Deep fake is an artificial intelligence that creates convincing videos, images, and audio hoaxes. This term is used to describe the content created and the technology employed, hence the term deep learning and fake i.e. deepfake. Deepfakes are synthetic media that have a video or image of a person being replaced by someone with the same likeness.

Although the creation of fake videos is not relatively new, deepfake employs a powerful tech from artificial intelligence and Machine learning to create audio or visual content that can deceive others. Some may argue that deep fake technology is a threat to society, others believe it has positive impacts when used in the right way.

The truth is that technology would always be misused, no matter how ethical the rules governing it should be. So we can expect that in time this Artificial intelligence would only have positive applications alone.

How Does Deep Fake Technology Work?

This face swap artificial intelligence takes a few methods to accomplish. Here are some of these methods and how they work:

Auto-encoders

These are neural networks that can copy their input. Auto-encoders compress data based on what they are trained to do. Auto-encoders comprise three elements that include; an encoder, a code, and a decoder.

You can run several face shots of two different people through this algorithm called the encoder. The encoder would then learn the similarities between the faces and compress them based on their common features.

The next algorithm, decoder, would recover the compressed images. Since the images are not the same, you would have to use two decoders when recovering the faces from the compressed images. The last step in completing the deep fake is to take the compressed image of A onto the one of B. The decoder would then reconstruct B into the orientations and expressions of A. For every video, this procedure is to be followed through.

Generative Adversarial Network (GAN)

Another method that can be used to create deepfake is called generative adversarial network. A GAN puts two different neural networks against themselves. These networks are called the generator and discriminator. The generator is an algorithm that turns random noises into images. The image created is added to other real life faces using the discriminator. At first instance, the generated images would not resemble the real photos. So the generator would make sure that it deceives the discriminator. The process would be repeated till the discriminator recognizes the generated images as authentic and real. With enough training and resources, GAN's would produce more realistic photos. This method is mostly employed in generating photos than creating videos.

Positive Applications of Deep Fake Technology

Some of the positive impacts of this tech include the following:

Art Industry

Deepfake Artificial intelligence can be used in making famous portraits to talk. Russian researchers created something like that by using the Da Vinci popular portrait 'Mona Lisa'. Other museums have also done a similar recreation using some of their artwork portraits to attract visitors.

Movie Industry

In movies, it is necessary to change the face of some actors. For instance, in the movie Rouge One, this AI was used in creating the face of Grand Molf Tarkin and Princess Leia. Currently, the Disney studios are working on their deepfake effects artificial intelligence. This is projected to reduce the financial costs and time spent in hiring an actor or actress. Also, in cases where an actor dies, this face swap AI can be used to recreate the faces so that they would remain in the movie.

Negative Effects of Deep Fake Technology

The negative effects of this include:

Blackmail

This Artificial intelligence can be a sinister tool to those who want to denigrate or blackmail others. Fake audios, pornography, and other videos have been created by these fraudsters to blackmail and scam others. With no credible evidence of what is true or not, innocent people and celebrities have been on the receiving ends of such scams.

Politics

Deepfake Artificial intelligence has been grossly used in misrepresenting well-known politicians. For instance: In April 2018, comedian Jordan Peele worked with Buzzfeed to make the deepfake of Barack Obama using Peele's voice. In 2018, various videos of politician’s face swaps were made. Some included the Argentine President Macri Mauricio who was replaced with Adolf Hitler. Then there was Angela Merkel, Germany’s Prime Minister who was swapped with Donald Trump. In 2020, a Belgian Rebellion group published a deepfake video of the Belgian Prime Minister on Facebook. Many who viewed this felt it was authentic and condemned the Prime minister for what he didn’t do.

How Can You Spot a Deep Fake Artificial Intelligence?

As the technology improves, it is getting harder to detect a deepfake. Before, some researchers believed that the faces in deep fake do not blink. No sooner as the research went viral that these swapped faces began to blink and make other facial movements.

However, poorly made deepfakes are quite easy to spot. Here are some key things that you'd notice:

• The skin tone would be patchy and the lip sync would be so bad.
• There could also be flickering on the edges of the compressed faces.
• The hair strands are also visible on its fringes.
• Pieces of jewelry and teeth are also big giveaways.

Can Artificial Intelligence Help Tackle This?

Artificial intelligence can help to detect fake videos and flag them up. There could be an online blockchain that holds original records of videos, pictures, and audio so that their manipulations can be countered. Biometrics, an AI facial recognition algorithm can also help to identify the original faces of individuals in the deep fake.

Is Deepfake a Threat or Revolutionary?

Some are helpful, while others are a big threat to society. When used ethically, this Artificial intelligence can be a promising and revolutionary technology to the world.

Summary

The deep fake AI is in its early stage and yet is very promising. It has its pros and cons, like other technologies. Over time, we would have many ways to curb the negative effects it can cause and promote its useful applications in various fields.