Free/Libre AI

Interest in Free & Open Source Software has increased dramatically over the past six months because of the rapid growth of Generative AI models. Machine Learning, and in general AI, has been around for more than a decade now, but Generative AI thrusts AI into the hands of the layperson. The FOSS ecosystem already suffers from chronic open-washing, which has been exacerbated by the GenerativeAI sector. Because AI involves multiple data and software components - data, pre-training scripts, serializable models, inference, and deployment - GenerativeAI model creators have been able to play it fast and loose with the term "open-source model". The Open Source Initiative (OSI) created an Open Source AI Definition (OSAID) (https://opensource.org/ai/open-source-ai-definition), which is endorsed by a lot of influential organizations (https://opensource.org/ai/endorsements), but it has also received criticism (https://sfconservancy.org/blog/2024/oct/31/open-source-ai-definition-osaid-erodes-foss/). And the pace at which the GenerativeAI ecosystem is moving makes it hard for most of the software developer ecosystem to keep up.

In this panel discussion, we would like to discuss what a "Free/Libre AI" would look like, a la Free Software. The ideological differences between the Free Software Movement and the Open Source Software Movement significantly impacted the adoption and sustainability of the FOSS ecosystem. Given the adoption, interest, and potential impact of GenerativeAI models, we need to understand if "Free/Libre AI" is even possible, how we could go about building it, and why we need policies to prioritize Free/Libre AI over closed (or even "open-source") models.

1. Understand the spectrum of "openness" when it comes to AI models, particularly GenAI models, i.e., open weights models, models that adhere to the OSAID by OSI, and models that achieve an ideal "Free/Libre AI" status, like those from the Allen AI Institute
2. Get an idea of the role of datasets in building AI models, and the modes of licensing available for datasets at the moment
3. What technical hurdles need to be overcome to build Free/Libre AI models
4. Why we (the public) should be demanding "Free/Libre AI" models for citizen services