Fake news: The computers fighting fakes

Computers are taking on the fight against 'fake news' by learning to detect fake videos... even if they are really believable.

Software can now create fake videos that look and sound so realistic people may not actually be able to spot that they aren't the real thing.

They can be created for lots of different reasons, some good and some bad.

They can, for example, help create special effects for movies. But the technology could also be used to create harmful, made-up stories that people might pretend is genuine news.

So to help fight fake news, tech experts are training computers to spot when something just isn't quite right.

Sometimes it might be a harmless joke or to show just how powerful the technology can be.

In the film industry it could be used for special effects and can be cheaper than other techniques.

It could also be used when translating a programme into different languages. Imagine one Newsround presenter speaking English, Welsh, French, Italian, or whatever language you wanted to have your news spoken in.

In the future, this technology could potentially make that happen.

There are lots of people trying to fight fake news and fake videos.

They're training computers and creating software to detect if a video is real or fake - even if you wouldn't be able to tell the difference by just looking at the video yourself.

The computers are trained by looking at lots of real videos and lots of fake videos so that when it comes across as fake, they know it's fake.

The computer first 'learns' how the person's face and mouth move when they say different words.

It then 'learns' how the mouth should move if the person said a new sentence - this can be done by a computer watching a video of an actor speaking that sentence.

It can then merge them together to create a new video.

A simple way would be to get someone who can do different impressions of people's voices and record their voice over the video.

Or, in a similar way to how the computer 'learnt' how a person's mouth moves when they say different words, the computer can 'learn' how a person's voice changes as they make different sounds, and then recreate it.