0:01

Ted Audio Collective. Talking

0:14

about AI feels like falling down a rabbit

0:17

hole sometimes. It's easy to

0:19

dig deep into ideas like neural networks

0:21

and machine learning, or complex concepts like

0:24

algorithms and big data. But

0:26

in the process, we lose sight

0:28

of one of the most important parts of

0:30

the conversation. How AI

0:33

tangibly affects our everyday lives.

0:36

If we were to explain AI to a

0:38

six-year-old, or someone like

0:40

my 92-year-old grandfather, we

0:43

might just point to everyday tools

0:45

we use. For instance, Siri

0:47

and Alexa on our smartphones and devices.

0:50

That's AI. Or

0:52

the way our favorite music streaming service knows

0:54

exactly which songs to recommend based

0:57

on our listening history. Well, that's

0:59

also AI. It's

1:01

true that AI is a tool that

1:03

we deliberately and intentionally use, and

1:06

that this technology is constantly developing.

1:09

But AI's everyday integration into our

1:11

lives is what makes it a

1:13

real revelation, both in our

1:15

present and our future. I'm

1:18

Shirelle Dorsey, and this is TED Tech.

1:24

Microsoft AI CEO Mustafa

1:26

Suleiman is one

1:28

of the primary architects of the AI models many

1:31

of us use today. He

1:33

takes to the TED stage to share his

1:35

understanding of this new digital entity and

1:38

how it will play a significant role in the rest

1:40

of our lives. This

1:48

episode is brought to you by Progressive Insurance.

1:50

What if comparing car insurance rates was as

1:53

easy as putting on your favorite podcast? With

1:55

Progressive, it is. Just visit the Progressive

1:57

website to quote with all the coverages

1:59

you need. you want. You'll see Progressive's

2:02

direct rate, then their tool will provide

2:04

options from other companies so you can

2:06

compare. All you need to do is

2:08

choose the rate and coverage you like.

2:10

Quote today at progressive.com to join the

2:12

over 28 million drivers who trust Progressive.

2:14

Progressive Casualty Insurance Company and affiliates, comparison

2:16

rates not available in all states or

2:18

situations. Prices vary based on how you

2:20

buy. Support

2:23

for TED Tech comes from Odoo. What

2:25

is Odoo? Well, Odoo is a lot of

2:27

things. Odoo is award-winning management

2:30

software. Odoo is total control

2:32

of your entire company in one place.

2:34

Odoo is a suite of fully

2:37

integrated applications for CRM, accounting, sales,

2:39

HR, inventory, manufacturing, and everything in

2:41

between. Basically, Odoo is what your

2:44

business needs to succeed. So

2:47

if you're ready to

2:49

get more done in

2:51

less time, visit odoo.com/TEDtech.

2:54

That's odoo.com/TEDtech. Odoo.

2:56

Business management made simple. Sometimes,

3:01

things in the world of technology

3:03

are complicated and need careful explaining.

3:07

Sometimes, they just need a little hard truth.

3:10

I don't think anyone is going to buy

3:12

a banana with crypto at any point in

3:14

the foreseeable future. I'm Lizzie

3:16

O'Leary, the host of Slate's What Next?

3:18

TBD, your clear-eyed guide to technology, power,

3:21

and the future, Friday and Sunday, wherever

3:23

you get your podcasts. I

3:30

want to tell you what I see coming. I've

3:33

been lucky enough to be working on AI for

3:35

almost 15 years now. Back

3:37

when I started, to describe it

3:40

as fringe would be an understatement.

