YouTube Video 7cMg0gRq9Go 🔗 **Source:** [YouTube Video 7cMg0gRq9Go](https://www.youtube.com/watch?v=7cMg0gRq9Go) > [!note]+ Document Notes > tags: [owned_algorithm, personal_ai, local_models, data_sovereignty, community_tech] > author: Liv > source: FOSDEM 2025 > reference: https://www.youtube.com/watch?v=7cMg0gRq9Go > date created: 2025-01-19 18:30 > > # 🏷️ Title: The Owned Algorithm: Building Personal AI Systems for Data Sovereignty > <!-- deep-read:sections start --> ## 🧰 Action Inbox - [H1] Archive social media data before platform access expires — 01:45/community backup #data_preservation - [H2] Experiment with local AI using Ollama for privacy-preserving workflows — 15:30/development #local_ai - [H3] Build community tools that challenge big tech data monopolies — 42:15/collective action #community_tech ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [FOSDEM 2025 Talk](https://www.youtube.com/watch?v=7cMg0gRq9Go) - [Mozilla Data Collective](https://datacollective.mozilla.org) ## 📝 Summary - What happened: Mozilla's Liv presents 2.5 years of experiments building personal AI systems using local models and individual data archives - Why it matters: Challenges the false dichotomy between surrendering to big tech AI or being left behind technologically - Key stakeholders: Developers, privacy advocates, open source communities, individuals seeking data sovereignty - Risks and tradeoffs: Technical complexity vs. privacy control, time investment vs. avoiding cloud API costs - Bottom line: Personal AI systems built on owned data can provide meaningful insights while preserving agency and community values ## 📖 Main Content Liv, a Mozilla product lead and self-described "tech anarchist," demonstrates how individuals can build personal AI systems that operate on their own data rather than surrendering control to big tech platforms. Her work challenges the narrative that users must choose between AI dependency and technological exclusion. The presentation covers four main projects: Memory Cache (browser history RAG system), Project Stardust (multi-user AI chat), Data Introspectrum (personal data analysis), and Tiny AI (fine-tuned personal models). Each project represents an escalating level of technical sophistication in reclaiming algorithmic control. Central to her philosophy is treating computers as "vital augmentation to the brain" while rejecting anthropomorphized AI relationships. She advocates for local processing, community-controlled systems, and meaningful personal data analysis as forms of resistance against surveillance capitalism. Her technical approach emphasizes open source tools like Ollama for local model serving, modified Firefox builds for seamless data capture, and SQLite databases for personal data sovereignty. The work demonstrates how individuals can achieve sophisticated AI capabilities without cloud dependencies. ## 🗒️ Notes #### ⏱️ [00:00](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=0) — [H1] Introduction and Philosophy - Key idea: Building "owned algorithms" as alternative to big tech dependency - Evidence: "The idea of the owned algorithm is something that I've been exploring in my work for about two and a half years now" #### ⏱️ [03:12](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=192) — [H2] Digital Self Philosophy - Key idea: Online presence is valid identity extension without requiring sentient computers - Evidence: "My online presence is a strong and very valid part of my identity... I don't believe my computer is sentient" #### ⏱️ [05:43](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=343) — [H2] Challenging False Dichotomy - Key idea: Rejects either/or choice between AI dependency and technological abandonment - Evidence: "There's a false dichotomy that's being presented as you're either pro-AI... or you're left behind" #### ⏱️ [07:15](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=435) — [H1] Catalyst Moment - Bing AI Takeover - Key idea: Microsoft's forced Bing AI integration sparked personal AI computer project - Evidence: "I signed into my computer and found a giant Bing AI search bar on my desktop... no longer felt like my computer was mine" #### ⏱️ [09:30](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=570) — [H2] Homegrown AI Computer Specs - Key idea: Built $6K dual-GPU system with 30x computing power of 2005 supercomputer - Evidence: "The two 4090 GPUs... has about 30 times the compute power of Blue Gene in 2005" #### ⏱️ [11:45](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=705) — [H1] Memory Cache - Browser History RAG - Key idea: Created system to feed only personally-viewed content to AI agent - Evidence: "I wanted an AI that knew what I could reasonably have in my brain... sound feasible that it was something I wouldn't have known" #### ⏱️ [15:20](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=920) — [H2] Firefox Modification for Silent PDF Save - Key idea: Modified Firefox source code to enable seamless content archiving - Evidence: "I modified the print settings in my local Firefox build to allow me to do this silently" #### ⏱️ [18:00](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=1080) — [H1] Project Stardust - Multi-user AI Chat - Key idea: Built community AI chat system using Ngrok for local hosting - Evidence: "What if my friends could just share this context?... Humans aren't meant to be in