AI can now create life-like human faces from scratch

Artificial intelligence is now capable of of creating images of human faces that are impossible to differentiate from real photos, new research has shown. 

Fast-paced developments in AI over the last four years have enabled computers to generate false images that are impossible to tell apart from reality.

This technology, called Generative Adversarial Networks (GANs), is behind the wave of digitally manipulated videos and photos called "deep fakes".

The results are that famous people such as actors or politicians can be made to say and do whatever creators want and are difficult to prove as fake. 

GAN technology was modified by AI company Nvidia to create a high quality database of  thousands of pictures of human beings, all generated by a computer. 

Researchers trained their computer with 70,000 photos of real people from Flickr, which included different ages, ethnicity and image backgrounds. 

Using these images as a base, the computer was able to learn and segment aspects of different people - such as hair colour, face shape, or skin colour -  and generate completely new images. 

Researchers have argued that these photos are better than those generated by current GANs. 

The technology is also able to detect and reproduce accessories like glasses, sunglasses or hats and create an infinite number of images of completely new people, including children. 

As GANs become more sophisticated, senior US and UK politicians have expressed alarm at the threat posed by deep fakes to spread misinformation or stir up conflict. 

Scientists are battling to create new algorithms to detect video and photo forgeries as the technology becomes too sophisticated for the human eye.