Artificial General Detector

Artificial General Detector (AGD) is a free, open-source tool designed to find content created by artificial intelligence. It can check text, images, audio, and video. AGD uses a combination of 27 different detection methods. Instead of just giving a simple yes or no answer, it shows exactly where and how each method found something unusual. This makes it very reliable because if one method doesn't work, the others can still do the job.

Benefits

AGD offers full transparency by showing an anomaly map. This map explains what each detection technique found, making it easier to understand the results. Its design makes it strong against attempts to trick it. By using many independent methods, it remains effective even if some are bypassed.

Use Cases

This tool is useful for anyone needing to verify if content is AI-generated. It can be applied to various forms of media, including text documents, pictures, sound recordings, and videos. The framework includes specific techniques tailored for each type of media, such as analyzing pixel data in images, word patterns in text, sound artifacts in audio, and frame consistency in video.

Vibes

AGD has been tested on various datasets like HC3 for text and CIFAKE for images, showing it performs well compared to other tools. It is also tested on real internet audio and video to ensure it works with compressed media.

Additional Information

To use AGD, you need Python 3.9 or newer. The necessary components can be installed using pip. The project also provides scripts for testing and evaluation, and it can be set up using Docker. The AGD repository is well-organized with modules for different media types and tools for creating reports.