0:00

Science Friday is supported by

0:02

Progressive. Now, most of you

0:04

aren't just listening right now. You're

0:06

driving, cleaning, even exercising. But what

0:09

if you could be saving money

0:11

by switching to Progressive? Drivers who

0:13

save by switching save nearly $750

0:15

on average, and

0:19

auto customers qualify for an average

0:21

of seven discounts. Multitask

0:24

right now. Quote today at

0:26

progressive.com, Progressive Casualty Insurance Company

0:28

and affiliates. National average 12

0:30

month savings of $744 by

0:34

new customers surveyed, who saved with Progressive

0:37

between June 2022 and May 2023. Potential

0:42

savings will vary, discounts not

0:44

available, in all states and

0:46

situations. Science Friday is supported

0:48

by Dana-Farber Cancer Institute. Thanks

0:51

to Dana-Farber's foundational work, protein

0:54

degradation can target and

0:56

destroy cancer-causing proteins right

0:58

inside the cell. That's

1:01

how Dana-Farber is working

1:03

to treat previously untreatable

1:06

cancers. Learn more at

1:08

danafarber.org/everywhere. NYC

1:10

Studios is supported by MathWorks,

1:13

creators of MATLAB and Simulink,

1:15

software for technical computing and

1:17

model-based design. MathWorks, accelerating

1:20

the pace of discovery in engineering and

1:22

science. Learn more at mathworks.com.

1:28

Listener supported, WNYC

1:31

Studios. What

1:37

does one of the leading experts on

1:39

artificial intelligence see on the horizon? The reason

1:41

that we don't get smarter is we have

1:44

a limited size for our brain, but if

1:46

we can actually go outside of our brain

1:48

and into the cloud, the cloud right now

1:50

is expanding exponentially. We'll be able to

1:52

think much deeper thoughts. It's Wednesday,

1:54

June 19th, and just

1:57

like every day, today is Science

1:59

Friday. I'm

2:03

Sci-Fi producer Kathleen Davis. Ray

2:05

Kurzweil has been in the field of

2:07

artificial intelligence for more than 60 years,

2:11

longer than probably anybody else.

2:14

One of his biggest accomplishments is

2:16

coining the term, the singularity, the

2:18

idea that man and machine will

2:20

merge as the next stage of

2:23

evolution. Ray's new book,

2:25

The Singularity is Nearer when we merge

2:27

with AI, is out this month. Here

2:30

he is talking to Ira Flato. Joining

2:32

me now is futurist and inventor Ray

2:35

Kurzweil, author of the new book, The

2:37

Singularity is Nearer when we merge with

2:39

AI. Ray has worked with artificial intelligence

2:41

in that field for more than 60

2:43

years. I'm going to guess longer than

2:45

probably any other person alive. Welcome back

2:48

to Science Friday, Ray. Yeah, great to

2:50

be here again. Nice to have you.

2:52

When you look at the last 20

2:54

years, is it hard for you not

2:56

to just yell, I told you so?

2:58

I mean, has it gone the way

3:01

you thought it would? Well,

3:03

other people are saying that. In 1999, I

3:05

predicted by 2029, we'd reached the Turing test

3:10

and also AGI, where

3:12

basically a computer can do anything a human can

3:15

do. People met that

3:17

with a great deal of skepticism. Stanford

3:19

had a conference. Apparently

3:21

about a thousand people came from around the

3:23

world. And everybody agreed

3:25

that this would happen, but

3:28

they didn't think it would happen in 30 years. The

3:30

average was 100 years. And

3:32

some people said several hundred years. Nobody

3:35

said it would happen within 30 years. But

3:38

now everybody is saying that. Well, why

3:40

did you get it right? And they

3:43

didn't? Because I have a

3:45

graph, price performance of computation.

3:47

And it's a straight line

3:49

on a logarithmic graph.

3:52

So as you go up the graph, it represents an

3:55

exponentiation of the speed. So 1939, we...

