Skip to main content

6 deepfake questions (with answers and explanations)

 

Deepfakes


The internet is flooding with new deep fake videos coming out every single day

you must be having some questions regarding them in your mind 

I will help you out 

here's a summary of this article 

What are deepfake videos?

a deep ake video is an altered video which appears real 

How deepfake works?

it works using 2 AIs one of them makes the video frames while the other one checks how real they look

why is deepfake dangerous?

they are dangerous they appear very real, so they can misguide roughly anyone

how is deepfake made? 

a face is applied on a pre-recorded video, by using computer programs they are then adjusted further 

Can deepfake be detected?

yes about 96% of them can be detected by some softwares, and a lot of research is going on to improve the detection process 

will deepfakes technology destroy democracy?

Not exactly, but it will become difficult for the people to know the true facts regarding elections 

What are deepfake videos?

A deep fake video is a fake video featuring a person (mostly a celebrity or a famous personality) who never did or said as shown in the video

There was a deep fake video of former American president Barak Obama; it appeared so real that very few people were able to tell the difference

The main reason for deep fake videos appearing so real are that, a lot of research is done before making such a video, they are not like regular photoshopped images

in regular photoshopped images simply a face is placed on someone else’s body, which is highly likely to  be spotted, while these fake videos are made after a lot of analysis about the person, who will be featured in the video

KEY POINT- deep fakes are seemingly very real fake videos.

How deep fake works

Deepfake works using 2 AIs (artificial intelligence) one of them makes, deep fake video frames, while other checks how well they match with the real person who is being featured in the video.

Here’s the complete process for the working of a deep fake

1-     An algorithm is trained with roughly 5000 images of the person being featured

2-     The algorithm then understands the facial expressions, lip movement, and many other details

3-     One AI helps to make video frames based on its understanding

4-     The second AI now checks these frames generated by the first AI

5-     If the second AI can find differences from the original picture, that particular frame is rejected

6-     The rejected frames are improved further, until they pass the second AI test

7-     This process continues until the best frames are selected

8-     These frames are now applied on a pre-recorded video

Such a technology which uses 2 AIs is known as GAN (generative adversarial network)

It is clear from its name itself that,

·        one of the algorithms generates something

·        While the other one finds flaws in it

KEY POINT- a combination of 2 AIs makes these videos; one generates video frames and another check on how real it looks.

Why is deepfake dangerous

Deep fakes are dangerous because it makes we question the saying “seeing is believing”

With the advancements in technology, deep fakes are getting harder and harder to be distinguished from the real video

Here are some of the reasons why deepfakes are dangerous

Readily available data-

·        A lot of data about almost any person is available on social media nowadays

·        This data makes it easier to create deep fakes of many people

·        As the algorithms are improving they may even understand a lot from fewer images only

·        From even a single video a lot of images can be extracted to teach the AI

·        As the trend of posting short videos is increasing, indirectly creation of such fakes is even easier

 

 Unlock face-locks-

·        it would be very easy to unlock a face recognition using deepfakes

·        most of the phones don’t have a hardware-based facial unlock system, so they can easily be unlocked with even, fake images

·        with the help of video frames, a lot of images from various angles can be extracted

·        these images can then be used to make a 3D model of face

·        so even if a phone has a hardware-based 3D scan face lock can be unlocked

 

Videos can become useless proofs-

·        fake videos can divert investigation of a crime

·        CCTV footages have helped so far to solve a lot of cases

·        But with the advancements in fake video technology, these video proofs may become useless

·        Because anyone would be able to make a fake video in his/her favor

·        Either a parallel technology has to be developed, to distinguish these fakes from real ones

·        Or a video proof will no longer be proof in the court

Influence politics-

·        These fake videos may also end a healthy political competition

·        Any political party can make these videos against a politician(s)

·        These videos may affect their public image

·        At the same time, these videos can be used to build a good image of someone

·        With a good image of a particular leader and a bad one of the others may lead to a continuous win for a particular leader

