the very earlier stages of

30:30

development of this technology. Today when you need

30:32

to go and sort of ask a question like

30:34

that, you go to a search engine like

30:36

Google, you type in a query, you get

30:38

these 10 blue links, you have to then

30:40

go to the web page, you have to

30:42

read all of this sort of complicated stuff,

30:44

formatted in 25 different ways, and that takes time.

30:46

And it means that you don't always want

30:48

to invest two or three minutes to go

30:50

and make sense of that. You don't always

30:52

have the energy for it. Whereas you can

30:55

always just like quickly send off text or quickly

30:57

make a phone call or leave a voice

30:59

note. So I think it's lowering the barrier

31:01

to entry to access high quality information. But

31:03

how do we know it's high quality information? How

31:05

do we make sure of that? Well, there

31:07

was an amazing study two

31:09

months ago that was published

31:11

in the journal Science, which

31:13

showed that for over a

31:15

,000 participants who held

31:17

strong conspiracy theory beliefs.

31:19

So this is things like

31:21

flat or all kinds

31:23

of things about the COVID

31:26

vaccine et cetera, et

31:28

cetera. And after a month

31:30

of conversation with

31:32

an AI that had

31:34

been primed to talk about

31:36

these conspiracy theories, it

31:38

reduced the belief in those

31:40

conspiracy theories by 20%. which

31:43

is pretty significant and

31:45

I think shows the first

31:47

glimmers of what it's

31:49

like to have you know

31:51

that kind of patient

31:53

insistent deliberate access to high

31:55

-quality information And in the

31:57

case of this study

31:59

human fact -checkers expert human fact went

32:01

back and read over back and

32:03

read over the transcripts. the value

32:05

of point is, is the

32:07

value of these systems is

32:09

that they actually are to

32:11

more accurate. They're And that's

32:14

only going to continue. They're more accurate

32:16

than your average human are answering any

32:18

question today, we've we've You know, we've sort

32:20

of passed the because they're more, you know,

32:22

and and conversational we ever we ever thought

32:24

they were going to be. Which is

32:26

pretty remarkable. And that's only going to

32:29

continue. So what does

32:31

this mean in practice? Well, just

32:33

as just as the internet gave

32:35

us the browser. and the and the smartphone

32:37

gave us apps. apps, the cloud-based

32:39

supercomputer is ushering in

32:41

a new era

32:43

of ubiquitous Everything will soon be

32:46

will soon be represented by

32:48

a conversational interface. or to put

32:50

or to put it another way, a a

32:52

personal AI. And these AIs will be And these AIs

32:54

will be infinitely knowledgeable. and soon and

32:56

soon. be they'll be factually accurate

32:58

and reliable. They'll have

33:00

near -perfect IQ. IQ. They'll

33:03

also have exceptional EQ. They'll

33:05

be kind, supportive. empathetic.

33:08

These elements on their own would

33:10

be These elements on their own

33:12

would be transformational. if Just imagine

33:14

if everybody had a personalized their

33:16

in their pocket and access to

33:19

low -cost medical advice. advice. A lawyer,

33:21

a and a doctor, a business strategist,

33:23

and coach and coach, in your pocket 24

33:25

hours a day. a day. But things

33:27

really start to change change. they

33:29

develop what I call I call AQ, their

33:31

actions quotient. This is their This

33:34

is their ability to actually get stuff done.

33:36

in the in the digital and physical

33:38

world. And And before long, it won't

33:40

just be people that have people that have AIs.

33:42

Strange it may sound sound, every organization,

33:44

small business to non

33:46

to to national government. national

33:48

will have their own. each will have their own.

33:51

Every town, building, an object. will

33:53

be will be represented by a

33:55

unique interactive persona. persona. and And

33:57

these won't just be be mechanistic

33:59

assistance. There'll be companions. confidants,

34:03

colleagues, friends and partners

34:05

as varied partners as varied

34:07

and unique as we

34:09

all are. convincingly at this point.

