The most common 'deepfakes' and how they can be detected

The Austrian Institute of Technology (AIT) is developing a new tool to detect deepfakes - while also raising awareness of the problems they can cause. These manipulated images and videos can be very dangerous to public security, according to project coordinator Martin Boyer from AIT, which cooperates with the Austrian government. 

The Viennese AI expert says the software tool he is developing is aimed at helping journalists know whether pictures or videos are authentic.

What types of deepfakes are there?

A common example of a deepfake video is 'face swap' in which the face of the victim is placed on a different body. This could create a scene which is embarrassing or even illegal. 

Another deepfake is known as 'face puppetry' in which the victim's lip movements are manipulated to literally put words into their mouths. To create a convincing deepfake of this kind, roughly 20 minutes of video footage are needed. 

Experts also consider AI-generated still images as deepfakes. Several websites will generate fake images by utilizing images available on the internet adapted to instructions put in by the creator.

‌

Is it possible to tell the difference?

No. According to Boyer, the naked eye can't tell the difference when a deepfake is well done. However, deepfake detection tools already exist which can show the likelihood of manipulation by checking various parameters, such as the image metadata. 

But Boyer says relying on detection tools isn't always enough. His team at AIT is not only creating a new webpage where users can upload content and get indications whether it has been manipulated, but also raises awareness by approaching businesses, senior citizens, and children and spreading knowledge about the dangers.

Why are deepfakes dangerous?

Creating such images can change and manipulate public opinion, which is often formed moments after the fake content is published. The consequences for a country are wide-ranging. Spreading disinformation or promotion of extremist groups could sway voters that would impact the outcome of an election. 

Politicians and voters are not the only potential victims. Criminals target CEOs or industry leaders by faking their identity, with the aim of convincing them or their associates to transfer money. A victim could for example receive a voice note by someone faking the partner's voice claiming to have been in an accident or a situation that required sending funds immediately.

READ MORE: Should I Worry About... deepfakes?

Subscribe to Storyboard: A weekly newsletter bringing you the best of CGTN every Friday