4:00

did

4:02

0.00007

4:04

calculations per second per

4:07

constant dollar. And the

4:09

last thing we did, 135 billion calculations

4:13

per second. So

4:16

that's a 75 quadrillion

4:18

fold increase for

4:20

the same amount of money. And

4:23

I basically used that chart, basically for

4:25

most of my predictions. I

4:27

figured we needed about a trillion calculations per

4:30

second to achieve what human beings could do.

4:32

And I figured that would

4:34

take place in 2029. And we're

4:37

right on that track. It's

4:39

an absolutely straight line on a

4:41

logarithmic graph. It's

4:43

pretty extraordinary. And we

4:45

were actually changing different types of

4:48

technology. In the beginning was relays,

4:50

then discrete transistors, then integrated circuits.

4:53

No matter what type of technology we're

4:55

improving, we made the same shift

4:58

in exponential terms every

5:01

year for 81 years. It's

5:03

pretty remarkable. Yeah. You

5:05

predicted that AI will pass

5:07

the Turing test in

5:10

2029. Does that prediction still stand or might

5:12

it be even earlier? Well, most people

5:14

now are saying it's going to happen earlier.

5:16

Elon Musk says we'll do

5:19

it within two years. So I'm

5:21

saying five years. I see no reason to amplify

5:23

my prediction, but it probably will happen

5:25

sooner than that. And in

5:27

terms of the singularity you predicted it will

5:29

happen in 2045. Is that still the

5:34

case or could that be earlier? I'm

5:36

sticking with that prediction. That basically

5:38

means we'll be able to shift

5:42

our awareness to what happens

5:44

in our own brain as well as what

5:46

happens in the cloud. And

5:48

the reason that we don't get smarter is we

5:50

have a limited size for our brain. But if

5:52

we can actually go outside of our brain and

5:55

into the cloud, the cloud right now is

5:57

expanding exponentially. It's twice as fair as it

5:59

is. every year for the same money,

6:02

we'll be able to think much deeper thoughts. It'll

6:04

feel the same, except we'll be able to do

6:06

what we do now. We do it

6:09

with our phone. Phones are a little slow. We'd

6:11

like to actually just have these

6:13

thoughts appear in our brain

6:15

without having to wait to type things

6:17

in and so on. Are you thinking

6:20

of a brain-computer interface here? Yes,

6:22

exactly. But if you go back

6:24

to 2005, we didn't even have phones. Very

6:27

few people had phones, and if you had a

6:29

phone that did very little, I've actually done polls

6:31

where I've asked people who has their phone,

6:34

everybody raises their hand. Nobody goes outside

6:37

without their phone. And so we'll have

6:39

a brain computer cloud interface that no

6:41

one will think twice about is what

6:43

you're saying. Right. And people

6:46

say, oh, I wouldn't do that. But

6:48

people didn't think they would use phones 20 years

6:51

ago either. Carry around this

6:53

thing and lose it. But

6:56

now everybody has a phone, and

6:58

we'll be able to interface our

7:00

brain wirelessly to

7:02

the cloud, which is growing exponentially.

7:04

But a lot

7:06

of things are happening now that we only

7:08

just thought about in 2005. For

7:11

example, we're doing now biology this

7:14

way. For example, Moderna, they

7:16

actually gathered several billion

7:19

different possibilities for what could

7:21

cause COVID. And

7:23

generally, we would go through that with

7:25

personal trials, but they actually did it

7:28

with simulated trials. And they did that

7:30

in two days, using

7:33

AI, different types of techniques to

7:35

eliminate various possibilities and came

7:37

out with one. And that's still

7:40

the solution that we're using

7:42

now for COVID was

7:44

done in two days, a few

7:46

years ago. And actually, the variation

7:49

of that is now dealing with cancer.

