Original subtitles

BEN: It's the DNA of the next tech revolution.

This is such a huge data set that there's no way that a human

or even a team of humans can look at all of it.

The race for artificial intelligence is on.

In quantum, you have this thing called a qubit,

which can be zero and one at the same time,

and that's where the power comes from.

AI might make our lives better...

The life expectancy for human civilization

might then easily measure in billions of years.

...but could it destroy the human race?

You ought to be really concerned about the strong AI

that has guns on it.

♪

As the host of Cyberwar, I've travelled the world to talk to

brilliant hackers, scientists and programmers,

and many of them have told me they're thinking about

artificial intelligence.

Today we're in the middle of an AI boom.

Huge advancements in artificial intelligence

have made things like Siri and self-driving cars possible.

AI is in video games, security surveillance systems,

smart home devices, and advanced weapons systems.

But as researchers race to make the next breakthrough,

a lot are warning that we haven't actually

thought this through.

And it's not necessarily the Terminator scenario

they're afraid of, where an AI system becomes self aware

and decides to kill us all.

The real threat could lie in the unintended consequences

no one sees coming.

I'm at Stanford University, where researchers are designing

AI that comes in a very non-threatening package.

Meet Jackrabbot, a cute little machine that's built to navigate

and move through crowds of human buddies.

This is a robot that's programmed to learn on its own,

through example.

This technique, called "deep learning",

has been fuelling a lot of recent AI breakthroughs.

Alexandre Alahi and Alexandre Robicquet are part of a team

that built Jackrabbot, and they say their little creation

is more advanced than a self-driving car.

So the robot enter the scene, analyze it, and then

navigate through it according to all the data that we gather.

So it's just watching people, and looking at people,

and what they're doing, and judging, you know,

the space and the time?

Yeah, and also trying to understand how they interact

between each other.

The main aspect would be to understand well, how...

what is a safe distance that I keep with someone else,

or how do I accelerate or decelerate when I get close

to someone and react accordingly?

So how do you gather that information?

You just like... this just looks like

some sort of surveillance footage.

So last summer, our team spent like 2 months gathering daily

at rush hour, top view data from the Stanford crowd,

so we can actually understand how they avoid,

how they behave and how they navigate in such a crowd.

Let's go, robot.

Free!

So wait, he's being controlled by you?

For right now, yeah.

So does he have a fully autonomous mode?

Yeah, he could.

And what does he do? He just follows?

No, we actually can give him-- like if you map the place

and the area, he can go from point A to point B,

but we're just having a simple collision avoidance right now.

But like you train to avoid every single obstacle

and go where you want him to go.

Can we drive him around a little bit?

Yeah, absolutely.

And those-- that spinning thing, that's just like a 3D sensor?

Right, right.

So we're capturing depth data, visual data,

and we're combining all these sensors to detect humans,

understand its surrounding to look at itself,

and predict also where people will go.

Do people freak out when they see it a lot?

No, they like it a lot.

They come hug it, they talk to it.

He doesn't... He cannot talk yet.

Is that what you hope the rise of the machines is?

It's a friendly rise of the machines?

For that one, yes.

- Yeah. - Absolutely.

Jackrabbot's creators hope that as AI advances,

machines like this will be built to carry luggage

through airports, or help the blind navigate

through pedestrian traffic.

The Stanford team is working to get AI to the point

where Jackrabbots can one day do

what most people can just do by nature.

But today's computers are already performing tasks

no human is capable of performing,

even if it's not quite AI.

That technology continues to evolve,

and the so-called supercomputer is at the cutting edge.

Bryan Biegel works in NASA's Advanced Supercomputing

Division, where they're working with machines that can actually

help us see into the future...

Let's go to the vislab, and I'll show you some of the things

that our users do with our supercomputers.

...using code so advanced it took 15 years to write.

Yeah, so this is actually a simulation of Earth's oceans.

This is... It uses measured data from NASA's satellites

and puts it into a really huge computer model

on the supercomputers, and predicts the ocean behaviour.

So this single simulation actually used up to 70,000

processors, and generated over 3 petabytes of data.

That's like 10,000 times what you would have

on your entire laptop.

Let me show you this in full scale.

This is actually scaled down so you can see it on the monitors,

but if I go to the next one,

you can see what the full-scale version of this simulation is.

