Jason Brooks, a member of the Fedora Project Governing Board, has submitted a draft of the guidelines for the use of AI tools in Fedora development. LinuxTwo weeks have been allocated for discussing the rule changes and making amendments, after which the governing board members will hold a vote to approve the new requirements.
AI tools are seen as a potential way to improve the platform, but there are also concerns related to privacy, security, ethics, and quality. It is proposed that the use of AI assistants should not be prohibited, provided that human oversight of their results is mandatory and the developer remains responsible for the code. The following rules for AI use are proposed:
- A developer using AI tools to prepare changes for Fedora is responsible for the submitted change and is obligated to review, validate, and test it. AI-generated content should be considered a recommendation, not final code or text. Submitting unverified or low-quality AI content is unacceptable and creates additional workload for those reviewing the changes.
- When submitting changes prepared using AI tools, you must add the "Assisted-by: name of the AI assistant" tag to the commit or pull request note.
- The use of AI tools to overcome language barriers or clarify thoughts during communication is encouraged.
- When reviewing other people's changes, it is recommended to limit the use of AI tools. AI can be used to assist in generating reviews, but it should not be used to fully automate the review process. The final decision on approving or rejecting changes should be made by a human.
- In project management, AI tools can be used to automate routine tasks such as spam filtering and note-taking, but their use is considered unacceptable in areas such as assessing code of conduct complaints, funding requests, nominating candidates for leadership positions, and preparing conference papers.
- The AI capabilities offered in the distribution, especially sending data to external servers, should be disabled by default and activated only after user notification and consent (opt-in). Useful applications of AI tools include their use in accessibility tools, such as translation, transcription, and speech synthesis.
- The creation of packages containing tools and frameworks necessary for research and development in the field of artificial intelligence is encouraged.
- Data generated by the project may be used to train AI models, provided the licenses are followed and the author is acknowledged. Aggressive data scraping, which places a significant load on the infrastructure, is prohibited (for effective access to the data, please contact the infrastructure team).
Source: opennet.ru
