This work was done during one weekend by research workshop participants and does not represent the work of Apart Research.
ApartSprints
Governance
Accepted at the 
 research sprint on 
Accepted at the 
Governance
 research sprint on 
January 7, 2024

Model Cards for AI Algorithm Governance

Here we present a framework to improve Algorithm Governance in AI Systems. We expect this framework to increase the transparency of the frontier AI Systems on an international level and at the same time increase the accountability of the AI Labs who are releasing these systems. Most of the countries are in the process of creating the first national regulations for frontier AI Systems which makes this time particularly critical to take advantage of the plasticity of the regulation and design systems, tools and frameworks that will foster international collaboration. However, so far these countries have had different focuses on their governance styles. which makes coordination and collaboration between them harder. On top of that the AI Labs are not constrained to any regulation or framework when it comes to the documentation of their AI Systems, which makes transparency and comparison of these models very hard for other entities different than the Labs themselves. Since China has taken the lead on algorithm governance for AI Systems and they have a tool to impose unified structure in the documentation (the algorithm repository) we believe that the EU and the US should present an equivalent framework in their new regulations that can be easily mapped to the Chinese algorithm repository. Here we present an initial proposal for such a framework based on Model Cards. We also suggest a mapping from the Chinese algorithm repository to the suggested Model Card framework and show an example of a possible Model Card for the AI System Claude 2.

By 
Jaime Raldua Veuthey; Gediminas Dauderis; Chetan Talele
🏆 
4th place
3rd place
2nd place
1st place
 by peer review