For example, this is the ice over Antarctic Ocean,

and you can see where it cracks,

and it sends ripples out into the ocean.

And they're working on...

they're still continuing to advance the accuracy of this

model so they can predict decades into the future.

Would you say this is artificial intelligence?

It seems like it's getting there, right?

It's so complex, but we have programmed in this case

every single line of this code.

It's a little different than artificial intelligence,

where it's more of a-- even more of a black box,

where we don't know exactly how it's gonna work

and how that artificial brain is gonna evolve.

Why would NASA be interested in artificial intelligence?

Well, there are a few reasons.

Of course NASA's trying to send artificial probes further and

further out into the solar system, and eventually beyond.

We can't program those for all of the things that

it's gonna encounter, so we need them to be able to do their job

without us constantly telling them everything to do.

Even in interpreting data like this,

this is such a huge data set that there's no way that

a human or even a team of humans can look at all of it.

Computing is gonna continue to expand dramatically over even

the next 10 years, and so we'll be able to do an even better job

of modeling the history of the universe,

the future of the universe, the history and future

of our planet, and go further in exploring our universe.

to artificial intelligence could be creating

systems that take cues from the way our own brains work,

using a process called "deep learning".

I'm here at Berkeley to meet one of the leading minds

in the development of AI.

Stuart Russell has been at this for more than 20 years.

He co-wrote THE definitive textbook on AI,

and he understands both its promise and its dangers.

So the brain has an enormous network of neurons.

So a neuron is a cell that has these long,

thin connections to other neurons,

so it kind of looks like a big tangle of electrical spaghetti.

And we have tens of billions of neurons,

and deep learning networks are much, much, much simpler.

What's similar about them is that both the brain and these

deep learning networks can learn how to perform a given function,

and they learn by being given lots of examples.

But deep learning networks, for example,

if you want to learn to recognize cats and dogs

and candles and... staplers and things like that,

you can show it millions of photographs with these things,

with labeled things is what they are.

So on tasks like recognizing a wide range of categories,

the computations they run now have a thousand different

categories of objects.

And in the last five years, we've gone from systems that

might get 5% accuracy on that thousand category task

to systems that are getting 98% accuracy.

So you've been at this for a while.

I mean, how does it make you feel, that the progress now

is becoming exponential?

The exponential word is a dangerous one, 'cause...

(Laughing)

'cause it tends to suggest that things will continue to

accelerate without end, but it may turn out that it will

plateau, and that for other tasks we need new breakthroughs.

Quantum computing?

Quantum computing possibly,

but that's kind of a cheat in some sense.

That says that rather than really understand

how the brain manages to do these amazing tasks

with what is really not that much hardware,

and so I almost hope that quantum computing...

Doesn't happen.

Doesn't happen, or happens a long time in the future where--

and you know, give us a chance to keep working on finding

the secrets of intelligence, the things that make us smart,

and gain real understanding.

'Cause just brute forcing it isn't really understanding

what's going on.

Quantum computing may be the next giant leap

in building the artificial intelligence of the future.

It's a technology so powerful that we really could use it

to develop AI without having to understand

how human learning works.

And though Stuart Russell hopes that quantum computing

won't happen for a long time,

Rupak Biswas can't wait - and he isn't.

He runs the Quantum Artificial Intelligence Lab at NASA's

Ames Research Center, where scientists are working on

a powerful experimental computer called the D-Wave.

It's built on the principles of quantum mechanics,

which I was hoping Rupak could explain to me.

If someone tells you that they'll explain to you

what quantum mechanics is, you should run away from them

because no one really understands this field.

- (Laughing) - It's a very complex field.

Rupak agreed to show me NASA's quantum computer.

This is the D-Wave quantum annealing system.

So what you'll see here is basically a black box,

and that is where the D-Wave processor is.

This is the Star Trek computer that's more powerful--

million of times more powerful than anything Microsoft's got.

Well sure, yeah, for certain classes of problems, yes.

This is something that is really more powerful,

and we expect to get a lot of research done on this.

NASA is hoping to use the quantum computer to develop

quantum algorithms, and algorithms are a key component

of the code running AI.

In layman's terms, an algorithm is a kind of recipe

for solving a problem.

Basically, it's the set of step-by-step instructions

given to a computer to help it accomplish a task.