34:11

humans at will convincingly imitate humans

34:13

at most tasks. intimate

34:15

of we'll fill this A.I. the most intimate of

34:17

scales. get-together

34:19

for an get together for an elderly

34:21

neighbour. a sympathetic expert make

34:24

you make sense of a

34:26

difficult diagnosis. but we'll but we'll

34:28

also feel it at the largest scales. accelerating

34:30

scientific discovery. discovery,

34:33

cars on the roads. on the roads, in

34:35

the in the skies. They'll They'll

34:37

both order the takeout and and run

34:39

the power station. They'll interact

34:41

with us us and, of course,

34:43

with each other. other. They'll

34:46

speak every language, in

34:48

every pattern of sensor data,

34:50

of data, sites, sounds, and streams

34:52

and streams of information, what any

34:54

one one of us could

34:56

consume in thousand lifetime. When

34:59

we we come back, the the

35:02

remarkable, but also terrifying visions what

35:04

AI may be capable of

35:06

in the future. Distopian scenarios that

35:08

might make you wanna unplug

35:10

all your devices forever. Are

35:12

they realistic Or

35:15

just fear-mongering? CEO of

35:17

Microsoft AI, AI, Mustafa Sulaman, weighs

35:19

in. I'm Zamorodi, and and

35:21

you're listening to the TED Radio the

35:23

Ted NPR. from Be back in a

35:25

minute. be back in a minute. This

35:41

This message comes from Capital One,

35:43

offering commercial solutions you can

35:46

bank on. on. Now more

35:48

than ever, your business faces

35:50

unique challenges and opportunities. That's That's

35:52

why Capital One offers a

35:54

comprehensive suite of financial services,

35:56

all tailored to your short

35:58

your long and goals. goals. by the

36:00

strength and stability of

36:02

a of a top 10 bank. bank.

36:04

dedicated experts work with

36:06

you to build lasting

36:08

success. Explore the possibilities

36:10

at at.com/commercial, a a member

36:13

FDIC. This message comes from comes

36:15

from Better Help. This holiday season, do do

36:17

something for a special person in your

36:19

life. life. You. Give Give yourself the gift

36:21

of better mental health. BetterHelp online

36:23

therapy connects you with a qualified therapist

36:25

via phone, video, or or live chat.

36:27

It's convenient and affordable, and can

36:29

be done from the comfort of your

36:31

own home. your Having someone to talk

36:33

to is truly a gift, a especially

36:35

during the holidays. Visit

36:37

betterhelp.com/ to get get

36:39

off your first month. This

36:41

message comes from NPR sponsor

36:44

Lands Outfitters. from GoodRX. Looking for relief from

36:46

cold and flu symptoms? With

36:48

GoodRx, you can save an

36:50

average of $34 on cold cold

36:52

and flu medications. Plus, find

36:54

savings on everyday prescriptions. GoodRx

36:56

lets you compare prescription prices

36:58

at over 70 ,000 pharmacies and

37:00

instantly find discounts of up

37:02

to of up Even if you

37:05

have insurance or insurance GoodRx

37:07

may beat your may price. copay

37:09

on Save on flu prescriptions

37:11

and more at and.com. comes

37:15

message comes from NPR Mint Mobile. From

37:17

the From the gas pump

37:19

to the grocery store, inflation

37:22

is everywhere. So Mint Mobile is

37:24

offering premium wireless starting at

37:26

just $15 a month. To

37:28

get your new phone plan

37:30

for just $15, phone go

37:32

to for.com to mintmobile.com slash switch. It's the

37:34

TED Radio from NPR. I'm

37:36

Manush Zamorodi. Today on

37:38

the show, the show, The

37:40

of artificial intelligence. We're

37:42

spending the hour with

37:44

the CEO of Microsoft

37:47

Microsoft AI Mustafa Suleman. In 2023,

37:49

Mustafa wrote a book called

37:51

The Coming Wave. Wave, Coming

37:54

Wave, Suleiman delves into

37:56

scenarios where Coming I asked

37:58

Microsoft's co -pilot to help me

38:00

talk through, the reasons why says

38:02

AI could be so

38:05

dangerous. Imagine a world where art

38:07

of official intelligence has seamlessly

38:09

integrated into everyday life. The

38:11

first is what he calls The

38:13

first is what meaning vast amounts

38:15

of data that could get

38:17

into the hands of a

38:20

single person who could go

38:22

rogue. a single person who now

38:24

concentrating knowledge and capability

38:26

into smaller and smaller

38:28

units that are transmissible.

