Transcript of YouTube Video: Computers Can Predict When You're Going to Die… Here's How

Transcript of YouTube Video: Computers Can Predict When You're Going to Die… Here's How

Welcome to our collection of transcripts of YouTube videos, where we provide detailed text versions of "Computers Can Predict When You're Going to Die… Here's How" content.

00:00

(peaceful music) (Joe sighs)

00:02

- I shouldn't do this.

00:03

I'm just gonna take one quick peek at the comments. Okay.

00:07

Oh.

00:08

"I always learn so much from Joe." That is so nice.

00:12

"Is it me or is he getting old?"

00:18

Old!

00:19

You're gonna die!

00:20

You're gonna die!

00:21

Hey, old man. You're ancient.

00:23

Old.

00:24

Death, death, death.

00:29

(Joe screams) (blender whirring)

00:31

Hey, smart people. Joe here, but for how long?

00:35

I'm just trying to cheat death

00:36

with this life-extending smoothie recipe.

00:39

I found it on TikTok.

00:40

(Joe gulps)

00:44

Tastes like youth and burning plastic.

00:49

Let's face it. We're all a little bit like Barbie.

00:51

We think about death a lot.

00:53

According to a 2022 survey,

00:55

half of all Americans think about death monthly,

00:58

and one out of three Gen Z'ers thinks about death daily.

01:01

Should probably start doing these without the box.

01:04

Death anxiety, or thanatophobia,

01:07

is a perfectly natural human feeling.

01:09

'cause, let's face it, being alive is pretty cool

01:12

when you consider the alternatives.

01:14

The global health and wellness market

01:16

is estimated at $1.8 trillion.

01:19

Yes, that's trillion with a T.

01:22

Truth is, no matter how hard we try to cheat death,

01:24

it could happen at any moment.

01:27

We can't predict our death, or can we?

01:32

Come on, folks. This is a science show.

01:35

You didn't really think that I was gonna.

01:37

Right now, there are people out there

01:39

predicting your death and mine,

01:42

distilling our lives into data points,

01:44

feeding it into lifeless machines,

01:46

and calculating with an uncanny level of accuracy

01:49

when someone exactly like you or me is gonna die.

01:53

I'm talking about predictive analytics.

01:59

(lively theme music)

02:01

Predictive analytics is a branch of mathematics

02:04

that uses historical data

02:05

to make predictions about future outcomes,

02:08

and it's everywhere.

02:10

Shopping, sports, social media algorithms, fraud detection,

02:14

politics, and deciding if you'll see this YouTube video,

02:18

because if a government or business can know

02:20

what's gonna happen before it happens, that's pretty useful.

02:24

It turns out we've been using math

02:26

to predict people's deaths for centuries.

02:29

By the 1600s, humans were shipping goods around the world

02:32

and you could make serious bank doing it

02:34

as long as your ship didn't sink.

02:36

Captains and the people who funded their voyages

02:39

had more to worry about than just weather.

02:41

The late 1600s were also the golden age of piracy.

02:45

To the Lloyd's Company of London,

02:47

these hooks-for-hands hooligans looked like an opportunity.

02:51

They started crunching numbers,

02:52

using past data to help predict

02:54

how dangerous a particular sea voyage would be.

02:57

Then, Lloyd's would offer insurance

02:59

to help cover the risk of the trip.

03:01

The more risk the calculations showed,

03:03

the higher the insurance would cost,

03:05

and lo and behold, the insurance industry was born.

03:09

Today, Lloyd's of London is

03:11

one of the largest insurers in the world,

03:13

and the predictive analytics they pioneered

03:15

are still used to predict risks and outcomes today,

03:19

only now it's powered by computers and they're good at it.

03:23

If you're thinking,

03:24

"Hey, I'm a complex and free will-having snowflake.

03:28

You can't predict me."

03:30

Think again.

03:31

- We tend to think that we're pretty unpredictable,

03:34

but if you think about it, if I were to guess for you

03:37

or almost anyone else where you'll be at 4:00 AM tomorrow

03:42

or a month from now, or a year from now,

03:44

you're gonna be at your house in bed,

03:46

and it's every single day.

03:48

Where are you gonna be in the daytime?

03:49

Well, you're gonna be at your work,

03:51

but that's one example of

03:53

where we have a lot of predictability,

03:55

but we're kind of blind to it.

03:56

- That's right.

03:57

Every day, you leave invisible breadcrumb trails

04:00

of data and behavior that you don't even think about.