·        This way the very spirit of democracy may dissolve  

Manipulated social media posts-

·        A lot of fake information keeps on spreading through social media

·        With the development of this fake video technology, the posts can be manipulated to influence people

·        There is almost no crosschecking for the the fakeness of a video, even on leading social media platforms

·        To protect people from the effect of these fake videos, social media platforms such as Facebook is deploying a dedicated team of engineers to check the fakeness of videos posted

FACT-Face book has spent 10 million dollars to detect these fake videos

Live deepfakes-

·        A lot of research is going on to create life fake videos

·        These live fakes can make it possible to manipulate people in real-time

·        Authority can be impersonated to manipulate people    

KEY POINT- these videos look so real, so they can misguide people

How deep fake is made-

Here is a basic outline of the procedure that is followed in making a deepfake video

·        A lot of data about the person being featured in the video is gathered

·        If a celebrity fake video is to be made, this data can also be gathered from their interviews and social media

·        New setup and dialogs are now chosen, and a a person is made to act, as per video requirement

·        The actor must very close facial features as the celebrity, and try to act the celebrity as closely as possible

·        Now these 2 data sets are entered into a deepfake software, i.e. celebrity’s data, and the actor’s video

·        This software has 2 algorithms, one will generate fake images based on celebrity data, and the other one checks how closely they resemble with the celebrity

·        images that didn’t pass the 2nd algorithm’s test are improved further

·        this process continues until they don’t pass the 2nd algorithm’s test

·        this way a nearly flawless and indistinguishable fake video is made  

This is just a broader outline of the process

To create a near-perfect fake video a lot of other skills are required, they are-

·        computer coding knowledge

·        acting skills

·        video editing skills

·        photo editing skills

FACT-in fact, one of the deepfakes of President Donald Trump took 4 days to complete

This procedure is defiantly not enjoyable and fun          

But

With improving technology, very powerful software is available for free, which can create these videos, in very less time and from a normal computer

If you want to create a high-quality fake, in 10 minutes, all you need is

·        a Google account (creating a new account is preferred)

·        1 png and 1 mp4 file of a celebrity you want to fake

·        One of the videos where you would like that celebrity to be (square aspect ratio is preferred)

·        Follow the instructions in this 6 minutes video

this is what he is talking about in the video

1# this is the link 2# !pip install PyYAML==5.3.1 3# first-order-motion-model 4# !ffmpeg -i /content/gdrive/My\ Drive/first-order-motion-model/*******.mp4 -s 256x256 -c:a copy result.mp4 5# /content/first-order-model/cropped.mp4

Can deepfake be detected-

A lot of research is being made to detect fake videos; companies like Facebook, Microsoft, and Google are coming forward for the development of detection software for these videos.

Spotting fake videos are getting harder and harder, with the developing technology

And such videos are also spreading at a fast pace, especially on social media

To detect a perfect fake video, some software is required

But

To detect a not so good fake video, all you need is common sense

Here the points that you must keep in mind detect deepfake videos

·        Look for distorted facial features, especially the eyes, eyebrows, and lips

·        Look for misplaced shadows, see if shadows are where they should be or not

·        Listen, whatever is said should be said or not

·        Check for the source of the video, if it is from an untrusted source it’s likely to be fake

For detecting near-perfect fakes, the software is being made; they mostly look for the clues, which machine learning currently cannot analyze

Some of these are

·        Slow the video down, and look for glitches

·        How do people move

·        How expressions change

·        Lip movements

·        Eye movements

The basic principle behind detecting deepfakes is training AI in a reverse way, meaning training it to find fake videos rather than creating others

This is a kind of race between creation and detection of deep fakes

The AI for detection of the videos works like this-

1-     Different lines and dots on the face are drawn by the AI

2-     These lines and dots helps to analyze the

a)    Movement of the head

b)    Different expressions

c)    The direction of the eyes

3-     Real and the fake videos are now compressed

4-     Detection AI now finds anomalies in the video

5-     The video is now given a score on the fakeness scale

6-     The more the score on the fake side, the more the fakeness of video

7-     This method of detection is highly successful till now and can differentiate up to 96% of fakes

This is a broader outline of the process involved for deepfake detection

But

Still, there is are 4% videos which cannot be detected by this method

For which, Research is continuously going to train the detection AI more and more

KEY POINT- 96% of these videos can be detected.