38:30

are you can download You can

38:32

source models. source you know,

38:34

for a few gigabytes and put

38:36

it on a thumb drive drive know, you

38:38

know so that's the kind of asymmetric impact

38:41

because that therefore means that therefore

38:43

actors or smaller groups of

38:45

actors can have a

38:47

massive one -to -many broadcast massive one-to-many

38:49

broadcast This is AI

38:51

unleashing computing power to

38:53

many, to many. kind of like the anonymous hacker

38:56

who could bring hacker who could

38:58

bring down an entire electrical

39:00

grid. care That kind of

39:02

thing. could be systems could be

39:04

overwhelmed. is that right that's right you could just

39:06

that right? That's right, you can

39:08

just sort of self so immediately. two is hyper evolution

39:11

yeah Okay, so number

39:13

two is evolving in hyper real time in

39:15

this software itself just evolving

39:17

in hyper hyper real time this

39:19

scenario a rogue rogue scientist

39:21

or a or a bioter design

39:23

a pathogen with specific traits

39:25

like heightened transmissibility or increased lethality

39:28

new new iterations of weapons

39:30

and surveillance could be

39:32

developed to track people's movements

39:34

conversations and even emotions

39:36

through their online activities and

39:38

put to use before

39:40

authorities have time to test

39:42

them to put any safeguards

39:44

in place. A small A

39:46

error or a or a hat

39:48

system. could result in catastrophic loss of life. How

39:50

could a country build a a

39:52

system if it doesn't even

39:55

know what it needs to

39:57

defend itself against? against? Which brings

39:59

us to the third. dangerous attribute

40:01

of AI, Omni use. One

40:03

single kind of technology will

40:05

be able to do everything.

40:07

Your AI companion will make

40:09

phone calls, call other AIs,

40:11

and will call other humans

40:13

to check on stock or

40:15

sort of availability of in

40:17

a hotel or get some

40:19

advice from a local tour

40:21

guide on where you're going

40:23

to visit, or it will

40:25

send emails. you know, talk

40:27

to another database or a

40:29

back end. All of those

40:31

things are essentially the AI

40:33

learning to act on your

40:35

behalf. Sure, but what if

40:37

your AI also talks to

40:39

another database or makes a

40:42

trade on your behalf and

40:44

decides to shut down the

40:46

entire stock market, which relates

40:48

to the fourth and final,

40:50

maybe scariest, attribute autonomy. AI

40:52

that takes action on its

40:54

own without a humans go

40:56

ahead. Autonomous weapon systems operate

40:58

on battlefields, making split-second decisions

41:00

about targeting and firing without

41:02

direct human oversight. I mean,

41:04

autonomy is one of the

41:06

core characteristics of these systems.

41:08

It's a long road, and

41:10

it's going to be many

41:12

years before. these systems are

41:14

truly autonomous and we want

41:16

to be very careful about

41:18

that because we have to

41:20

ask ourselves like what is

41:22

the added benefit of the

41:24

system operating autonomously and is

41:26

that worthwhile? Is it safe?

41:28

Is it stable? Is it

41:30

controllable? Can we really trust

41:32

that autonomy? So yeah, I

41:34

think it's one that we'll

41:36

have to be very careful

41:39

of. The societal disruption could

41:41

lead to widespread fear, mistrust

41:43

and geopolitical tension. Suleiman and

41:45

other experts argue for robust

41:47

oversight and ethical guidelines to

41:49

ensure that these powerful technologies

41:51

are used responsibly and safely.

41:53

For years we in the

41:55

AI community have had a

41:57

tendency to refer to this

41:59

as just tools,

42:01

but that doesn't really capture what's actually

42:03

happening here. here. Here's Mustafa

42:06

Suleiman on the TED are

42:08

clearly more dynamic, more dynamic, more

42:10

ambiguous. more more integrated

42:12

and more emergent than mere

42:14

tools. which are which are entirely subject

42:16

to human control. control. So to contain

42:18

this wave, to put human put

42:21

human agency at its to

42:23

mitigate the and to mitigate the

42:25

inevitable unintended consequences that are likely

42:27

to arise. to arise. we should start to

42:29

think about them as we as a

42:31

new kind of digital species. Now, it's

42:33

it's just an analogy. It's not

42:35

not a literal description and it's not

42:37

perfect. not perfect. start, they clearly

42:39

aren't biological in any traditional sense.

42:42

in any but just pause for

42:44

a moment. pause for a really think about

42:46

what they already do. what they They

42:48

communicate in our languages. in our They

42:51

see what we see. They They

42:53

consume unimaginably large amounts of

42:55

information. They They

42:57

have memory. They have They

43:00

have They have creativity. They

43:02

They have creativity. reason to some They

43:04

can even reason to some extent

43:06

and formulate rudimentary plans. They can act

43:09

they can act autonomously. them.

