What is deepfake? How to coexist with deepfakes ! Challenges and Future of Deepfakes

AI can create fictitious videos of people moving and talking as if they were real people.

For example, a deepfake video of Mr. Zuckerberg, co-founder, chairman and CEO of Facebook, speaking recently was uploaded to Instagram and became a hot topic.

In this way, deepfake is a technology that allows you to easily create videos using people without actually using them.

If deepfakes develop, they could revolutionize news, movies, and just about every other industry.

So, this time I would like to introduce you to deep fakes.

What is deepfake

The technology used is GAN

The images used for deepfakes are GANs (Generative Adversarial Networks).

A GAN is a type of generative model that uses two types of AI, a “generator” and a “discriminator”, to generate images.

A generator is an AI that tries to create a fake, and a discriminator is an AI that sees through this fake.

The two AIs repeat the process of making fakes and finding out.

In the beginning, the amount of learning is insufficient, and discriminators can easily spot fakes. However, as the generator learns more and more, the fake becomes more sophisticated, and finally the discriminator becomes difficult to spot the fake.

In this way, GAN increases the accuracy of products.

Deepfakes are such GAN-generated videos.

Can be generated with or without existence

Deepfakes can be generated by real people as well as non-existent people.

Therefore, it can be said that it has a wide range of applications in the field of video.

For example, it may be possible to use deepfakes to accurately generate suitable people as advertising models at low cost.

Deepfakes can be abused

Anyone can easily create a deepfake video using a PC or smartphone. Therefore, many malicious deepfake videos have been created and are regarded as a problem.

This is not to say that deepfake technology itself is bad.

It’s just that deepfakes can be used to disguise the existence of actions or statements that the target person doesn’t actually make.

Therefore, if someone with bad intentions uses deepfake, it is possible to create an adult video that does not actually feature celebrities or celebrities.

In addition, if a government official or a person with political or economic influence is created, the deceived media may spread as fake news, leading to political instability and diplomatic problems.

Deepfake criminality and illegality

The first arrests have been made in connection with deepfakes.

According to the Metropolitan Police Department’s announcement, the suspect has released a video on a pay site that replaces the face of a sexy actress with the face of a female celebrity. According to reports, two suspects were charged with the arrest.

1. defamation
2. Copyright infringement
3. Infringement of privacy/infringement of portrait rights
4. The law has not kept up with technological developments

defamation

In this case, the video is posted on Twitter or paid video sites and can be viewed by an unspecified or large number of people.

As a result, it was determined that the female entertainer, who was the victim, gave an unspecified number of viewers the impression that she was appearing in an adult video, and defamed her reputation.

Copyright infringement

In addition, the copyright of the adult video belongs to the adult video production company that produced it.

One type of copyright is the right to maintain integrity of the work, which is the right not to alter the work against the will of the copyright holder.

In this case, the content was altered without permission from the production company of the adult video, so it can be said that the production company’s right to maintain identity was violated.

Infringement of privacy/infringement of portrait rights

Deepfake is the use of another person’s face in a video without permission, and usually falls under violation of portrait rights.

In addition, infringement of privacy rights is, simply put, the structure that “it is illegal to disclose an individual’s private matter (regardless of whether it is true or not) without permission.” Publishing an event, even falsified, is usually an invasion of privacy.

The law has not kept up with technological developments

However, not all deepfakes are criminal for defamation or copyright infringement.

Basically, the current law does not assume that advances in AI technology will enable the creation of advanced deepfake videos.

Therefore, under current law, there are some crimes that cannot be held criminally responsible.

Like the video that led to the arrest this time, if it is made available to unspecified or large numbers of people, and if the synthesized video has content that generally lowers the image of celebrities, it will be charged with defamation. There is a possibility.

For this reason, at this time, if many entertainers are victimized, it is likely that they will claim that the offense of defamation will be established.

5 Deepfake Cases

President Obama speech video

In April 2018, a fictitious speech video of former President Obama was posted on Facebook.

The video was courtesy of American news outlet BuzFeed and film director Jordan Heal.

In the video, former President Barack Obama lashes out at the Trump administration.

David Beckham’s anti-malaria video

RG/A, Ridley Scott Associates, and Synthesia jointly created a malaria prevention video for the NPO Malaria No More.

This campaign has already exceeded 400 million impressions around the world and has received a great response.

In this way, by expanding the use of celebrities and deepfakes, it is possible to create campaign videos that have an impact on society.

Elon Musk childhood video

This is a fake video made by predicting the face of American businessman Elon Musk in his childhood from his current age.