Will deepfake technology destroy democracy

Possibly Deepfakes cannot destroy democracy, because this technology is available to every political party, thus creating a possibility of deepfake competition amongst political parties

·        If one party creates a fake video

·        Another party may counter it by making another fake video or providing other evidence against the video

Here are the main ways how deep fake technology can affect democracy

·        Destroying the image of a leader amongst voters 

·        Spreading fake news

·        Influence voters against a political party/leader

·        Manipulate voters to vote for a specific party/leader

·        Continuous and collective manipulation can lead to regulars wins of a particular leader  

These videos do not seem a threat to democracy unless used all political parties target them against a particular leader or party

If everyone has equal access to advanced fake video creation technology, they will use it in their favor, thus creating a completion

However, it would be difficult for the voters to know the actual truth

Example- conceder 2 political parties A and B

1.     party A makes a fake video against a politician of party  B

2.     party B can either

a)    defend by proving the fakeness of video

b)    make another fake video against someone from part B

3.     if party B used the second option, damage caused to them may be neutralized

4.     But, voters won’t be able to know the actual truth and a chain of fake videos may build-up, with each party defending itself by fake videos.

 KEY POINT- deepfakes may be a lesser threat to democracy than to truth.

Conclusion-

I hope you would have learned, exactly what are deepfake videos

If you have any confusion or you want to add anything

Head over to the comments, and let me know!


Labels:

Comments

Post a Comment

Popular posts from this blog

quantum computers explained simply (complete guide)

   Quantum computers, this a term which you probably have heard at some point, even if you are not much into technology this computing is probably the future of mankind In this article, we will be covering the following topics Do quantum computers exist? yes they exist, but they are not very powerful  Quantum computers are now a thing of reality, but the existing quantum computers are at the very initial phase of development, the current quantum computers are the ones born right after theory  Leading technology companies such as IBM and Google have been constantly researching on quantum computing and now they have successfully developed quantum computers which are able to solve certain problems At the end of this article we will also link you to the IBM’s quantum com[puter experience, so you yourself can actually experience the technology of future There is even a company that sells quantum computers, called D-Wave  D-Wave  quantum computers  a compute...
Post a Comment
Read more

transistor

Transistors are present almost every electronic equipment that you come across in your day to day life There are nearly 100 million transistors in a mobile phone and nearly over 1 billion transistors are present on a laptop or a personal computer in this article, we will be dealing with the following topics what is transistor science of semiconductors n-type doping p-type doping working of a transistor transistor as a switch transistor as an amplifier  What is transistor Transistors in its simplest form can be thought off as a switch with special features The mechanical switch can control the flow of electric current and can be turned on and off Likewise, the same thing can be done with a transistor Unlike a mechanical switch, a transistor does not have any moving parts, does not need any human operator and can operate at a very fast rate Transistors are do all this because of the advancements in the science of semiconductors Science o...
Post a Comment
Read more

Linux (everything)

In this article, you will get to know what is Linux , what are its applications, and 10 ways in which  Linux  is better than its competitors. If you have been wanting an operating system with a lot of stability and performance and privacy, the answer to all these questions is Linux based operating systems So let’s start by understanding what exactly is Linux What is Linux What is   Linux ? ,  Linux  is a family of open-source operating systems based on  Linux  kernel, such as Ubuntu, android,  Kali Linux , etc. A kernel is a part of an operating system which can control anything in the system, kernel helps in providing a link between hardware and software A kernel can control how much ram can an application use, how what permissions are given to an application and everything an application wants from the hardware is in the control of the kernel. For example- a camera application wants permission to access the...
Post a Comment
Read more