7:52

We have our first trial with

7:54

a thousand people in England,

7:56

where we're testing pancreatic and

7:59

breast cancer. cancer and several other

8:01

different types of cancer using something

8:03

very similar and getting very positive

8:05

results so far. We don't have the final results

8:08

yet, but that never could

8:10

take place before. This is actually the very

8:12

first time that this has happened. So

8:14

we're going to go through all kinds

8:16

of diseases using simulated

8:18

biology. As of two

8:21

or three years ago, we had gone

8:23

through 170,000 proteins. That was actually done by

8:25

person, but that was

8:28

a very small fraction of them. And

8:30

then alpha fold two in 2022 went through 200 million

8:35

proteins and got their shape exactly correct.

8:38

And that's going to help come up with new

8:41

medications. So there's a lot

8:43

of things that we had imagined or that I

8:45

imagined in 2005 that

8:48

are now actually coming true. You had

8:50

also imagined, if I remember way back in the

8:52

day, that people might live forever.

8:55

That's not exactly what I'm saying.

8:57

We're going to reach longevity escape

9:00

velocity. Right now you go through

9:02

a year, you use up a year of your longevity. However,

9:05

scientific progress is also progressing

9:07

and you get back a

9:09

certain number of months from

9:11

that. Right now you're getting back about four

9:14

months a year. So you lose a year, but you

9:16

get back four months or only losing eight

9:18

months. However, the scientific

9:20

progress is progressing exponentially. Here

9:23

between 2029, 2035, you'll get back not four months, but a full year. So you

9:25

use up

9:29

a year, but you get back a year and

9:31

you'll basically stay in the same place as far

9:34

as your longevity is concerned. Now

9:36

that's not a guarantee. You could have a

9:38

20 year old and you can compute their

9:40

life expectancy as many decades and they could

9:42

die tomorrow, let's say from an

9:44

accident. Although we're also dealing

9:46

with accidents. For example, cars right

9:48

now we lose 40,000 lives

9:51

from humans driving cars. When

9:54

a computer drives a car, it's

9:56

going to be much safer. In fact, that's one of

9:58

the reasons it's held up. We don't want to. on

10:00

any accidents. And so we're going to get rid of

10:02

a lot of accidents in cars and, and

10:05

other places. But it's not

10:07

a guarantee. You could develop a

10:09

disease that we don't have a cure for yet. But

10:12

as we go through a year, you're not going to be

10:14

using up a year of your longevity. Because

10:17

because we'll have new medicines and vaccines

10:19

and things and, and body parts. Yes.

10:22

Because we wear out our body

10:24

parts, right? Absolutely. In fact,

10:26

I'm involved with a company where we're

10:28

creating artificial kidneys. And

10:30

people are working on artificial hearts, artificial

10:33

lungs. So we're going to

10:35

be recreating body parts as well. And we're

10:37

going to be making progress at a much faster

10:40

pace than we've done before. As

10:43

we watch artificial intelligence, AI

10:46

skyrocketing, we're also

10:48

aware that it uses a lot of energy,

10:51

right? It's it's using more energy

10:53

than Bitcoin mining these days.

10:56

Can we achieve the singularity without

10:58

completely depleting our energy sources and

11:00

our grids? Well, we're

11:02

also making fantastic progress on converting

11:06

renewable resources

11:08

into energy. Right now

11:10

solar cells are 99.7% cheaper.

