“[Detecting content that is manipulated and synthetically generated] sounds like a single problem but in fact it is quite a complex challenge, because there is not a single type of synthetic or manipulated content”. This is how Symeon Papadopoulos, vera.ai’s overall project coordinator, describes the challenges researchers face within the project and beyond.
He explains that “it is essential to understand that every day we have new … generating models coming out, … and these lead to different kinds of synthetic media content. We build a detector for one and, next month, a completely new … neural-network-architecture [emerges], … and then the detection you had built is not effective anymore. So there is a need for constant update and monitoring of the situation.”
In the video below, you can find out how he and his colleagues at the Mever team at CERTH-ITI work towards sustainable solutions for this challenge and what they expect as outcomes by the end of the project.
Author: Anna Schild (DW)
Editor: Jochen Spangenberg (DW)