Facebook and Twitter, YouTube has banned deepfakes.
Facebook and Twitter, YouTube has banned deepfakes.

The devil’s in the detail in times of deepfake attacks

  • Facebook and Twitter, YouTube has banned deepfakes
  • Deepfakes are synthetic videos generated using artificial intelligence (AI) algorithms called generative adversarial networks (GANs)

Over the past month or so, Facebook and Twitter banned something called deepfakes. As the US presidential election slated later this year picks up pace, tech giants are increasing their efforts to manage the spread of misinformation on the internet. Mint takes a look.

How is a deepfake video created?

Deepfakes are synthetic videos generated using artificial intelligence (AI) algorithms called generative adversarial networks (GANs). Using these one can take a video of a person who resembles a celebrity, politician, etc. and make it seem that the celebrity is saying the words in the video. GANs have two neural networks—a generator and a discriminator. The generator creates a fake, while the job of a discriminator is to catch that fake. The generator continues doing so till it can beat the discriminator, which then becomes the output of the algorithm. These are also used to create artificial humans for movies, advertising, etc.

Graphic by Paras Jain/Mint
Graphic by Paras Jain/Mint

Have I ever seen a deepfake video?

Remember the viral video of Barack Obama calling US President Donald Trump a “complete dipshit"? That was a deepfake created by BuzzFeed and comedian Jordan Peele to demonstrate the concept. Artists Bill Posters and Daniel Howe created a video of Mark Zuckerberg bragging about controlling “billions of people’s stolen data", which also went viral. Indian actors Priyanka Chopra and Deepika Padukone were victims of deepfake porn, when fake videos of them showed up on Pornhub, one of the largest porn websites in the world. Journalist Rana Ayyub was also a victim of such an attack.

What action have major social platforms taken?

Besides Facebook and Twitter, YouTube has banned deepfakes. However, the social platforms put deepfakes under the overarching term “manipulated media", so as to cover all forms of manipulated content. So their bans include anything from photoshopped images to advanced AI-generated videos. Biometric morphing is a version of deepfake used for images.

Is there a way to spot deepfake content?

Deepfakes first appeared in 2017 and have improved a lot, making them difficult to catch. Experts say one should focus on a subject’s eyes, on signs such as blinking and the rate of blinking, which is less in such videos. You should look for blurring at some points of the face, mainly near the nose and lips. A mismatch between expressions and emotions against what is being said is also a giveaway. Ironically, the best way to catch a deepfake is through AI algorithms. That’s how Twitter, Facebook, YouTube, etc. are planning to tackle the menace.

Are there positive uses of this technology?

Yes. As with anything tech, there is a light side to deepfakes, too. Samsung’s Neon and US-based Soul Machines used GANs to create artificial humans, which can be used for advertising, movies and more. The underlying technology can be used for creating artificial voices for people who have lost theirs. They can also be used to make AI voice assistants like Amazon’s Alexa more lifelike and relatable. Researchers from universities around the world have been working on finding the benefits of this technology.

My Reads Logout