AI weights are not open “source”

AI licensing is a complex process that goes beyond traditional software licensing. AI consists of multiple components, such as source code, weights, and data, each requiring different licenses. Additionally, AI licensing involves socio-ethical considerations, which has led to the need for more restrictions like behavioral use and distribution restrictions. In order to standardize the discussion around AI licensing, I have created categories for different license types. These categories include proprietary, collaborator, available, ethical, and open licenses. It is important to distinguish between source code and weights when licensing AI, and to have separate licensing frameworks for each component. The term “open source” should only be applied to licenses that meet the specific open source definition. It is also important to use precise terminology to avoid confusion and to accurately reflect the level of openness in AI licensing. By using the right nomenclature, such as “Open Weights” and “Ethical Weights,” we can develop industry standards for each category.

To top