# FSFs Criteria for Free Machine Learning Applications > [! note]- > The content of this page is generated by audio/video transcription and text transformation from the content and links of this source. Source: [https://fosdem.org/2025/schedule/event/fosdem-2025-4818-fsf-s-criteria-for-free-machine-learning-applications/](https://fosdem.org/2025/schedule/event/fosdem-2025-4818-fsf-s-criteria-for-free-machine-learning-applications/) <video src="https://video.fosdem.org/2025/k1105/fosdem-2025-4818-fsf-s-criteria-for-free-machine-learning-applications.av1.webm" controls></video> ## Summary & Highlights: **Introduction to FSF's Criteria for Free Machine Learning** The session introduces the Free Software Foundation's (FSF) efforts to define criteria for free machine learning applications. It focuses on the concept of user freedom within machine learning, contrasting it with traditional software freedom. The FSF emphasizes the importance of understanding machine learning's unique elements, such as training data, and discusses the ethical implications of these technologies. **Understanding User Freedom in Machine Learning** The FSF outlines the challenges of defining user freedom in machine learning, which includes both software and non-software elements. They highlight the importance of training data and other parameters in ensuring freedom. The discussion also touches on the ethical considerations of using non-free machine learning systems and the potential for these systems to be 'free but trapped' due to dependencies on non-free elements. **The Role of Free Software in Society** The session reiterates the significance of free software in empowering users to control their computing. It highlights the historical context of the FSF's efforts, including the development of the GNU General Public License and the Free Software Definition. The FSF's commitment to maintaining these principles in the context of machine learning is emphasized as a continuation of their mission. **Challenges and Future Directions** The FSF acknowledges the complexities of machine learning and its societal implications, such as environmental impact and ethical concerns. They invite feedback and collaboration from the community to refine their criteria. The session concludes with a call to action for ongoing dialogue and input from stakeholders to ensure that machine learning technologies align with the principles of freedom and transparency. ## Importance for an eco-social transformation The session is highly relevant for eco-social transformation as it addresses the ethical and social implications of machine learning, such as user freedom, privacy, and data transparency. For eco-social designers, the FSF's criteria can guide the development of machine learning applications that prioritize ethical considerations and user empowerment. Challenges include navigating technical complexities and potential conflicts with existing copyright laws. The session encourages collaboration and input from diverse stakeholders to address these issues and advance sustainable technology practices. ## Slides: | | | | --- | --- | | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_001.jpg\|300]] | The first slide introduces the Free Software Foundation's criteria for free machine learning applications, presented by Zoë Kooyman and Krzysztof Siewicz. The focus is on defining user freedom in the context of machine learning. | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_002.jpg\|300]] | The second slide provides a disclaimer that the FSF's work on free machine learning criteria is a work in progress. The presenters acknowledge they are not machine learning experts and are open to collecting questions and feedback from the audience. | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_003.jpg\|300]] | The third slide highlights the Free Software Foundation's mission to promote computer user freedom. Founded in 1985 by Richard Stallman, the FSF supports the GNU Project and maintains the Free Software Definition and the GNU General Public License. | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_004.jpg\|300]] | The fourth slide emphasizes the importance of free software in allowing users to control their computing. Free software is integral to software development, with the GNU GPL and AGPL among the most popular licenses on platforms like GitHub. | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_005.jpg\|300]] | The fifth slide discusses the Free Software Definition, which states that access to source code is a prerequisite for the four freedoms: to run, study, modify, and share software. | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_006.jpg\|300]] | The sixth slide celebrates 40 years of free software, noting the publication of the Free Software Definition in 1986 and the evolution of the GNU GPL to include three versions. The AGPL is noted as increasingly popular. | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_007.jpg\|300]] | The seventh slide presents the growth of the machine learning market, which has surged by over 60% since 2020 and is projected to grow by 535% by 2030, reaching a value of over half a trillion dollars. | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_008.jpg\|300]] | The eighth slide discusses the rise of computing done with machine learning, highlighting its role as an intermediary between users and software. The slide warns against 'freewashing' attempts and notes the FSF's role in providing guidance. | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_009.jpg\|300]] | The ninth slide addresses the freedom-related challenges of machine learning, noting that these systems lack true 'intelligence' and are composed of both software and non-software elements. The slide emphasizes the importance of understanding these elements for studying or modifying machine learning systems. | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_010.jpg\|300]] | The tenth slide outlines the FSF's aspirational criteria for freedom-respecting machine learning applications, emphasizing the importance of free training data and the four freedoms, with a focus on studying and modifying. | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_011.jpg\|300]] | The eleventh slide asserts that users should have all necessary elements to control their computing with machine learning. It highlights the importance of free training data and the need for retraining from scratch rather than incremental training. | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_012.jpg\|300]] | The twelfth slide poses open questions regarding the moral justification of using machine learning applications without available training data and the concept of 'free but trapped' applications. The FSF invites audience feedback. | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_013.jpg\|300]] | The thirteenth slide outlines the FSF's next steps, including discussions with technical and philosophical experts and the development of a statement of criteria. The FSF aims to create an aspirational goal for the future. | ![[FOSDEM 2025/assets/FSFs-criteria-for-free-machine-learning-applicatio/preview_014.jpg\|300]] | The final slide thanks the audience for their attention and participation in the session. ## Links [FSF’s criteria for free machine learning applications](https://fosdem.org/2025/events/attachments/fosdem-2025-4818-fsf-s-criteria-for-free-machine-learning-applications/slides/238472/Kooyman_S_W6BKphT.pdf)