11:13

Since 1975. We've increased the amount of energy

11:18

coming from that 2 million fold. And

11:21

the total amount of energy we're getting

11:23

from solar and wind, it's

11:25

doubling every four years. So

11:28

at that rate, in about 10 years, we'll

11:30

be getting all of our energy that way and it'll

11:32

keep getting cheaper. So ultimately, it'll

11:35

be very inexpensive. We

11:37

actually have 10,000 times more sunlight

11:39

that falls on the earth than we

11:41

need to meet all of our energy needs. Some

11:43

of it's hard to get to, but we

11:46

have plenty of headroom and we'll be able to

11:48

do more than do that within 10 years. Support

11:55

for Science Friday comes from the Rita Allen

11:57

Foundation. More at Rita allen.org. Do

12:04

we have enough headroom for making

12:07

computer chips that all this stuff is going to

12:09

need? I mean, the basic elements

12:11

needed for computing. What about mining of

12:13

the materials we need for it? That's

12:16

all part of the process. I mean, it's gone on

12:18

for 81 years. We find the

12:20

materials, we're able to create

12:23

many more computations per second,

12:26

include in the book an analysis

12:28

of what the ultimate would be. We call

12:31

that computronium, and we're very,

12:33

very far from that. So we can keep

12:35

this exponential growth going

12:38

for a very long time. There's

12:41

a climate change argument against

12:43

the massive use of computing

12:46

and AI. Can they work together? Well,

12:49

that's only true if we use

12:51

non-renewable sources. If we use oil

12:53

and coal and so on that

12:56

add to carbon dioxide,

12:59

these new forms of solar

13:01

and wind don't use any

13:03

of that. And we also really

13:05

can go and use all of our

13:07

energy that way and completely

13:09

eliminate adding carbon dioxide to

13:11

develop energy. That's where we're

13:13

headed. Do you think

13:15

we could ask AI to help us solve

13:18

the climate change problem? Well,

13:21

it's doing that by converting

13:23

things that don't add carbon dioxide to

13:26

create energy, and that's going

13:28

at an exponential rate. So as

13:30

we get to the 2030s, we'll actually be able to reduce

13:35

the amount of carbon dioxide we're creating.

13:38

Moving on to the dominance of

13:41

the AI industry, do you think there needs

13:43

to be checks and balances so

13:45

that AI doesn't get out of hand? I mean,

13:47

how can we make progress without

13:49

losing control in this race to the top?

13:51

Or do you expect that AI will take

13:54

us over in that singularity? Well,

13:56

yes and no. I mean, we do need some more

13:59

guidelines. we actually had

14:01

an acilimar meeting, because

14:03

the acilimar meetings on biotechnology have worked

14:05

out very well, and

14:08

we're doing that now for AI. There

14:11

is a lot of guidelines already.

14:14

I mean, any of these major companies, if

14:17

they put out something that causes a great deal

14:19

of harm, there'll be a lot of liability. So

14:22

they're very motivated to avoid that. And

14:25

I would say in all of these new changes,

14:27

they put more effort into avoiding

14:29

problems and creating the new developments.

14:33

So the insurance companies, as they

14:35

always do. Yeah, well, that's actually

14:38

part of the motivation. And

14:41

sometimes people look at the problems we're

14:43

creating and they can imagine the problems

14:46

and they don't imagine any new solutions. And they

14:48

go, gee, if somebody put this out,

14:51

we'd have a big problem. But

14:53

we're actually making advances in how to avoid

14:55

problems at the same time that

14:57

we're creating these new problems. So

15:00

far it's worked out well. We've

15:02

had this concern. I mean, the light movement started

15:04

200 years ago. They

15:07

were correct that things like the cotton genie and

15:09

so on did eliminate lots of

15:11

these jobs. We created new

15:13

jobs. And if you ask, well,

15:15

gee, what are the new jobs gonna be? And

15:18

you ask that 200 years ago, no

15:20

one could answer it because we hadn't even invented these

15:23

new jobs in the new industries that they were employed

15:25

in. So I'm

15:27

optimistic, but I mean, there are definitely

15:30

problems. Intelligence is really

15:32

the most important resource

15:35

we have. And if you have somebody that's not

15:38

of right mind who has a lot of intelligence,

15:40

that could be a problem. So

15:42

I'm not saying there's no issues. I mean, we really have

15:44

to pay attention to this. Speaking

15:46

of paying attention, what about the

15:49

upcoming elections, the possibility of phony

15:51

AI, AI generated statements and photos?