04:04

Like, you might have apps on your phone that track

04:06

how many hours you slept last night,

04:08

or a metro card you use to catch the bus,

04:11

and you ordered coffee on the way to the bus stop.

04:14

I hate to tell you this, but somewhere out there,

04:16

somebody knows about all the websites you've visited.

04:19

Yes. Even that one.

04:22

That's all data, and it turns out, so are we.

04:26

- The idea, Joe, is that, you know,

04:29

any one person is just a data point.

04:32

Probabilities of mortality or longevity

04:35

are gonna play out for that person individually.

04:37

- In order to make predictions about

04:40

us walking human data points,

04:42

the computers have to pile us together

04:44

and create a sort of statistical Franken-human

04:47

that represents the whole bunch.

04:50

This is the mathematical theory

04:51

called the law of large numbers.

04:54

Basically, the larger your data sample is,

04:57

the more likely it is that the average of that sample

04:59

will reflect what actually happens.

05:02

- Again, the benefit might be in studying, you know,

05:05

10,000, 100,000 people

05:07

who have potentially some similar characteristics to you.

05:11

They might be of a similar age.

05:13

They might be male.

05:14

They might have a general same health and wellness.

05:16

- This is where things get really complicated.

05:19

In order for us to get accurate predictions of the future

05:23

based on past data, we first have to figure out

05:26

all the potential outcomes

05:27

that could happen around an event.

05:29

Here's how predictive analytics works in the simplest terms.

05:33

Say I have a bag with 20 marbles in it,

05:35

where some are red, some are yellow, and others are blue.

05:39

If I pull one marble out,

05:41

I can't accurately predict which color I'll grab,

05:44

but if I were to draw 100,000 times,

05:48

I could calculate the likelihood

05:49

of anyone pulling a particular color with extreme accuracy,

05:54

as long as I don't lose my marbles first.

05:57

To be even more accurate with our prediction,

05:59

we could even start factoring in other data

06:02

like the weight or size of different marbles

06:04

and how that may affect their distribution in the bag.

06:07

The point is that multi-factored data,

06:09

considering all of the different factors

06:11

and how big or small their influence is on the outcome,

06:15

that can improve our ability to predict the future.

06:19

So marbles are great, but when are we gonna die?

06:23

There are so many potential factors

06:25

we have to predicatively analytic-size.

06:28

Have a chronic illness? Love scuba diving?

06:31

These are potentially negative factors.

06:34

Exercise regularly? Eat well?

06:36

Got access to good healthcare?

06:38

Looking at you here if you're up in Canada.

06:40

Well, these are all positive factors

06:42

in the mathematics of mortality,

06:45

but some of these things are more likely than others

06:47

to put you in a speedboat across the River Styx,

06:51

so some factors get more weight in different scenarios.

06:55

If that sounds like there are

06:56

an almost overwhelming number of factors to consider

06:59

when predicting the future, you're right.

07:02

The future is complicated. At least, I think it will be.

07:05

That is why scientists are using machine learning

07:08

to look at more and more complex factors

07:10

and figure out which ones are actually important.

07:13

When they said AI was gonna be responsible for our death,

07:16

I don't think this is what they meant.

07:18

So what kind of data do you need

07:22

in order to construct an algorithm like this?

07:25

- So if you're predicting how long someone will live,

07:28

you look at their age and lots of other properties

07:31

that the insurance companies, you know, have tallied out,

07:35

but the algorithm that we do, we do something different.

07:37

We basically say to you,

07:38

we're gonna put your whole life in the mathematical model,

07:41

and then the model will tell what's important.

07:43

So you can put in lots of stuff

07:45

that you might not think was important,

07:47

but that the model will then learn

07:49

that actually is one of the things

07:50

that tells you something about our future behavior.

07:53

- [Joe] Typically, human analysts would select

07:55

the factors that they think are likely

07:57

to predict some outcome,

07:59

and they'd test how much weight they should be given

08:01

in the calculation,

08:03

but what factors are selected or not selected

08:06

may be affected by human bias.

08:08

That machine-learning algorithm instead feeds

08:11

all possible factors into the system

08:13

and lets it select and weight factors, free from human bias.

08:18

The algorithm analyzes a person's life

08:21

the way a large language model analyzes words.

08:24

Where a language model calculates patterns of words

08:26

that are likely to be associated with each other

08:29

and uses those to create future language,

08:32

Sune's mortality model looks for patterns of behavior

08:35

and demographics that are likely to be associated with death

08:39

and instead of language, it writes the story of a life.