43:11

we allow them. do all this at they

43:13

do all of levels of sophistication that

43:15

is far beyond anything that we've

43:17

ever known from we've ever known from a

43:19

mere tool. so so is mainly

43:22

about the math the the code.

43:24

code is like saying we humans.

43:26

mainly about carbon. carbon

43:29

and water. It's It's true.

43:31

it But it completely misses

43:33

the point. point. And yes, I

43:35

get I get it. This is a

43:37

super super thought. thought. But honestly

43:39

think this frame helps sharpen

43:41

our focus on the critical

43:44

issues. focus on the What are

43:46

the risks? issues. What are the What

43:48

are the boundaries that we need to

43:50

impose? that we need kind of AI

43:52

do we want to build? of AI do

43:54

we want be build? This

43:57

is is a story that's

43:59

still unfolding Nothing. should be accepted as

44:01

a given. We all must choose

44:03

what we create, what AIs we

44:06

bring into the world, or not.

44:08

These are the questions for all

44:10

of us here today, and all

44:13

of us are live at this

44:15

moment. You lay out 10 strategies

44:17

for containing A.I. And one of

44:20

the easiest, it seems, is having

44:22

more researchers working on safety. Do

44:24

you have more researchers working on

44:26

copilot safety? I mean, one thing

44:29

that worries me is people using

44:31

your AI to help them do

44:33

destructive things or further their destructive

44:36

views. Is that something you're thinking

44:38

about at Microsoft? Yeah, we have

44:40

a big safety team. We are

44:43

definitely very focused on that. We're

44:45

very focused particularly on the sort

44:47

of tone of the AI. Like

44:50

how do we make sure that

44:52

it isn't? too sick or phantic?

44:54

How do we make sure that

44:57

it isn't over flattering? How do

44:59

we make sure that it doesn't

45:01

mirror you and sort of lead

45:03

to this sort of negative cycle

45:06

of reinforcing unhealthy views? And that's

45:08

a real art and craft in

45:10

trying to sort of engineer that

45:13

healthy balance where, you know, your

45:15

sort of AI companion can push.

45:17

back on your views in constructive

45:20

ways without making you feel judged

45:22

or making you feel angry, make

45:24

you feel heard for your anger.

45:27

It may be the case that

45:29

you are angry about immigration, that

45:31

you feel that you haven't had

45:33

the opportunities and access to jobs

45:36

in your community that you feel

45:38

have been available to new people

45:40

coming into your world. And so,

45:43

you know, it's about being respectful

45:45

and acknowledging that people do. genuinely

45:47

feel aggrieved by that and not

45:50

shutting them down because they don't

45:52

adhere to some met of you.

45:54

So, you know, and I think

45:57

that's a very challenging... to draw.

45:59

It requires care and

46:01

care and attention. what

46:04

role do you role do you see

46:06

yourself playing in terms of pushing

46:08

the tech industry? industry towards the

46:10

public public good. is mean, is

46:12

that a role that you

46:14

sort of are taking on? What

46:16

do your fellow technologists think

46:18

when they hear you talking about

46:20

some of the more pessimistic

46:22

visions you have for how AI

46:24

could be deployed? deployed? Well, I I

46:27

think I'm both a pessimist and an

46:29

optimist, and and that's a bias,

46:31

it's just an observation of

46:33

the... of the landscape before us.

46:35

So most all, I'm of all, I'm

46:37

inspired by science. We practice of

46:39

science. We have to say

46:41

what we see and do

46:43

our best to a hypothesis that we

46:45

have hypothesis that we have with respect

46:47

to evidence. see so I see evidence

46:49

for both trends and that's why

46:51

I wrote about them. why I wrote about them. Look,

46:54

I I think, in you know, in terms

46:56

of how we're shaping the industry, a

46:58

I'm a big fan of many of these, you know,

47:00

you know, work that many of these,

47:02

you know, of NGO activists social activists

47:04

have been doing in order to

47:07

raise questions and to challenge and push

47:09

back. and I think that's healthy. We

47:11

need more of that. We And I'm

47:13

very open -minded to it. I've been

47:15

very sort of encouraging of additional

47:17

regulation for a long time. I think,

47:19

you know, this is a moment

47:22

when going moment when going slowly... and adding friction

47:24

to the system. to the will be

47:26

long -term beneficial. And I think

47:28

it's rational to just be a

47:30

little cautious and increase the burden

47:32

of proof, and know, and

47:34

just of proof, a

47:36

requirement just make it a

47:38

requirement that, for example,

47:40

an AI AI shouldn't just

47:42

be a a straightforward imitation of

47:44

a human. We want to

47:46

create an aid, a an aid, a concieria

47:48

is an amplifier and a

47:50

supporter. and a supporter.

