What is OSAID 1.0
OSAID 1.0 (The Open Source AI Definition – 1.0) is the official standard released by the Open Source Initiative (OSI), used to clarify the conditions under which an AI system becomes open source. Organizations involved in formulating OSAID 1.0 include Microsoft, Google, Amazon, Meta, Intel, Samsung, the Mozilla Foundation, the Linux Foundation, the Apache Software Foundation, and more than 25 other organizations. OSAID 1.0 requires that AI systems must freely provide the necessary code, data, and parameters under legal terms recognized by OSI, ensuring that users can use, study, modify, and share the AI system without restrictions. OSAID 1.0 particularly emphasizes the importance of data information and code in machine learning systems, supports the exclusion of certain training data under legal restrictions, and promotes transparency and collaborative development of AI technology.
The specific definition content of OSAID 1.0
- Free to use: An open source AI system should support users in using the system for any purpose without requesting permission.
- Free to study: Users can study how the system works and inspect its components.
- Free to modify: The system can be modified, including changing its output, to meet any purpose.
- Free to share: Users can share the system with others, whether modified or not, for any purpose.
These freedoms apply to the complete system and also to the independent elements within the system. The premise is being able to access the preferred form for modifying the system. OSAID 1.0 specifically points out that for machine learning systems, the preferred form for modification must include the following elements:
- Data information: Provide sufficiently detailed data information so that a skilled person can use the same or similar data to build a substantially equivalent system. This includes a complete description of all data used for training, including the source, scope, and characteristics of data that cannot be shared, as well as how data is obtained and selected, labeling procedures, data processing and filtering methods, etc. It is also necessary to list all publicly available training data and how to obtain them, and all training data available from third parties and the ways to obtain them (including paid access).
- Code: Provide the complete source code used to train and run the system. The code should show the complete specification of how to process and filter data and how to perform training. The code should be provided under an OSI-approved license.
- Parameters: Provide model parameters, such as weights or other configurations. Parameters should be provided under OSI-approved terms.
For machine learning systems, an AI model consists of the model architecture, model parameters (including weights), and the inference code that runs the model. AI weights are a set of learned parameters that generate output based on a given input. "Open source model" and "open source weights" must include the data information and code used to derive these parameters.
Project address of OSAID 1.0
- Project address：opensource.org/ai/open-source-ai-definition
Who OSAID 1.0 is for
- Open source community: Provide clear standards and guidance for open source AI projects, helping to improve project quality and reliability and attract more developers to participate.
- Enterprises: Influence enterprises' release strategies for AI models, requiring adjustments to ensure compliance with the OSAID standard, including adjustments involving technical, legal, and business aspects.
- Developers: Provide more choices and flexibility; developers can choose and use open source AI systems according to the OSAID standard, promoting innovation and customized development.
- Educational and research institutions: Based on open source code, educational institutions and researchers can more easily access and use AI technology, promoting the development of AI education and research.
- Legal and Policy Making: Providing policymakers with a reference framework helps in formulating and adjusting AI-related laws and policies, ensuring that the legal system keeps pace with technological development.
Industry Advantages of OSAID 1.0
- Clarity: OSAID 1.0 provides a clear set of definitions and standards to help developers, enterprises, and users understand what truly open-source AI is, and how to ensure that AI projects comply with open-source principles.
- Promoting Innovation: By encouraging openness and sharing, OSAID 1.0 helps promote innovation in the AI field. Developers can freely access and modify the code of open-source AI projects, driving the development of new technologies and applications.
- Enhancing Trust: Following the OSAID 1.0 standard can increase users' trust in AI systems. Open source means transparency; users can verify how the system works, which helps build trust in AI technology.
- Collaboration Opportunities: Open-source AI projects can attract participation from developers worldwide. OSAID 1.0 projects provide a common framework, making collaboration easier and more efficient.
Scope of Impact of OSAID 1.0
- Impacting the development and use of AI models: OSAID stipulates that open-source AI models must provide sufficient information so that anyone can "substantially" reconstruct the model, including the source of training data, how it is processed, and how it is obtained.
- Promoting AI transparency and openness: OSAID emphasizes the importance of transparency, requiring open-source AI models to disclose the complete source code used for training and running AI, as well as model parameters.
- Promoting the healthy development of AI technology: It provides clear guidance and standards for the development and application of open-source AI.
- Impact on existing AI models: Many AI models that call themselves "open source" do not actually meet OSAID standards, which may have an impact on the development and use of these models.
- Legal and copyright issues: It has an impact on the legal status and copyright issues of AI models, especially on the question of whether models can be protected by copyright.