DeepFake can impact Humanity, more than you can imagine
Technology has caused great changes in the way we experience and consume media, and it continues to do so! Recently, we invited Sezer Karaoglu, CTO and Co-Founder of 3DUniversum, in our RTL Cinema to tell our RTL-colleagues all about the fascinating technology ‘DeepFake’. Here is a summary of the session.
Deepfake: A new way of creating content
According to the famous Dutch dictionary Van Dale, the definition of DeepFake is:
‘digitally manipulated photos or videos of a person doing and saying things that he/she did not actually say’.
It is considered to be part of ‘synthetic media’, an all-encompassing term for the artificial creation or modification of media by “machines”.
Deepfakes (or synthetic media) has been applied in the movie industry for years. Back in the days, you needed a lot of resources (e.g. time, money, and skills) to create synthetic media, making it mostly available for Hollywood and not for individual users. However, as time passed, even consumers started getting access to this technology on their smart devices via various apps, enabling them to alter images of themselves (e.g. swapping their gender on Snapchat). With today’s technology it is possible to turn a single image into a talking and moving image.
Imagine what all could be created if multiple videos of a person are available.
How to create and recognize a DeepFake?
Artificial Intelligence and CGI (computer-generated imagery) technology play a large role in creating DeepFakes. If you take a picture, the picture compresses a lot in information from a scene into an image (e.g. geometry, lighting, reflectances). Artificial Intelligence (computer vision) can extract this information from an image. Then, CGI technology can create a perfect environment where you have all the extracted components. It is then possible to manipulate these components in order to re-render a scene. This process is called ‘neural re-rendering’.
It is important to understand how DeepFake is created, know where it fails, in order to recognize DeepFakes.
There are primarily two ways a DeepFake is used:
- Replace/Change existing images: uses of this type of DeepFake-creation are FaceSwap and Face Synch. With a FaceSwap, an image of a person is replaced with another image. AI is used to detect the existing face, extract key points of the face in order to relocate and place them back on another image. These DeepFakes can be recognized by failures that occur due to differences between the images (e.g. light conditions, facial boundaries and gender). In case of Face Synch, the head position or facial expression of a person in an existing image or video is manipulated.
- Generate new images: it is also possible to create DeepFakes from scratch. In this case, an algorithm learns from an enormous amount of data how an object or structure can be created. An algorithm can easily learn to create faces from many existing examples, as human faces have a common structure in their facial geometry. Unstructured parts of the body (e.g. how hair is structured and the movements of hair) are more difficult for the algorithm to create.
We now know where to look if we would like to recognize a DeepFake. However, DeepFakes can be very close to reality. Therefore, 3DUniversum has developed a tool which can help you identify DeepFakes (try it out here).
New opportunities that are created by DeepFake
Imagine taking a physics class from Albert Einstein, or being welcomed to the Dalí-museum by Salvador Dalí himself. It is all possible due to DeepFake!
For media companies, new opportunities arise as well. In the future, news content might be recorded once, and will be adjusted with DeepFake according to the news of the day. DeepFake can also be used to anonymize people’s faces in a documentary, without having to blur their faces, to protect one’s identity.
One day, you may be watching your favourite show on a streaming service, and DeepFake enables you to become the main actor in the show. Synthetic media also allows content creators to change the original content, creating possibilities such as changing daytime into nighttime, or a summer landscape into a winter landscape.
Another powerful AI-related development is the opportunity to create visuals based on text. Have you been searching for an image of a bowl of soup which is a portal to another dimension? You have just found it! All you have to do is type in the text and the visuals will be created (explore it here). Also, DeepFake therapy exists, which allows people to communicate with a DeepFake of their loved ones who passed away.
Threats and questions while exploring various DeepFake use cases
As new opportunities are created by DeepFake, new threats and questions also arise. With today’s technology we can create a talking and moving image from one single image. Imagine the impact this can have on identity fraud, taking into account that current systems ask you to smile or look to the side to define your identity.
From a Legal perspective you could wonder who owns the rights of the content, when existing content is used to create new DeepFake content. Various ethical questions also rise when we talk about DeepFake and it’s usecases. Will there be a time when people will have to give their consent before a DeepFake of them can be created? What is out there to protect us?
In this new era, we will probably have many more questions to answer. Sezer’s final note to the audience was…
“See it but don’t believe it.“
Stay critical, and have your own judgement on the content that is given to you.
A big thanks to Sezer Karaoglu, for opening our minds and educating us on synthetic media and DeepFake. Feel free to invite him if you want to get inspired on how Deep Fake can help Humanity.