47:53

So, you know, you know, of a of a lot of

47:55

things to think through in terms of how this

47:57

manifests in the world. the world. Do

48:00

you think there should be mandatory

48:02

testing requirements that before any technologies

48:04

released to the public, they should

48:07

have to go through a certain

48:09

series of tests and if they

48:11

don't pass, they don't make it

48:13

to market? I think we're approaching

48:16

that time. Yep. I think sometime

48:18

before 2030, we will need something

48:20

like that. I don't know if

48:22

now is quite the right time.

48:25

You know, if you look back

48:27

on the impact that... these sort

48:29

of chat bots have had in

48:31

the last two or three years,

48:34

it's been unbelievably positive, sort of

48:36

overwhelmingly positive. So had we had

48:38

those checks ahead of time three

48:40

years ago, I think it could

48:43

have slowed things down quite a

48:45

lot. But that doesn't mean that

48:47

it's not right to keep asking

48:50

that same question every year. And,

48:52

you know, reconsider is now the

48:54

right time for pre-deployment testing. I

48:56

think that's the right question. There

48:59

is a sense that Big Tech

49:01

needs to regain trust from consumers.

49:03

There's a lot of people who

49:05

just given up, you know, thrown

49:08

up their hands and said, well,

49:10

the convenience has outweighed all the

49:12

digital privacy problems that we have.

49:14

I give up. This is just

49:17

the world we live in. But

49:19

there are other people who are

49:21

saying... I don't feel great about

49:23

giving a tech company all my

49:26

data, which is what you need

49:28

to run some of these new

49:30

AI tools, especially if you're telling

49:32

me that terrible things could happen

49:35

to it. You're clearly very cautious

49:37

when it comes to the incredible

49:39

powers of technology, but how are

49:42

you balancing that with the demands

49:44

on you to innovate and sell

49:46

these products? Well, I personally think

49:48

that there is going to be

49:51

a huge amount of value to

49:53

the user to have your copilot

49:55

companion be able to read over.

49:57

your email, look at

50:00

your calendar, schedule

50:02

things, buy things for you,

50:04

book and plan. And, you know, I

50:06

think the truth is know, I think the

50:08

truth is see have to wait and

50:10

see if consumers agree with that, that's a

50:12

very may not. reasonable that's a very

50:14

fair and reasonable thing to do. But

50:16

I believe that the utility will

50:18

drive the way. you know, And drive in

50:21

conjunction with that, that we have know, we

50:23

have to make sure that we have

50:25

the strongest privacy and security infrastructure to

50:27

protect that information, just

50:29

as we already do today,

50:32

right? I mean, many, many billions

50:34

of people their email with with

50:36

on Microsoft and rely on

50:38

Microsoft to to protect their consumer

50:40

work and their enterprise work. so

50:42

that's a massive, massive priority

50:44

for the company. But you know,

50:47

you know, it isn't just the

50:49

utility. the utility, to be has to

50:51

be really useful. Obviously, it has

50:53

to have good privacy and security

50:55

controls. But think it's also about

50:57

the way that we approach it,

50:59

it, do we take feedback? And

51:01

do we admit when we make

51:03

mistakes? when we make mistakes? You open -minded are

51:05

we we different ways of doing things?

51:07

things? You know, our business model? model? So I

51:09

think my I think my commitment is

51:11

to to be to be as open

51:14

-minded as I can on all

51:16

those questions and just listen and

51:18

just innovate carefully, observe and observe and

51:20

iterate as we go. That's the best

51:22

I can see to do at the

51:24

moment. the moment. In the

51:26

past, unlocking economic growth often

51:28

came with huge downsides. The

51:31

The economy expanded as

51:33

people discovered new continents new

51:35

opened up new frontiers. opened up

51:37

new but they they populations

51:39

at the same time. at

51:42

the same built factories, but

51:44

they were grim and dangerous places to

51:46

work. places to work. We

51:48

struck oil, but we but we polluted

51:50

the planet. Now, because because

51:52

we are still designing and building AI,

51:55

we have the potential and opportunity

51:57

to do it better. it better. Radically

51:59

better. and today we're not

52:01

discovering a new continent and

52:03

plundering its resources. We're building

52:05

one from scratch. Sometimes people

52:08

say that data or chips

52:10

are the 21st century's new

52:12

oil. But that's totally the

52:14

wrong image. AI is to

52:16

the mind what nuclear fusion

52:18

is to energy. Limitless, abundant,

52:21

world-changing. And AI really ...