The question here though is that the algorithm could be different

on a supercomputer than on a quantum computer,

even though you are trying to solve the same problem.

So then what is the difference?

Because, you know, to a lot of people when you say

supercomputer and quantum computer,

aren't they all supercomputers?

What's the difference?

Right, so supercomputers are basically--

it's based on this transistor, and the transistor is...

in layman terms, is basically a very small switch.

So it's either zero or one.

So a bit - and in a traditional computer, it's called a bit -

it's either zero or one, whereas in quantum you have this thing

called a qubit, which can be zero and one at the same time.

And that's essentially the difference between a computer or

a supercomputer or what we would consider classical computing

versus quantum computing, where the quantum computer allows you

to be in two stages at the same time.

And that's where the power comes from.

A quantum computer could help solve problems

that no technology can solve today;

finding cures for diseases or designing space colonies.

But what if we choose to use that awesome computing power

for destructive purposes?

You know, if you think about nuclear energy,

you can use nuclear energy to solve world's energy problems,

but you can also use it for bad things.

You know, so all of these things,

if they are used improperly and get in the wrong hands,

it could lead to trouble.

BEN: I'm in Oxford to meet the founding engineer of Skype.

Since leaving the company, Jaan Tallinn has become

one of the most prominent voices

in the field of artificial intelligence.

He sees great advantages in developing AI,

but he also warns of the risks of making the technology more

powerful without understanding its potential harms.

Once you have systems that are basically smarter than humans

when it comes to developing further intelligent systems,

then you have intelligent systems developing

intelligent systems that in turn go on to develop

even more intelligent systems.

And you have this intelligence explosion.

He thinks the next step in AI

is the creation of so-called "general intelligence".

What's the difference between what we're producing now and

"general intelligence" of the future?

If you think about a chess-playing computer,

it's just modelling the chessboard in its memory

and looking at scenarios how this game can play out,

and what are the actions that it can do on the chessboard.

So it actually would be good from a chess-playing perspective

to not only model the chessboard,

but also what's going on in the brain of your opponent.

So it's one thing for a computer to understand

the game of chess, but if it can look up from the board

and understand me, that's a whole new level of scary.

But I'm not the only one that's worried.

In 2015, a group of prominent thinkers signed an open letter,

warning researchers of the risks of making AI more powerful

without understanding its potential harms.

The signatories include Stuart Russell,

Jaan Tallinn, Stephen Hawking, and Elon Musk.

Nick Bostrom also signed the letter.

He's a Swedish philosopher who's concerned that

a mega-powerful AI that is capable of fulfilling the goals

we give it could cause our extinction.

What if we don't understand the full consequences

of what we're asking it to do?

What if we leave a crucial detail out?

Take the myth of King Midas.

You know, he asked...

everything he touches should be turned into gold,

which sounds like a great idea because you'll be very wealthy

if you can turn coffee mugs into gold.

Then he touches his food, it turns into gold,

he touches his daughter, turns into a gold sculpture,

so not such a cool idea.

Turns out that it's actually quite difficult to write down

some objective functions such that it would actually be good

if that objective function were maximally realized.

It seemed to me that this could be the most important thing

in all of human history.

What happens if AI succeeds at its original ambition?

Which has all along been to achieve full general

intelligence, where we have potentially artificial agents

that can strategize, can deceive, that can form

long goals and find creative ways of achieving them.

And at that point, that kind of AI is not necessarily

best thought of as merely a tool, merely another gadget.

At that point, really talking about creating

another intelligent life form.

So could this potentially be - AI - the last human invention?

Yeah, so once you have general intelligence at the human or

superhuman level, then it's not just that you have made

a breakthrough in AI, but you have made indirectly

a breakthrough in every other area as well.

So the AI can do research, science,

development, all the other things that humans do.

So potentially what you have is a kind of telescoping of

the future, where all those possible technologies that,

you know, maybe we would have developed given 40,000 years

to work on it - you know, space colonization,

cures for aging, all these other things.

So in other words, we could be this hyper-invincible,

space-hopping species with AI,

but we also could go extinct by it?

Yeah.

I think... that machine superintelligence is -

depending on how optimistic you feel on a given day -

either the keyhole through which

Earth-originating intelligent life has to pass,

and we could crash into the wall

instead of actually going through this.

But if you make it through there,

then the life expectancy for human civilization might then

easily measure in billions of years.

