Audience: Apple. Kenneth Cukier: Apple. Of course it is.

How do we know it?

Because of data.

You look at supermarket sales.

You look at supermarket sales of 30-centimeter pies

that are frozen, and apple wins, no contest.

The majority of the sales are apple.

But then supermarkets started selling

smaller, 11-centimeter pies,

and suddenly, apple fell to fourth or fifth place.

Why? What happened?

Okay, think about it.

When you buy a 30-centimeter pie,

the whole family has to agree,

and apple is everyone's second favorite.

(Laughter)

But when you buy an individual 11-centimeter pie,

you can buy the one that you want.

You can get your first choice.

You have more data.

You can see something

that you couldn't see

when you only had smaller amounts of it.

Now, the point here is that more data

doesn't just let us see more,

more of the same thing we were looking at.

More data allows us to see new.

It allows us to see better.

It allows us to see different.

In this case, it allows us to see

what America's favorite pie is:

not apple.

Now, you probably all have heard the term big data.

In fact, you're probably sick of hearing the term

big data.

It is true that there is a lot of hype around the term,

and that is very unfortunate,

because big data is an extremely important tool

by which society is going to advance.

In the past, we used to look at small data

and think about what it would mean

to try to understand the world,

and now we have a lot more of it,

more than we ever could before.

What we find is that when we have

a large body of data, we can fundamentally do things

that we couldn't do when we only had smaller amounts.

Big data is important, and big data is new,

and when you think about it,

the only way this planet is going to deal

with its global challenges —

to feed people, supply them with medical care,

supply them with energy, electricity,

and to make sure they're not burnt to a crisp

because of global warming —

is because of the effective use of data.

So what is new about big data? What is the big deal?

Well, to answer that question, let's think about

what information looked like,

physically looked like in the past.

In 1908, on the island of Crete,

archaeologists discovered a clay disc.

They dated it from 2000 B.C., so it's 4,000 years old.

Now, there's inscriptions on this disc,

but we actually don't know what it means.

It's a complete mystery, but the point is that

this is what information used to look like

4,000 years ago.

This is how society stored

and transmitted information.

Now, society hasn't advanced all that much.

We still store information on discs,

but now we can store a lot more information,

more than ever before.

Searching it is easier. Copying it easier.

Sharing it is easier. Processing it is easier.

And what we can do is we can reuse this information

for uses that we never even imagined

when we first collected the data.

In this respect, the data has gone

from a stock to a flow,

from something that is stationary and static

to something that is fluid and dynamic.

There is, if you will, a liquidity to information.

The disc that was discovered off of Crete

that's 4,000 years old, is heavy,

it doesn't store a lot of information,

and that information is unchangeable.

By contrast, all of the files

that Edward Snowden took

from the National Security Agency in the United States

fits on a memory stick

the size of a fingernail,

and it can be shared at the speed of light.

More data. More.

Now, one reason why we have so much data in the world today

is we are collecting things

that we've always collected information on,

but another reason why is we're taking things

that have always been informational

but have never been rendered into a data format

and we are putting it into data.

Think, for example, the question of location.

Take, for example, Martin Luther.

If we wanted to know in the 1500s

where Martin Luther was,

we would have to follow him at all times,

maybe with a feathery quill and an inkwell,

and record it,

but now think about what it looks like today.

You know that somewhere,

probably in a telecommunications carrier's database,

there is a spreadsheet or at least a database entry

that records your information

of where you've been at all times.

If you have a cell phone,

and that cell phone has GPS, but even if it doesn't have GPS,

it can record your information.

In this respect, location has been datafied.

Now think, for example, of the issue of posture,

the way that you are all sitting right now,

the way that you sit,

the way that you sit, the way that you sit.

It's all different, and it's a function of your leg length

and your back and the contours of your back,

and if I were to put sensors, maybe 100 sensors

into all of your chairs right now,

I could create an index that's fairly unique to you,

sort of like a fingerprint, but it's not your finger.