Although it is still an eerie image, it is reproduced quite realistically.

If the accuracy improves, it may be possible to reproduce the childhood of historical figures whose only adult photographs remain.

Generate a full-body movie of a fictional character! data grid

A high-resolution, high-quality fictitious full-body model image can be generated from a large amount of full-body model image training data.

It is expected to be used as a virtual model for advertising and apparel EC.

In the future, we will continue to further improve the accuracy of the full-body model automatic generation AI and research and develop motion generation AI.

AI makes even amateur drawings realistic! Nvidia’s GAU GAN

Quote: https://www.nvidia.com/en-us/research/ai-playground/

It is a deepfake that converts even a picture drawn by a beginner into a realistic landscape image.

It’s amazing how even lines that look like scribbles can look professional.

The creation of landscape images, which used to be left to professionals, may now be possible for anyone.

Profitable uses of deepfakes

I think deepfakes have gotten a bad impression recently, probably because of the frequent news reports that they have been exploited.

But deepfakes are, in fact, a highly profitable technology if used correctly.

For example, when making a movie, deepfakes can be used to create more elaborate videos.

In addition, the production that made Hibari Misora ​​appear on the stage as if she had come back to life became a hot topic.

In this way, deep fake technology makes it possible to bring the deceased person back to life.

Deepfake technology can be used in ways that have never been done before, but also in ways that have never been done before.

How to coexist with deepfakes

These days, anyone can create a deepfake image. Of course, at present, only specialists can make very sophisticated objects, but I think we will eventually be able to do it.

Therefore, we have to make it possible to choose whether information is true or false, and coexist with deepfakes.

That is why DARPA (the US Defense Advanced Research Projects Agency) is heavily funding deepfake detection.

In addition, twitter has made efforts to have users write notes, and Facebook has launched a deepfake detection technology competition called “Deepfake Detection Challenge” with more than 10 million dollars. teams are participating.

In today’s world where information is highly valued, the coexistence of deepfakes is inevitable.

It seems that there are still many challenges, but many companies are doing their best. We may be able to coexist with deepfakes in the near future.

How to make a deepfake

Earlier, we talked about what deepfakes are.

So, here, I would like to introduce how to create a deepfake.

Deepfakes are created using a method called GAN (Generative Adversarial Networks).

GANs (Generative Adversarial Networks) are networks that create fakes and determine if they are fakes.

To explain in more detail, GAN consists of two network structures: Generator and Discriminator.

Generator creates fakes with quality close to the real thing.

The Discriminator is a Generator that discriminates whether or not the created fake is fake.

Through these two processes, we create a deepfake that is as close to the real thing as possible.

As a result, even to the untrained eye, it becomes a deepfake that is indistinguishable from the real thing.

An app for creating deepfake videos

Here are four apps for creating deepfake videos.

1. Turn your face into a movie star! “Zao” (for smartphones)
2. You can make videos with other people’s faces! xpression (for smartphones)
3. Deepfakes Web Beta (for PC)
4. FakeApp (for PC)

I will explain each.

Turn your face into a movie star! “Zao” (for smartphones)

Zao is an app that allows you to easily create a deepfake with a single face photo.

It gained popularity by being able to create videos that fit any face to the video of movies and TV programs, and won the first place in the popularity ranking of free apps in the iOS App Store for China.

You can make videos with other people’s faces! xpression (for smartphones)

xpression is an application that uses AI to replace the face of a person in a video with your own face in real time.

For example, if you use a presidential speech video on YouTube, you can easily create a very realistic video that allows you to pretend to be the president and give a speech freely.

Deepfakes Web Beta (for PC)

Deepfakes web beta is an application that allows you to easily create deepfake videos by simply uploading an image or video that substitutes for your face.

It costs $2 per hour to create, but you can create deepfake videos with high accuracy.

It is also attractive that you can create high-quality deepfake videos regardless of the specifications of your PC.

FakeApp (for PC)

FakeApp is a free PC application.

Japanese is not supported, but there is an English explanation video, so you can understand the procedure without any problems.

Challenges and Future of Deepfakes

This time, I introduced how deepfake works, how to make it, and the problems when it is abused.

I hope you have understood that while it is a technology that expands the possibilities of video content production, it also has the risk of developing into a major problem if it is misused.

In order to prevent deepfakes from being exploited, it is important for each and every one of us to accumulate knowledge about AI and take appropriate measures.

It is also important for individuals to have an awareness not to easily spread content created by deepfakes.

Now that AI technology continues to evolve, why not take measures to protect yourself from malicious crimes and use AI correctly?