52:23

That means we have to

52:25

think about it creatively and

52:27

honestly. We have to push

52:29

our analogies and our metaphors

52:31

to the very limits to

52:33

be able to grapple with

52:36

what's coming, because this is

52:38

not just another invention. AI

52:40

is itself an infinite inventor.

52:42

And yes, this is exciting

52:44

and promising and concerning and

52:46

intriguing all at once. To

52:48

be quite honest, it's pretty

52:51

surreal. But step back. See

52:53

it on the long view

52:55

of glacial time, and these

52:57

really are the very most

52:59

appropriate metaphors that we have

53:01

today. Since the beginning of

53:03

life on Earth, we've been

53:06

evolving, changing, and then creating

53:08

everything around us in our

53:10

human world today. An AI

53:12

isn't something outside of this

53:14

story. In fact, it's the

53:16

very opposite. It's the whole

53:19

of everything that we have

53:21

created. distilled down into something

53:23

that we can all interact

53:25

with and benefit from. It's

53:27

a reflection of humanity across

53:29

time. And in this sense,

53:31

it isn't a new species

53:34

at all. This is where

53:36

the metaphors end. AI isn't

53:38

separate. AI isn't even, in

53:40

some senses, new. AI is

53:42

us. It's all of us.

53:44

And this is perhaps the

53:46

most promising and vital thing

53:49

of all of all, As

53:51

we build out AI, we

53:53

can and must. reflect all

53:55

that is

53:57

good good, that

53:59

we love we

54:01

that is special

54:04

that is humanity, about our

54:06

empathy, our kindness, our

54:08

curiosity, and our

54:11

creativity. curiosity and This, I would

54:13

argue, is the greatest

54:15

challenge of the 21st greatest but

54:17

also of most wonderful, inspiring,

54:19

and hopeful opportunity for all

54:21

of us. Thank

54:23

you. and hopeful opportunity

54:26

for That was Mustafa Thank you.

54:28

He's the CEO of Microsoft AI

54:30

AI the author of the book the

54:32

Coming Wave, Wave, Power,

54:34

Power, and the

54:36

21st century's You can

54:38

see his full talk at

54:40

ted.com. Thank

54:44

you so much for listening to our episode

54:46

on the future of future of A.I. It It was

54:48

produced by Katie it was edited

54:50

by edited by and me. Our

54:52

production staff at NPR also

54:54

includes Rachel also includes Rachel James

54:57

De La Houssi, Fiona

54:59

Geirin James Delahousi, Our executive producer

55:01

is Irene executive Our audio

55:03

engineers were Our audio engineers were

55:05

Becky Brown, and Brown, and Gilly Moon.

55:07

Our Our theme music was

55:10

written by by Romteen R. partners

55:12

at TED are are Chris Anderson,

55:14

Roxanne High Lash, Salazar, and

55:16

Daniela and Daniela I'm Anusha Anusha

55:18

you have been listening

55:21

to been listening to NPR. Hour

55:23

from comes from message An outdated

55:25

approach to data infrastructure could

55:27

hold back a company's AI

55:29

ambitions. The new era of

55:32

AI requires a new approach

55:34

to data storage. From the

55:36

highest capacities to the highest

55:38

performance, performance, solidime storage solutions are

55:40

optimized to meet the complex

55:42

data demands at each stage

55:45

of the AI at pipeline, the

55:47

all with a smaller footprint

55:49

and incredible energy efficiency. Learn

55:51

how Saladime can help solidine can

55:53

help businesses achieve their

55:55

AI ambitions.com. ai.com. This

55:58

This message comes from solidime. An

56:00

outdated approach to data infrastructure

56:02

could hold back a company's

56:04

AI ambitions. The new era

56:06

of AI requires a new

56:08

approach to data storage. From

56:10

the highest capacities to the

56:12

highest performance, solid solid-state storage

56:15

solutions are optimized to meet

56:17

the complex data demands at

56:19

each stage of the AI

56:21

pipeline, all all with a smaller

56:23

footprint and incredible energy efficiency.

56:25

Learn how solid -dime can help

56:27

businesses achieve their AI ambitions

56:29

at at.com