So what could we do with this?

Researchers in Tokyo are using it

as a potential anti-theft device in cars.

The idea is that the carjacker sits behind the wheel,

tries to stream off, but the car recognizes

that a non-approved driver is behind the wheel,

and maybe the engine just stops, unless you

type in a password into the dashboard

to say, "Hey, I have authorization to drive." Great.

What if every single car in Europe

had this technology in it?

What could we do then?

Maybe, if we aggregated the data,

maybe we could identify telltale signs

that best predict that a car accident

is going to take place in the next five seconds.

And then what we will have datafied

is driver fatigue,

and the service would be when the car senses

that the person slumps into that position,

automatically knows, hey, set an internal alarm

that would vibrate the steering wheel, honk inside

to say, "Hey, wake up,

pay more attention to the road."

These are the sorts of things we can do

when we datafy more aspects of our lives.

So what is the value of big data?

Well, think about it.

You have more information.

You can do things that you couldn't do before.

One of the most impressive areas

where this concept is taking place

is in the area of machine learning.

Machine learning is a branch of artificial intelligence,

which itself is a branch of computer science.

The general idea is that instead of

instructing a computer what to do,

we are going to simply throw data at the problem

and tell the computer to figure it out for itself.

And it will help you understand it

by seeing its origins.

In the 1950s, a computer scientist

at IBM named Arthur Samuel liked to play checkers,

so he wrote a computer program

so he could play against the computer.

He played. He won.

He played. He won.

He played. He won,

because the computer only knew

what a legal move was.

Arthur Samuel knew something else.

Arthur Samuel knew strategy.

So he wrote a small sub-program alongside it

operating in the background, and all it did

was score the probability

that a given board configuration would likely lead

to a winning board versus a losing board

after every move.

He plays the computer. He wins.

He plays the computer. He wins.

He plays the computer. He wins.

And then Arthur Samuel leaves the computer

to play itself.

It plays itself. It collects more data.

It collects more data. It increases the accuracy of its prediction.

And then Arthur Samuel goes back to the computer

and he plays it, and he loses,

and he plays it, and he loses,

and he plays it, and he loses,

and Arthur Samuel has created a machine

that surpasses his ability in a task that he taught it.

And this idea of machine learning

is going everywhere.

How do you think we have self-driving cars?

Are we any better off as a society

enshrining all the rules of the road into software?

No. Memory is cheaper. No.

Algorithms are faster. No. Processors are better. No.

All of those things matter, but that's not why.

It's because we changed the nature of the problem.

We changed the nature of the problem from one

in which we tried to overtly and explicitly

explain to the computer how to drive

to one in which we say,

"Here's a lot of data around the vehicle.

You figure it out.

You figure it out that that is a traffic light,

that that traffic light is red and not green,

that that means that you need to stop

and not go forward."

Machine learning is at the basis

of many of the things that we do online:

search engines,

Amazon's personalization algorithm,

computer translation,

voice recognition systems.

Researchers recently have looked at

the question of biopsies,

cancerous biopsies,

and they've asked the computer to identify

by looking at the data and survival rates

to determine whether cells are actually

cancerous or not,

and sure enough, when you throw the data at it,

through a machine-learning algorithm,

the machine was able to identify

the 12 telltale signs that best predict

that this biopsy of the breast cancer cells

are indeed cancerous.

The problem: The medical literature

only knew nine of them.

Three of the traits were ones

that people didn't need to look for,

but that the machine spotted.

Now, there are dark sides to big data as well.

It will improve our lives, but there are problems

that we need to be conscious of,

and the first one is the idea

that we may be punished for predictions,

that the police may use big data for their purposes,

a little bit like "Minority Report."

Now, it's a term called predictive policing,

or algorithmic criminology,

and the idea is that if we take a lot of data,

for example where past crimes have been,

we know where to send the patrols.

That makes sense, but the problem, of course,

is that it's not simply going to stop on location data,