AI Action Summit 2025: PEReN and Viginum open collaborative artificial content meta-detector project
With the AI Action Summit just a month away, PEReN and Viginum are inviting researchers and experts to participate in the development of a digital commons consisting of a meta-detector for artificial content. Based on an open-source architecture, the tool uses a collection of state-of-the-art detectors.
The AI Action Summit, to be held in Paris on February 10 and 11, 2025, will focus on concrete actions to ensure the development and deployment of artificial intelligence for the benefit of our societies, economies and environment, in the general interest and with respect for the common good.
To contribute to the work of this high-stakes international meeting, PEReN and Viginum have developed and shared with international players in the fields of AI trust and information integrity, an open-source architecture for an artificial content meta-detector.
As part of an open, collaborative approach to creating a true digital commons, this project can be freely tested and installed, and anyone with expertise in the field is invited to contribute to its development and refinement.
Simultaneous interrogation of a collection of artificial content detectors
How does the project work? To develop the software building blocks needed to aggregate several, or even many, existing or future artificial content detectors (currently text and images) within a single tool.
This meta-detector has three objectives:- To be easy to use.
Simply, the content to be tested is sent to the meta-detector for analysis by all or some of the available detectors. Additional detectors can be easily integrated by following the documentation. - Take advantage of the strengths, weaknesses and complementarities of aggregated detectors.
Different types of content are addressed to all detectors to determine which are the most effective for each type of content. Is one detector more effective for faces and another for landscapes? For texts, is one detector better for a short format and another for a given language? Once this information is known, the content can be pushed directly to the most relevant detectors, saving both time and reliability. - Increase robustness and detection performance by judiciously combining detectors.
In a forthcoming phase, a large volume of content can be sent to a large number of detectors, so as to identify, on the basis of all the data generated, the association of detectors that will take best advantage of the characteristics of each. In the shorter term, analysis of the behavior of detectors in the face of typical alterations to the content present on networks will increase their robustness.
An open-source tool, under permissive license, to which anyone can contribute
The first software bricks structuring this artificial content meta-detector are published as open-source on PEReN’s open software forge.
This project is intended to be both collaborative and dynamic, offering the code developed to all scientific, economic, public and civil society players. While PEReN and Viginum are continuing to develop the project in the run-up to the Summit, they are already inviting all interested parties to contribute directly online.
Don’t hesitate to share this project or contact us with any questions you may have!