3:43

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

3:45

on machine learning, because working on AI was

3:47

seen as way too out there. In

3:49

2010, just the very

3:51

mention of the phrase AGI,

3:54

artificial general intelligence, would

3:56

get you some seriously strange looks, and

3:59

even a cold shoulder. You're

4:01

actually building AGI, people would

4:03

say. Isn't that something out of

4:05

science fiction? People thought it was 50

4:07

years away or 100 years away if it was

4:09

even possible at all. Talk of

4:11

AI was, I guess, kind of

4:13

embarrassing. People generally thought we

4:16

were weird. And I guess in some ways we kind

4:18

of were. It wasn't long though

4:20

before AI started beating humans at a whole

4:22

range of tasks that people previously thought were

4:25

way out of reach. Understanding

4:27

images, translating languages,

4:29

transcribing speech, playing

4:31

Go and chess, and even

4:34

diagnosing diseases. People started waking

4:36

up to the fact that AI was going

4:38

to have an enormous impact. And they

4:40

were rightly asking technologists like me some

4:43

pretty tough questions. Is it

4:45

true that AI is going to solve the

4:47

climate crisis? Will it make personalized education available

4:49

to everyone? Does it mean we'll all

4:51

get universal basic income and we won't have to work anymore?

4:54

Should I be afraid? What does it

4:56

mean for weapons and war? And of

4:58

course, will China win? Are we in a race? Are

5:01

we headed for a mass misinformation

5:04

apocalypse? All good questions.

5:07

But it was actually a simpler and much

5:09

more kind of fundamental question that left me

5:11

puzzled. One that actually gets

5:13

to the very heart of my work every

5:15

day. One morning over

5:18

breakfast, my six-year-old nephew,

5:20

Caspian, was playing with Pi, the

5:22

AI I created at my last company, Inflection.

5:26

With a mouthful of scrambled eggs, he

5:28

looked at me plain in the face and said, but

5:31

Mustafa, what is an AI

5:33

anyway? He's such

5:35

a sincere and curious and optimistic little guy.

5:37

He'd been talking to Pi about how cool

5:39

it would be if one day in the

5:42

future he could visit dinosaurs

5:44

at the zoo and how he could make

5:46

infinite amounts of chocolate at home. And

5:49

why Pi couldn't yet play I Spy? Well,

5:52

I said it's a clever piece of software that's read

5:55

most of the text on the open Internet and it

5:57

can talk to you about anything you want. Right!

6:01

So like a person, then. I

6:04

was stumped, genuinely left scratching

6:06

my head. All my

6:08

boring stock answers came rushing through

6:10

my mind. No, but

6:13

AI is just another general-purpose technology, like

6:15

printing or steam. It'll

6:17

be a tool that will augment us and

6:19

make us smarter and more productive. And

6:22

when it gets better over time, it'll

6:24

be like an all-knowing oracle that will

6:26

help us solve grand scientific challenges. You

6:29

know, all of these responses started to feel,

6:31

I guess, a little bit

6:34

defensive, and actually better suited

6:36

to a policy seminar than breakfast with a

6:38

no-nonsense six-year-old. Why am

6:40

I hesitating, I thought to myself? You

6:43

know, let's be honest. My

6:45

nephew was asking me a simple question

6:48

that those of us in AI just

6:50

don't confront often enough. What

6:53

is it that we are actually creating?

6:57

What does it mean to make something

6:59

totally new, fundamentally

7:01

different to any invention that we

7:03

have known before? It

7:05

is clear that we are at an inflection

7:07

point in the history of humanity. On

7:10

our current trajectory, we're headed

7:12

towards the emergence of something that we

7:15

are all struggling to describe. And

7:18

yet, we cannot control what

7:20

we don't understand. And

7:23

so the metaphors, the mental models,

7:25

the names, these all

7:27

matter if we are to get the most

7:29

out of AI whilst limiting its potential downsides.

7:32

As someone who embraces the possibilities of

7:34

this technology, but who has also

7:37

always cared deeply about its ethics, we

7:40

should, I think, be able to easily describe

7:42

what it is we are building, and

7:44

that includes the six-year-olds. So

7:46

it's in that spirit that I offer up

7:48

today the following metaphor for

7:51

helping us to try to grapple with what this

7:53

moment really is. I think

7:55

AI should best be understood as

7:57

something like a new digital...

