What Regulatory Framework for Open-Source AI Systems?
Inès Bedar · Zenodo (CERN European Organization for Nuclear Research) · 2024
Open source commonly refers to software made publicly accessible online under licenses that allow users to run, copy, distribute, study, modify, and improve models and data, either free of charge or for a fee. The core value of this approach lies in its ability to enable the decentralised and collaborative development of technologies. Open-source artificial intelligence (AI) plays a crucial role in today’s technological landscape. Most AI models currently available on the market are built upon open-source models released by major technology companies such as Meta, Google, or Microsoft, and subsequently reused hundreds of thousands of times. Examples include models such as LLaMA 2, Gemma, and Mixtral-8x7B. These companies release their models in open source in order to benefit from the intellectual dynamism of the open-source community, which helps challenge and improve their research and development efforts. Reusers, in turn, benefit from access to models they could not have developed from scratch due to the substantial infrastructure and resources required. The community can thus rely on existing models, refine them, and adapt them to specific use cases, while collectively advancing the state of the art. Undeniably, open-source AI enables a form of horizontal collaboration between actors that has few, if any, equivalents. Entire volumes could be devoted to open-source AI, given both its technological impact and what its ecosystem reveals about contemporary society. The following sections aim to provide key analytical insights into the current regulatory debates.