Explainable AI can enhance deepfake detection transparency

A new study by SRH University emphasizes the benefits of explainable AI systems for the reliable and transparent detection of deepfakes. AI decisions can be presented in a comprehensible way through feature analyses and visualizations, thus promoting trust in AI technologies.

A research team led by Prof Dr. Alexander I. Iliev from SRH University, with key contributions by the researcher Nazneen Mansoor, has developed an innovative method for detecting deepfakes. In the study recently published in the journal Applied Sciences, the scientists present the use of explainable artificial intelligence (Explainable AI) to increase transparency and reliability in the identification of manipulated media content.

Deepfakes, i.e., fake media content such as videos or audio files created using artificial intelligence, pose an increasing threat to society as they can be used to spread misinformation and undermine public trust. Conventional detection methods often reach their limits, especially when it comes to making the decision-making processes of AI models comprehensible.

In its study, the SRH University team carried out extensive tests in which different AI models were tested for their ability to reliably identify deepfakes. Particular attention was paid to explainable AI, which makes it possible to present the basis for the models’ decisions in a transparent and comprehensible manner.

This is done, for example, using visualization techniques such as “heat maps,” which highlight in color which image areas the AI has identified as relevant for its decision. In addition, the explainable models analyze specific features such as textures or movement patterns that indicate manipulation.

Prof Dr. Iliev, Head of the Computer Science—Big Data & Artificial Intelligence Master’s program, explains the importance of these approaches: “Our aim was to create technologies that are not only effective, but also trustworthy. The ability to make the decision-making process of AI transparent is becoming increasingly important—be it in law enforcement, the media industry or in science.”

The study shows that explainable AI not only improves recognition accuracy, but also promotes understanding and trust in AI technologies. By showing how the decisions were made, weaknesses in the models can be identified and future systems can be optimized in a targeted manner. This is a crucial step in strengthening the responsible use of AI in society.

Provided by
SRH University