BEN: The scientists and philosophers working on

artificial intelligence are sure that AI can improve our lives,

yet the same people worry that it could destroy us.

But how would that happen?

- Hi, I'm Ben. - Hi!

- Nice to meet you, Ben. - Nice to meet you.

What if the AI we build is actually designed to kill?

Heather Roff has testified before the UN as an expert

on autonomous weapons.

I once sat next to a grad student on a plane,

and he told me that he was an AI researcher.

And so I was very interested, saying,

"What is it that you work on?"

And he said, "Well, I work on image recognition."

And I was like, "Really? Tell me more about that."

And he said, "Well, I'm looking at how we identify

different birds, and how we identify

different types of coral and starfish."

And I said, "Who funds your research?"

And he said, "The Office of Naval Research."

And I said, "Do you really think that your research

is going to be confined to birds and starfish?"

And he didn't actually have an answer for me.

He didn't really think through the next step.

How bad would it be if superintelligence is developed

by the military?

So I think if a superintelligence emerges,

we're all in trouble for a variety of reasons.

One of the things that we know about military applications

is that they're not for the benefit of humanity, right?

They're directed towards offensive harming.

We're not talking about creating an AI that's gonna be

trying to solve... climate change

or solve-- or create poetry.

We're not worried about those types of AI applications.

We could be worried about those ones becoming superintelligent,

but you ought to be really concerned about the strong AI

that has guns on it.

What does a superintelligent weapon look like?

A superintelligence could be connected to everything.

If it has a network, it has a capability of being connected

through Wi-Fi, or it could figure out new ways of

connecting itself if you shut off that Wi-Fi.

It could propagate itself and its software

on different servers and different things

so you could never really truly get rid of it.

It could hook itself into missile defense systems

and nuclear arsenals, and it could do whatever it liked.

I mean, that's the whole thing about being a superintelligence,

is you're everywhere.

You're, you know...

Think Skynet but scarier.

(Chuckling)

Wow, can't wait.

(Laughing)

Thanks to all those AI researchers for getting that.

Well, I mean, the AI researchers...

I think it's not fair to blame them.

I think they're trying to do things that are good with AI.

It's the moment someone takes a scalpel from the surgeon

and makes it a knife for killing.

But others think the greatest dangers will come from

unintended consequences, as Jaan Tallinn explained to me.

I don't think it's correct to say that AI is a technology

just like any other technology, or tool like any other tool.

No, it's a technology that can potentially create

new technologies itself.

Now, there are a lot of smart people, as we know,

looking into this problem and this issue, and...

Not enough though.

- Not enough? - No.

Like imagine that we are building a spaceship that's able

to carry the entire humanity.

And like the boarding has already begun,

and children are already onboard.

And then there's like a small group of people who used to be

completely ignored, who are saying that,

"Look, we're gonna need some steering on this ship."

And now people are going like, "Oh, wait a minute.

Yeah, steering might become handy."

(Laughing)

But in this case, steering is, what,

programming the AI to make sure that it doesn't kill us all?

I think the more general point is that whenever you're

a technology developer, you have the responsibility of...

thinking through the consequences of your actions.

♪

It might be an incredibly subtle process which eventually ends up

with the human race becoming sort of enfeebled and dependent

on machines in ways that leave us vulnerable

to any kind of unexpected event.

Is it possible right now we're almost tripping over ourselves,

in that we're coming up with discoveries about AI

and we don't really realize the full implications of

what we'd just discovered, and we just use it?

You know, for centuries or millennia or even--

"It'll never happen," you hear that, you know.

"We don't have to worry, it's just impossible."

The history of nuclear physics, there was a...

a speech by Rutherford.

He's the guy who split the atom.

In 1933, September 11th, he said, "There is no possibility

that we'll ever be able to extract energy from atoms."

Less than 24 hours later, Szilard invented

the neutron-based nuclear chain reaction,

and instantly realized what it would mean in terms of

the ability to create nuclear explosions.

So it went from never to 24 hours.

The scientist who split the atom didn't understand

that his discovery would so quickly lead to the atom bomb.

But the same discovery later gave us the energy that still

provides electricity to millions around the world.

When it comes to AI, it seems like we're about to split

the proverbial atom again.

The future of humanity could be at stake.

And how we build AI now - what we design it to do -

could make the difference.