15:54

I mean, that's probably the biggest problem

15:56

that AI is presenting for this

15:58

election. It probably will. take place

16:00

right before the election so there won't be time to

16:02

actually sort it out. Because if

16:05

it happened now, we'd be able to figure out,

16:07

okay, that's probably a fake

16:09

answer. I mean, you see Biden

16:11

and he says something, it looks real, but

16:14

it's completely fake. And

16:16

that could cause a lot of problems. It

16:19

hasn't really happened to

16:21

the major players. So I'm

16:24

not saying there's no problems. I mean, we really do have

16:26

to pay attention to this. I

16:28

think we'll make it through. But these

16:30

are definitely issues we didn't have in

16:32

the last election, for example. Yeah. And

16:34

as we move toward the singularity, what

16:36

concerns you most about that? We're

16:39

going to have to make changes much more quickly. You ask

16:41

actually people in the side polls

16:44

where people think that poverty

16:46

is getting worse. 80% of

16:48

the people think that poverty is getting

16:51

worse. It's actually been reduced to 50%

16:53

in the last 20 years. And

16:56

so we're actually doing much better than we

16:58

thought, really because of technology.

17:00

But we had time to figure

17:03

that out. I mean, how many jobs were lost

17:05

from the railroad? But it

17:07

took decades for that to happen. Now it's

17:10

happening very, very quickly. I think

17:12

people are actually getting used to this,

17:15

but it could be a problem with

17:17

just the speed of change. We

17:20

are actually getting wealthier. U.S.

17:22

personal income per capita is actually

17:26

10 times better than it was 100 years ago.

17:29

People think, oh, gee, I'd like to live in 1900, but

17:32

they don't realize how poor we were and

17:34

how little help we got from health

17:37

technology, for example. So

17:39

things were very, very poor 100

17:42

years ago. And that's going to continue.

17:44

But we're going to have to change our expectations

17:46

to what kind of jobs we can have and

17:49

so on. So this is going

17:51

to be very rapid change. And we've

17:53

seen this in the last two years. I mean,

17:55

the AI revolution started two years ago, but

17:57

you compare the technology two years ago.

17:59

to today, it's quite astounding

18:02

how much progress has been made. Well, do

18:04

you think our job training can keep up

18:06

with this as people will need to change

18:08

the kinds of jobs and the education they

18:10

get? Yes. And it's

18:13

not exactly clear what to do. I think

18:16

we will actually have some government programs so

18:18

that your baseline, I mean,

18:20

some people can't work, for example, it's

18:23

not going to be zero. And we're

18:25

going to move towards that. But

18:27

then the real concern is having a

18:29

job that gives you satisfaction and personal

18:32

reflection and so on, not

18:34

just being able to make a living.

18:36

So that's a nice problem to have.

18:38

But that's actually going to take

18:41

a lot of concern on people

18:43

who try to plan things

18:45

like job training and so on. Ray,

18:47

we have run out of time. It's always a

18:50

pleasure to talk to you. And I wish you

18:52

good luck with your book. It's a great

18:54

read. Yes, we're very much looking forward to it.

18:56

It's coming out on the 25th of this month.

18:58

Ray Kurzweil, inventor, futurist and

19:00

author of The Singularity is Nearer.

19:03

When we merge with AI, and if you

19:06

want to read an excerpt of the book,

19:08

you can do so on our website, sciencefriday.com

19:11

slash singularity. And

19:13

that's all the time that we have

19:15

for today. A lot of folks helped

19:18

make the show happen this week, including

19:20

Dave Peterson, Sandy Roberts, Beth Rami, Robin

19:22

Kasmer, and many more. Tomorrow,

19:24

we'll talk about why tick season

19:26

is getting longer. And

19:28

some tips for spotting those pesky

19:30

arachnids. But for now,

19:32

I'm Sci-Fry producer Kathleen Davis. Thanks for

19:35

listening. At

19:45

Highland, we're all about celebrating little wins

19:49

and little ways to innovate digital processes.

19:53

There's no customer pain point too small for us

19:55

to help with. Maybe that's

19:57

why more than half of the Fortune 100

20:15

Start your summer road trip at Midas and get up to

20:18

$30 off your next repair service. Plus get a free closer

20:20

look vehicle check to make sure you're road trip ready. So

20:22

if you need a brake service and alignment check or tune

20:24

up, hit up Midas for up to $30 off. For

20:27

more details, request your appointment at midas.com.