dhruvrathe

 

Friends this video of Obama which you have seen, this is a fake video!

Can't believe? But the technology has reached a point now where you can use anyone's face and make anyone talk anything.

and such high quality fake videos can be made.

You won't need any animation software for this neither you will have to spend lacs of rupees as done in the movies.

you can do it on your computer just at your home within a few hours.

This technology is being called as deep fake or synthetic video.

Deep fake technology uses neural networks to make such fake videos.

If you have studies artificial intelligence or machine language will know what is this Neural network.

Basically Neural networks are such computer networks whose designs are inspired by our brain's neurons.

In a normal computer function there is an input and an output and there's an intermediate function which transforms input into output.

This intermediate function is defined exactly by a programmer so as to what is has to be.

That means a programmer writes line by line the programs which will transform the input into the required output.

But in the Neural networks this intermediate function is not programmed by any human.

rather this Neural networks itself programs the function what is should be.

This is done by training it through various examples of input and the desired output.

This sounds complicated to hear but when I'll explain it through examples you will understand it better.

Here I've written  S H A R E that is share, these days computers systems are programmed in a way to recognise the printed texts.

so most of the computer systems will be able to recognise it as Share but if the same thing I will write in my own handwriting

then how many computer systems will be able to recognise it?

None of it right?

But a Neural network can even be trained to recognise my handwriting.

This can be done by showing examples for training.

I will write share for 10-12 times and show the Neural network that if I write share like this then it's a share.

and if I write in the other way then too it's  a share.

so with such 10-12 examples the Neural network will learn gradually.

It is obvious that with the more number of examples, Neural network's accuracy will increase more

to recognise that it is Share

So this intermediate function is where the Neural network is itself getting trained and recognising it.

This same Neural network is used in the Snapchat and Instagram filters.

Just like everyone has a different handwriting and a normal computer cannot recognise it

in the same way every person has a different face too.

How does a computer, a phone camera will know that if my mouth is here and the nose is here then another person's mouth and nose will be here too

because we all have different faces.

This Snapchat camera is been trained.

The Neural network of this Snapchat camera has been shown some 50-60 faces marked with pointers to recognise the mouth, the nose of that face.

after pointing again and again on the different faces then the Neural network had learnt

that it's nose is here only so people's nose is always here and mouth is always here.

so we have to place the Snapchat's filter's nose here and mouth here only.

Looksery was a company which was bought by Snapchat in 2015 and this was the expertise of this company

of how to use neural networks in face detection.

Deep fake friends also works in the same format.

To make that fake video of Obama, a lot of his video footage was collected.

to make it learn Obama's body language, his style of talking, the changes in his voice tone as he talks.

On the basis of that, this Neural network made this Obama's fake video.

Since Obama has many videos about his body language so Neural language gave more than one outputs.

Here you can see, 4 different times when Obama is saying the same thing in 4 different body language.

So Neural network can give more than one result too which can be true.

So you can see this technology has a lot of potential and it has a lot of positive uses.

I can think of movie industry which can save a lot of money due to this.

What can you think of? In which all fields these Neural networks can be used.

Let me know in the comments.

Naturally this can also be used in the most dangerous ways especially in our country

where already IT cell, paid media and fake news have created a ruckus.

In our country friends people cant recognise these videos whether they are fake or real how can they recognise such HD videos?

Its  a dangerous thing but you don't have to fear personally that someone can make your such fake videos.

because the initial and necessary requirement is that lots of video footage is necessary to train the neural network.

Someone who needs to fear are the people like me who has a lot of their videos available online

like journalists, celebrities, politicians,

if the availability of the video footage is in the large number than the neural network gets very well trained.

If the training videos are not available then the output of Neural network is somewhat like this.

You can see that the glitches are seen easily and the quality is not so good.

This video of Obama was released in 2017 by the researchers of University of Washington.

and since then a lot of Deep fake videos are available online

If you search deep fake on youtube you will get to see a lot of comedy videos

Like this actor Nicholas cage whose face is replicated at many places for the purpose of comedy.

Lat month friends when I attended Global media forum, a very good talk on deep fake happened there

I would like to show some clips from there

Through the medium of this video, I want to make you all aware of the existence of such technology friends.

So, you share this video with your friends and family and make all of them aware that such technology exists too.

For now, no misuse of this technology  has been done in India.

But only till now! Next year we have elections and these IT cell people can stoop down to any level!

So we need to be vigilant here

For solution, there's no solution available for this thing as it's a new technology.

One solution can be that to fight against the Neural network, we use the Neural network itself.

If a neural network can make fake videos then a neural network can be made which can recognise fake videos.

to recognise such things we can't see from the normal eyes.

This can be possible in future and I hope that this happens in future.

second solution can be the use of block chain to store the original videos

so that we always know what are the original videos and the others will be the fake ones.

If you know about bitcoins that you may know what is blockchains

blockchain is a technology where you can save the history of anything and it can't be tempered.

So it's a very good way to store the history and it can't be changed.

So these are the few solutions which can be thought as of now and if there's anything new I will update you

Even you have to be alert and also keep your friends and family vigilant to keep them safe from the lies.

Share this video with most of the people friends and if you like my work then you can donate on patreon.com/DhruvRathee

One bonus clip of Global Media ForumI would like to share here Friends whose teaser I've shown you in Pratik Sena's video

Many of you have asked me to show the complete video of what question did I asked and what was the answer given by the people.

So here is that clip

here we are talking about artivism that is Art + Activism.

Discussion was basically about how artist are helping in social activism these days.

Let's meet in the next video, thank you.