08:43

- If you look at, let's say, income,

08:46

it would say that, if all other things are equal,

08:49

if we kind of take you

08:51

and increase the income for your data point,

08:55

then you have a higher probability of surviving,

08:57

and that lines up with what we know

08:59

from existing social science, that if you're wealthy,

09:03

you basically have a better chance of living a long life.

09:06

- They first trained this model

09:07

on a large multi-factor data set

09:09

pulled from health and demographic statistics in Denmark

09:13

and compared this to actual death records

09:15

to gauge its accuracy.

09:17

They then tested the model by feeding it

09:19

a set of people where half survived and half died.

09:22

If we were to randomly predict

09:23

if a particular one of these people survived,

09:26

we would expect to get the answer right 50% of the time.

09:30

Their AI predictive mortality model

09:32

was able to guess right 8 out of 10 times.

09:36

Right now, this AI mortality model

09:38

is being used as a research tool

09:40

to create better models in the future

09:42

so I can't ask it when I'm gonna die.

09:45

So I decided to ask an actual actuary.

09:49

Well, as an actuary, have you ever missed a flight?

09:51

- I think I am 100% on making my flights.

09:55

- So I sent Dale a bunch of information about me,

09:58

like my height, age, and some of my habits.

10:01

All good ones, mind you, and Dale crunched the numbers.

10:04

- I'm gonna estimate, Joe, that you're around,

10:08

you know, say, 40 years old, give or take.

10:11

- That's a good estimate.

10:12

It's fine. It's close, yeah.

10:14

- And put that in there.

10:16

I'm going to then select that you're a male.

10:21

You do not smoke.

10:23

So I have, you given this longevity illustrator,

10:28

to be around a life expectancy

10:33

of 86.

10:35

You have a 37% probability of actually living to age 90.

10:39

You also, by the way,

10:41

have an 8% chance of living to age 100.

10:44

- This is the best news I've heard all day.

10:48

I thought you were gonna say like 70, 75,

10:51

something like that.

10:52

- Well, remember, some of those life expectancies

10:54

that you hear quoted are life expectancies at birth,

10:58

and so you've had the benefit of surviving

10:59

the first 40 or so years

11:01

and past some of the hazards or risks

11:05

that might unfortunately lead to some early deaths,

11:08

and so I would encourage you to do a little bit of thinking

11:11

of, all right, what are some of the planning

11:12

I might want to do should I live to that age?

11:15

- This is fantastic.

11:18

My gym membership has gotta be

11:20

the greatest investment I've ever made in my entire life,

11:23

and I hope all the YouTube commenters are listening.

11:25

You hear that? I don't look old.

11:28

Okay, anyway.

11:29

And even with all the data in the world,

11:32

there will still be outlier events

11:34

that we could never see coming, so-called black swan events

11:38

that are moments of totally unpredictable chaos.

11:42

That said, these things are accurate,

11:45

almost scary accurate.

11:47

As for me, I'm glad that I met with Dale

11:50

and that he gave me a number.

11:53

As a scientist, I love numbers, and as a person who's alive,

11:57

I love that I'll probably get to stay that way

11:59

for a long time.

12:01

The most important thing I learned is that,

12:03

even though the mathematical tools

12:05

that predict our lives and our actions

12:08

are uncannily accurate, we still have power to make choices

12:13

that can change those predictions,

12:15

to leave new breadcrumb trails of data

12:19

that might lead to different destinations.

12:21

At the end of the day, all of us only have

12:23

a little time on this blue rock we call home.

12:27

Math and science and predictive analytics

12:30

can help us make the most of it.

12:33

At the very least, it'll suggest some good videos

12:36

to watch while we're waiting.

12:39

Stay curious.

12:42

Hey, thanks for sticking around to the end of the  video. Hope you enjoyed that one. And as always,  

12:47

I would like to thank everybody who supports this  show on Patreon. If you don't like predictive  

12:52

analytics algorithms telling you about every  video that you should watch and you want to  

12:57

take some of that power back for yourself, well  Patreon's a great way to do that by signing up  

13:01

for Patreon. You'll find out about videos early,  you'll get to watch them before anybody else.  

13:06

And it's just you and me without any of those  computers in the way. I mean, there'll be a  

13:10

computer in the way 'cause you have to watch  it on a computer, but it's like it's a good,  

13:15

you know what I mean? Check out the LinkedIn  and description. I'll see you in the next video.

13:19

"Hank Green is my favorite."

13:22

I'm not.

13:24

I'll take it.

13:25

- [Producer 1] Those are great.

13:29

- [Producer 2] Yeah.

13:32

(crew chuckles)