8:00

digital species. Now

8:02

don't take this too literally, but

8:04

I predict that we'll come to see them as

8:07

digital companions, new partners

8:09

in the journeys of all our lives. Whether

8:11

you think we're on a 10, 20, or

8:14

30 year path here, this

8:16

is in my view the most accurate and

8:18

most fundamentally honest way

8:21

of describing what's actually coming. And

8:24

above all, it enables everybody to

8:26

prepare for and shape what

8:29

comes next. Now I totally

8:31

get this as a strong claim and I'm going to

8:33

explain to everyone as best I can why I'm making

8:35

it. But first, let me just try

8:37

to set the context. From

8:39

the very first microscopic organisms, life

8:42

on earth stretches back billions of

8:44

years. Over that time, life

8:47

evolved and diversified. Then

8:50

a few million years ago, something

8:52

began to shift. After

8:54

countless cycles of growth and adaptation,

8:58

one of life's branches began

9:00

using tools. And that branch grew

9:03

into us. We went on

9:05

to produce a mesmerizing variety of

9:07

tools. At first

9:09

slowly, and then with astonishing speed,

9:12

we went from stone axes and fire

9:15

to language, writing, and

9:17

eventually industrial technologies. One

9:20

invention unleashed a thousand more and

9:24

in time we became homo

9:26

technologicus. Around 80

9:28

years ago, another new branch of technology

9:30

began. With the invention of computers,

9:32

we quickly jumped from the first mainframes

9:35

and transistors to today's

9:37

smartphones and virtual reality headsets.

9:40

Information, knowledge, communication,

9:43

computation. In

9:45

this revolution, creation has

9:47

exploded like never before and

9:50

now a new wave is upon us, artificial

9:53

intelligence. These waves

9:55

of history are clearly speeding up

9:57

as each one is amplified and

9:59

accelerated. by the last. And

10:02

if you look back, it's clear that

10:04

we are in the fastest and most

10:06

consequential wave ever. The journeys

10:08

of humanity and technology are

10:10

now deeply intertwined. In

10:12

just 18 months, over a

10:14

billion people have used large language models. We've

10:18

witnessed one landmark event after

10:20

another. Just a few

10:22

years ago, people said that AI would never be

10:24

creative, and yet AI now

10:26

feels like an endless river of creativity,

10:29

making poetry and images and music

10:31

and video that stretch the imagination.

10:34

People said it would never be empathetic, and

10:37

yet today, millions of people

10:39

enjoy meaningful conversations with AIs,

10:42

talking about their hopes and dreams and

10:44

helping them work through difficult emotional challenges.

10:47

AIs can now drive cars, manage

10:50

energy grids, and even invent new

10:52

molecules. Just a few years ago, each

10:55

of these was impossible. And

10:58

all of this is turbocharged

11:00

by spiraling exponentials of data

11:03

and computation. Last

11:05

year, Inflection 2.5, our last model, used

11:09

five billion times more

11:11

computation than the DeepMind

11:14

AI that beat the old school Atari

11:16

games just over 10 years ago. That's

11:19

nine orders of magnitude more

11:21

computation, 10x per year, every

11:24

year, for almost a decade. Over

11:27

the same time, the size of these models

11:29

has grown, from first tens of millions of

11:32

parameters, to then billions of parameters, and

11:34

very soon, tens of trillions of

11:36

parameters. If someone did nothing but

11:38

read 24 hours a day for

11:41

their entire life, they'd consume eight

11:43

billion words. And of course, that's

11:45

a lot of words. But

11:48

today, the most advanced AIs

11:50

consume more than eight trillion

11:52

words in a single month

11:54

of training. And all

11:56

of this is set to continue the long

11:59

arc of technology. I

24:22

have always been running with Brooks, but now I

24:24

run with the Brooks Ghost 16. It

24:27

has soft cushioning, smooth transitions,

24:29

and there's an engineered air

24:31

mesh upper. It provides

24:33

the right amount of stretch and structure to make

24:35

sure that the shoe accommodates the movement of my

24:38

feet. The Ghost 16 is out now. I encourage

24:40

you to try them out. My

24:42

runs feel great. I love

24:45

these shoes. They're really breathable. You

24:48

can visit brooksrunning.com to

24:50

learn more.