Digital Avatars and the Future of Fake News

(Hail to the Chief)

My fellow Americans, I'm sorry.

I had no idea it would be this tough.

Nobody told me it would be this difficult

to be presidenting.

No one knew, no one had any idea...

[Jack Stewart] Okay, back up. (tape rewinding)

Something is obviously not right here.

This isn't really Donald Trump.

It's a sort of digital puppet

made by a Los Angeles startup called Pinscreen.

Right now you can Photoshop anyone

into any picture.

People will be able to do fake videos as well.

[Jack Stewart] With Pinscreen's technology, we created

a digital mashup of the real Donald Trump

with audio from Trump impersonator John Di Domenico.

Pinscreen calls this face retargeting.

It's one of the most sophisticated of a handful of

avatar-creation techniques that could hit the market soon.

In the not so distant future we'll all be able

to make digital versions of ourselves, of someone else,

or even create videos of things that never happened.

I think the ability to create

new contents give people the ability

to manipulate and create information that isn't real.

So that's something that we have to be careful about.

[Female Reporter] Because of fake news.

[Male Reporter] fake news.

[Female Reporter] Pizzagate.

Alternative facts.

[Jack Stewart] If you thought the problem of

false stories circulating on social media was bad enough,

the next generation of fake videos may be even

harder to spot.

You can take a celebrity,

you can take a politician and you can take his picture

and recreate his digital avatar.

[Jack Stewart] This sort of face-swapping that

Pinscreen did with Trump, isn't quite ready for consumers.

But Li's team has created a CG rendering program,

using neural networks

that can produce 3D avatars in seconds.

His research also contributed to the Animoji app

on the new iPhone 10.

Their goal is to develop

easy-to-use facial mapping and rendering technology

that could change the way we communicate online.

If you think of computer games or movies,

you see a lot of digital doubles of humans.

And what we're trying to do is,

we're trying to make this process really easy

for anyone to create a 3D character of yourself.

Oh no, here they come!

To find out how they do it, we visited their L.A. office.

What we did was email over

a couple of pictures to you, of me.

[Hao Li] Yeah.

And you've built them out into one of these avatars.

Basically, a virtual me.

Correct, if I take a picture of you, everything is 2D.

And part of your face is not visible.

So for instance if your face is slightly tilted

I won't be able to see what is on the other side.

So what the algorithm does is retrieve from

a huge database, what your most likely

complete face model is.

However, in our database, we for instance

never have your face texture, what we'll do

is we'll try to recombine from existing people

what is the most likely combination of a

complete face to reproduce what you would look like.

t really looks like me.

Particularly the bottom of my face here is,

I think really realistic,

the way the beard is rendered is really good.

In this picture, the one that we provided to you

I can see that it looks like my hairline is sort of

receding quite a lot at the sides there.

It looks like I have less hair and that's

what you've reproduced very faithfully.

[Hao Li] Right.

[Jack Stewart] My avatar looks a lot

like me but isn't an exact doppelganger.

It's more like something you'd see in a video game.

And there's a good reason for that.

The problem is that if you attempt

to do something that's fully real and

you can't 100% get there, if you're only at 95%

then it will be very off to some people.

So people will feel that the face doesn't look very natural.

Then you end up in this uncanny valley.

So what we do is we try to deliberately make it

slightly stylized, slightly animation-like...

[Jack Stewart] You're changing the person I am.

Stylized avatars, Li believes, will become the way

we choose to communicate in virtual and augmented reality.

While the tech still has a way to go,

it's improving day by day.

Just look at my avatar from a month ago

and now the very latest rendering the company produced.

Just imagine what would be possible in a few years.

We are already getting to the point where

technology can bridge the uncanny valley for one

aspect of human communication, speech.

[Male Voice] I kiss my dogs and my wife ...

[Jack Stewart] Last year Adobe which makes Photoshop

among other creative editing tools showed off a

prototype version of what it calls VoCo,

a sort of Photoshop for the voice.

[Male Presenter] And here we go.

[Male Voice] And I kissed Jordan and my dogs.

(audience laughter)

Well, you're a witch. (woman laughing)

I have decided--

[Jack Stewart] We had to hire an impersonator

for the Trump experiment, but new tools like this

could make it even easier to make avatars

say things that they never did.

But hopefully not one of the scary ones.

Couple it with advanced and inexpensive 3D avatar modeling,

like the kind Pinscreen is working on,

and it won't just be Hollywood studios

that can make convincing duplications of real people.

The implications are both thrilling and terrifying.

The important thing is that we educate

people that this is possible.

[Jack Stewart ] Li thinks we'll have to come up

with new ways of authenticating video

and even watermarking digital manipulations.

But as we've seen, it's challenging enough to

flag the fake news articles that go around on social media.

So get ready because you may be seeing

a lot more clips like this.

I love the media.

Especially Wired Magazine, the opposite of fake news.

(upbeat electronic music)