New Research in Detecting AI-Generated Videos

Researchers at Drexel University have developed a new tool called MISLnet to detect deepfakes, or AI-generated videos. The tool was created using years of data from detecting fake images and videos. Traditional methods of detecting fake videos look for changes made to digital video or images, such as added or moved pixels. However, AI-generated videos don't contain the same disparities between pixel values as camera-captured videos. The Drexel team's tools, including MISLnet, use a constrained neural network to learn and differentiate between normal and unusual values at the sub-pixel level. This approach allows the tools to detect deepfakes more effectively than traditional methods. MISLnet is the latest development in the ongoing arms race between creating and detecting deepfakes.