A definition and license for open source compatible licensing of neural network weights
In the realm of AI, there's a fundamental misunderstanding that needs to be addressed–the assumption that the principles of open source software licensing can directly apply to Neural Net Weights (NNWs). The misconception stems from conflating two different artifacts - software source code and NNWs.
The software community created the open source concept as a means to make source code available to anyone for use, modification, and distribution. Source code is human-readable, which means anyone with programming skills can understand, debug, and enhance it. It was an overwhelming success.
NNWs are different. They represent the 'knowledge' an artificial neural network has learned and are often stored as large matrices of numbers. Unlike source code, NNWs are not human-readable nor debuggable. In simpler terms, you can't look at a weight matrix and understand what it signifies or infer any specific knowledge from it.
Open source's foundational freedoms–to run, study, redistribute, and modify software–do not translate easily to NNWs. While you can run and distribute NNWs, studying and modifying them is non-trivial, or functionally impossible.
It is critical for the industry to develop and standardize on "Open Weights" licensing frameworks. These frameworks should align closely with the founding Four Freedoms of free software but should be specifically tailored for NNWs.
The technology community needs a standard for Open Weights that recognizes the unique nature of NNWs and provides legal and practical guidelines for their use, distribution and sharing. This requires collaboration from the entire AI community, including developers, researchers, legal experts, and regulatory bodies.
It will not work to abdicate this definition to governments. After two years of rhetoric, they have not come up with consistent or useful standards. The Open Weights definition follows the path of open source software movement, to create voluntary community standards. This approach is not intended to be inconsistent with regulation. If a consensus on regulatory standards emerges, the defintion may need to be re-examined for consistency with them.
Also, a definition of Open Weights should not import subjects such as privacy, human rights, or clearance of data inputs into its licensing principles. Those are important topics, but they will take time to figure out and are not easily addressed with licensing terms. This definition focuses instead on the original idea of openness, and preserving the original goals of Freedom Zero of free software and open source. Others will create their own standards for restrictions and ethical licensing, and to participate in the legislative process to set the standards of society for proper use of AI, the information used to train it, and the information it produces.
This definition should be developed in the open, much like open source software itself. Therefore this definition and license will be published on GitHub and the community will be invited to improve it.
This repository also contains a model license for permissive licensing that meets the Open Weights definition. There is no plan to release a "copyleft" or "sharealike" license, because these models cannot translate well to NNWs.
The text in this repository is dedicated to the public domain under CC0. If you contribute to this project, your contributions will also be under CC0.
Interactions for this project are subject to the Mozilla Community Participation Guidelines.