AI Can Create Interesting and Fake Faces

Welcome to the 21st century, where AI generated faces are indistinguishable from real ones. In simpler terms, these fake faces are so realistic that the average person cannot pick them out in a sea of faces.

How did we get here? Should we be scared? This article answers questions like these and others as we take a deep dive into the world of AI generated faces and images.

The media world is about to get its share of the immense possibilities of artificial intelligence. Content writers have been able to make their work smoother using new AI writing tools like INK and Jasper. It is now the turn of photographers and artists to use AI to their benefit.

How Are AI Generated Faces Created?

Generative Adversarial Networks (GANs) are the key to creating fake faces that are like real human faces. This machine learning framework was developed by Ian Goodfellow and his colleagues in 2014 

GANs comprise two neural components – the generator and the discriminator. These two are adversaries because the generator creates fake photos for the discriminator to assess and point out as fake or otherwise. 

The more the discriminator picks out fake pictures, the better the generator gets at creating the right face. The two neural networks continue this back and forth (training) until the generator creates a face that the discriminator accepts as real.

It is a zero-sum game where one party’s gain is the other party’s loss. In essence, the discriminator will train the generator to up its game till it can beat its detection system.

How Lifelike Are AI-Created Faces?

The use of AI-generated images and faces is more common than you think they are. These faces are so realistic that they are hard to point out. In a study, researchers found out that people found images of fake people made with AI trustworthy. The result was unexpected. One of the researchers noted that they initially felt that the synthetic faces would be less trustworthy than the real faces. Unfortunately, they were wrong.

The truth is that images generated with AI tools have surpassed expectations, and it is becoming harder to spot the use of this technology.

For instance, AI-generated art is hard to spot and is becoming more common every day. However, some AI-generated faces can be spotted on a closer examination like this one:

At first look, the two images are perfectly normal. Observe them a second time, and you’ll find some subtle signs that something is amiss in the second photo.

• There’s a white patch on the ear lobe
• The lenses are mismatched
• The background looks blurry

These are a few features that give out the second picture as fake. However, these features may not be present in other pictures. Other things you can look out for include: mismatched pieces of jewelry like earrings; asymmetric features may be exaggerated. Also, the end pieces of glasses may not match, and bright colors may bleed into the hair and skin from the background.

Conclusion

It is fair to say that faces created with artificial intelligence resemble real people. There are deep fake videos on the internet that show how much a human does resemble AI-created faces.

The last few decades have heralded great development in the field of artificial intelligence. It has disrupted several sectors, and the creation of faces of non-existent people is the next big thing. If Andrew Ng’s opinion about AI is anything to go by, the current advances are just the beginning.