Microsoft Unveils AI Tool to Detect Deepfakes Ahead of Elections
Microsoft introduces Video Authenticator, an AI-powered tool designed to analyze and score the likelihood of manipulated media in videos and photos. The tech giant releases this software to combat the spread of deepfake election propaganda.
Microsoft releases Video Authenticator, an AI-powered tool designed to combat deepfake election propaganda by analyzing manipulated images and videos. Developed in collaboration with Microsoft's Responsible AI team, the software automatically scans media to provide a confidence score indicating whether the content has been artificially altered. The system trains on the FaceForensics++ open-source dataset and undergoes testing using the extensive DeepFake Detection Challenge Dataset created by Facebook AI and other technology vendors.
Despite the promising premise, experts express skepticism regarding the actual effectiveness of the new detection tool. Forrester analyst Brandon Purcell notes that Microsoft does not release any specific metrics about Video Authenticator's accuracy. Since the winning algorithm of the recent DeepFake Detection Challenge only achieves an accuracy level of around 65%, there is significant concern that Microsoft's tool performs at a similar rate.
If the tool operates with only moderate accuracy, many deepfakes inevitably slip through the cracks while authentic content risks being falsely labeled as fake. Purcell emphasizes that Microsoft must reveal more technical details about the model itself for the public to place confidence in the software. However, Microsoft currently declines to provide data about the tool's accuracy or details on how users can access the technology.