AI Alliance Program Governance

Outline

1. Definitions
2. Collaboration and Contribution
3. Governance
4. Intellectual Property

Definitions

Core Project

A set of living artifacts with goals and a roadmap to address substantial cross-community challenges and are an opportunity for individual contributors and members to collaborate, build, and make an impact on the future of AI. Core Projects are managed directly by the AI Alliance and governed as described below. These are the foundational building blocks of the AI Alliance.

Affiliated Project

A set of living artifacts with goals and a roadmap aligned and important to the Alliance mission and are typically from member organizations who seek deeper collaboration and impact, often by building with or on Core Projects or other Affiliated Projects, but the project owners retain full management of the project. All Affiliated Projects must have open permissively licensed artifacts and clear community calls-to-action.

Initiative

A collection of projects within a common theme, often with common governance across multiple projects of the initiative.

Collaborator

Any individual that contributes to any project in nearly any substantive way. Collaborators can be from member organizations or the broader community. Collaborators are not just users.

Member

An organization that has committed to be an anchor sponsor of a major component of the AI Alliance program and is part of the AI Alliance overall governance based on the defined membership process.

Sponsor

An organization that has made a significant resource commitment to a project or initiative. This may include people, funding, compute, data, or other services.

Working Group

A group of collaborators working together toward a specific goal for a project or initiative. Typically a working group will be part of a Focus Area, but it may cut across more than one focus areas.

Focus Area

One of six goal-oriented thematic programs of the AI Alliance.

Collaboration and Contribution

The AI Alliance aims to accelerate and disseminate open innovation across the AI technology landscape to improve foundational capabilities, safety, security and trust in AI, and to responsibly maximize benefits to people and society everywhere. This is a broad heterogeneous program that embraces diversity of thought and action. As such we favor a lightweight operating and governing structure with maximum empowerment of individual collaborators and organizational members and sponsors to pursue projects in ways best fit to the project.

We do require that:

• Every Collaborator adheres to our community Code of Conduct.
• Every Member and Sponsor adheres to our Competition Law Guidelines.
• Every Project adheres to a set of minimum criteria and incubation process.

Grass roots collaboration is at the heart of the AI Alliance. Good ideas and contributions come from many people in many places. The AI Alliance seeks to discover, enable, and scale good ideas and contributions from across the global AI community. The processes and governance of AI Alliance are intended to support this.

Governance

Governance is a key ingredient in any successful project. At the same time, the optimal type and amount of governance will vary among Projects and Initiatives, especially with a portfolio as broad as the AI Alliance.

Governance must serve the community and our goals, not the other way around! It should be neither absolute nor one-size-fits-all.

Required Governance

All Core Projects and Initiatives should:

• have a published goal(s) to produce an open permissively usable artifact (often many).
• have at least two organizational Sponsors.
• have a named lead/co-leads with contact information and roles published.
• include a public description of the Project on the AIA website that lists Sponsors, leads/maintainers, Github link, how to get involved and contribute, and how decisions are made.
• be open and inviting for any person to become a Collaborator.
• Note: most, but not all of the above, can currently be accomplished by automated approaches and we highly encourage this to promote fast, efficient, enjoyable progress.

Affiliated Project Governance

The AI Alliance does not require specific forms of governance for Affiliated projects. 

Focus Area Governance

At least 2 co-chairs from two different member organizations organize, establish OKRs, and facilitate work and decision-making as servant-leaders. Focus Areas have a set of Projects (Core and Affiliated), Initiatives, and Working Groups within their scope.

• Note on Cross-listing: in some cases, Projects, Initiatives, or Working Groups may cut across two or more Focus Areas. In these cases a primary Focus Area is identified and the governance is addressed in the specific context by those involved.

Working Group governance

At least 2-co-chairs organize, establish OKRs, and facilitate work and decision-making as servant-leaders. Working groups typically exist within one of the Focus Areas but can sometimes be cross-cutting.

AI Alliance Steering Committee

A cross-Alliance advisory and decision-making function on matters of overall strategy, progress toward goals, and major decisions that affect people and projects broadly, for example the expansion of membership or official positions on policy.

AI Alliance Legal Entity

The AI Alliance is incorporated as a hybrid legal entity. A US 501(c)3 non-profit entity holds and manages the majority of work and projects. A US 501(c)6 holds and manages specific industry-focused work such as policy lobbying, and establishing commercial practices or standards. Each of these entities has a governing board with limited scope to decide matters of overall program operations including finance and legal matters.

Recommended Governance

The AI Alliance includes many leading individuals and organizations in open source and AI, with a large amount of collective experience and wisdom in open community-oriented work. We therefore recommend that some/many of the following elements or governance structures can be useful, depending on the nature of the project:

• Project or Initiative Steering or Advisory Committee: a group of individuals that include representatives from each Sponsor as well as key technical leaders from each working group.
•  Working Group: as defined above, a standing team with leads/co-leads responsible for a specific part of a Project or Initiative based on a defined charter and scope.
• Task Team: operate under the umbrella of a Working Group, consisting of at least one lead and Collaborators that carry out tasks supporting the relevant Working Group charter and roadmap, often doing so independently.
• Leads: individuals delegated with responsibility to make technical decisions on a specific scope of the project.
• Decision-making: what is delegated to who, and what is handled by which group, and how escalation is handled. For example, it is often useful to define an escalation path from individual to task team lead, to working group lead, to steering committee.
• Empowered Tech Governance: technical decision-making should be granular and distributed to those doing, i.e. the actual individual leads and Task or Working Groups. They should be empowered to make technical decisions quickly with a strong bias toward action.
• Meritocracy over Bureaucracy: people and groups who make progress and do things get more say in the trajectory of the project than those who are on the sidelines. Governance by Github commits vs. relying on panels and committees.
• Limited Non-technical Governance: should be kept minimal and responsive to only actual hurdles and challenges, such as legal blockers on the use of a data set or regulatory adherence for the release of a specific tool or model.
• Processes that Serve Goals: process exists only to enable faster, broader, and more impactful action.
• Open Governance: Governance of the projects are inclusive, inviting anyone from the community to join any working group or task force of interest, encouraging engagement and contribution to the group's efforts and decision-making processes.

Intellectual Property

Contribution of IP to an AI Alliance Core Project (“Core Project”) does not require or involve the contributor to give away the contributor’s ownership in the contributed IP. The Developer Certificate of Originality (DCO) used by the AI Alliance does not change ownership of the contribution (e.g. the AI Alliance does not become the “owner” of contributions”), but instead requires that anyone submitting content to a Core Project affirm that they can and are providing anyone and everyone a license to their contribution under the applicable open source license(s) for the project. This is true regardless of whether the contribution is submitted by an individual from a member or non-member of the AI Alliance.

Contribution of IP to an AI Alliance Affiliated Project (“Affiliated Project”) in general follows different processes, which are typically outlined in the relevant Github repository, web or other documentation associated with the project, and not the AI Alliance Core Project processes identified in this document. We ask the Affiliated Projects to clearly identify themselves as “Affiliated Projects” on the AI Alliance website, Github, Hugging Face, and wherever else they are listed in affiliation with AI Alliance.

The AI Alliance has standard open source licenses that it applies depending on the type of artifact contributed to the Core Project, and a Core Project should not modify these standard licenses without exception approval by the Steering Committee or by a Work Group or Lead delegated to by the Steering Committee. These licensees are:

• Code: Apache 2.0; “code” refers to code, software, or other similar content written in a computer language.
• Data: CDLA Permissive 2.0; information or other content contributed for the purpose of being part of a dataset.
• AI Model Weights: Apache 2.0; parameters, weights, and biases in an AI model.
• Documentation and similar content: CC BY 4.0; information or other content that is not Code, Data, or AI Model Weights contributed for the purpose of informing persons who read it.

For example, contribution of an AI model would call for contribution of the model weights and associated software under Apache 2.0, with any model card or other information about the model and its performance under the CC BY 4.0.

We acknowledge that certain artifacts may call for a deviation from these standard licenses, for example if a particular concern about safe use is identified. These situations are handled by the exception process through the Steering Committee.

When setting up Core Projects in a repository like HuggingFace or GitHub, Core Projects should reference the DCO used by the AI Alliance; as well as the standard licenses that apply to Code, Data, AI Model Weights, and Documentation. For GitHub the DCO is incorporated into the contributor workflow in a per-commit sign-off. Because the act of contributing the content to the Core Project triggers the contributor’s licensing of the content under the respective Core Project open source licenses, it is important that content is contributed to the Core Project by the contributor before it is used by the Core Project. 

Special Concerns with the Contribution of Data

Projects involving data should keep in mind that the provenance, dependencies and license terms of the source(s) of data, and relevant legal and regulatory statutes may affect the ability to contribute. For example, unlike open source software, it is rare for data publicly available on the Internet to be made available for use under a license.  Whenever there is a question about ownership and usage of data, please seek guidance from the relevant lead(s)/maintainer(s) of the project before proceeding. In particular, the AI Alliances Open Trusted Data Initiative was established to address the many challenges and opportunities in data for AI.

Confidential Information

Confidential information is at odds with open technology and open innovation and should not be exchanged or contributed as part of AI Alliance Core Projects, and more broadly should never be introduced into any Alliance project, discussion or event. Avoiding exchanging confidential information also supports antitrust compliance as laid out by the Competition Law Guidelines for the AI Alliance.

Pre-release Information and In-Progress Artifacts

While we aspire to have all work and all output of AI Alliance be open, accessible and permissively usable at all times, we appreciate that work in progress may not always be amenable to incremental release. In these scenarios we still encourage collaborators to release early and often, use Github and publicly-accessible documents to ensure broad access, and use reasonable effort to invite the broad community to participate when feasible.

Core Project’s repositories should include links to the AI Alliance’s Code of Conduct and the Competition Law Guidelines for the AI Alliance. Leads should reference Competition Law Guidelines for the AI Alliance in formal meetings by reading or showing the following language:

An important reminder is that all AI Alliance activities need to comply with applicable antitrust and competition laws and that you should comply with the Competition Law Guidelines for the AI Alliance, which are available from the AI Alliance website. There should be no discussions between competitors enabling agreements or concerted actions that restrain competition, including the exchange of confidential information that enables competitors to engage in illegal concerted action. Moreover, participation in AI Alliance activities does not commit any participant to use any specific technology, restrict the use or sale of any technology, or otherwise limit any participant’s ability to engage in similar activities with others. If you have questions about these matters, please contact your legal counsel for advice.

If you have questions about this guidance, please reach out to the relevant lead.