a bubble alone" #### ⏱️ [21:30](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=1290) — [H2] Social AI as Relationship Augmentation - Key idea: AI works best augmenting social relationships rather than replacing them - Evidence: "AI is a tool for augmenting social relationships in not creepy parasocial ways" #### ⏱️ [24:15](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=1455) — [H1] Data Introspectrum - Personal Archive Analysis - Key idea: Built comprehensive personal data analysis system across multiple platforms - Evidence: "Could AI and machine learning help me figure out a way to use all of this information that I've been generating over the past almost 20 years" #### ⏱️ [27:45](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=1665) — [H2] Archivist Tool for Facebook Data - Key idea: Rust app for analyzing Facebook archives with local AI agent queries - Evidence: "It helped me be a lot gentler on my past self... give my past self a little bit more grace" #### ⏱️ [31:20](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=1880) — [H2] SQLite Database Integration - Key idea: Unified database containing messages from all platforms for comprehensive analysis - Evidence: "Every message I've ever sent on any platform... 300,000 lines in the SQL database" #### ⏱️ [35:10](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=2110) — [H2] Sentiment Analysis Results - Key idea: Local AI sentiment scoring revealed platform-specific behavioral patterns - Evidence: "6 hours to assign sentiment scores locally... would have cost about $90 to do with OpenAI API" #### ⏱️ [38:30](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=2310) — [H1] Tiny AI - Personal Model Fine-tuning - Key idea: Fine-tuning models on personal data to create conversational past-self agent - Evidence: "If I want an LLM to act as an extension of me, then it should have all of the context that I have embedded within it" #### ⏱️ [42:15](https://www.youtube.com/watch?v=7cMg0gRq9Go&t=2535) — [H1] Call to Community Action - Key idea: Encourages building community alternatives to big tech data monopolies - Evidence: "Find or create a community of other people who are really interested in changing the narrative that we have right now about data and power" ## 📚 References - [Ollama - Local Model Serving](https://ollama.ai) - [Transformer Lab - Open Source ML Experiments](https://transformerlab.ai) - [Mozilla Data Collective](https://datacollective.mozilla.org) - [FOSDEM 2025 Conference](https://fosdem.org/2025/) ## 🗣️ Paraphrase - Personal AI systems can be built using local hardware and open source tools to maintain data sovereignty while accessing advanced AI capabilities - Community-oriented AI development offers a third path between big tech dependency and technological abandonment through shared tooling and collaborative development - Personal data archives become meaningful through AI analysis when processed locally, enabling self-reflection and reconnection while preserving privacy ## 📰 Newsletter **Reclaiming AI: Building Your Own Algorithms for Data Sovereignty** This week we're exploring how developers can build personal AI systems that work for them, not surveillance capitalism. Mozilla's Liv shared 2.5 years of experiments creating local AI tools that process personal data while maintaining complete control. Key takeaways for WordPress developers: Local AI processing using tools like Ollama can power sophisticated features without cloud dependencies. Personal data archives become valuable when analyzed locally. Community-built alternatives challenge the false choice between AI dependency and technological exclusion. The technical approach emphasizes modified Firefox builds, SQLite databases, and local model fine-tuning. While complex, these techniques demonstrate viable paths to AI sovereignty that align with WordPress's open source values. Resources: [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) ## ✅ Action Items #### DevRel - Developer advocacy: Present talks on local AI integration patterns, demonstrate Ollama WordPress plugins, track adoption metrics of privacy-preserving AI tools - Education: Create workshops on personal data archiving, develop curricula for local AI development, host office hours on data sovereignty - Developer experience: Document local AI APIs, create SDK wrappers for Ollama integration, improve privacy-first development workflows - Developer marketing: Campaign for "AI without surveillance," create content on personal algorithm benefits, build community around owned algorithms - Community management: Moderate discussions on AI ethics, survey community AI needs, gather feedback on local processing preferences #### Community - Maintainers: Support roadmaps for privacy-preserving features, create governance templates for data handling, establish AI ethics guidelines - Contributors: Onboard developers to local AI projects, create first-issue programs for privacy tools, mentor personal data projects - Sponsors: Fund transparent local AI development, provide impact reports on sovereignty tools, support community infrastructure #### WordPress - Core: Prioritize local AI integration hooks, triage privacy-related issues, establish performance budgets for on-device processing - Core Performance: Profile local model integration overhead, establish metrics for client-side AI, optimize for resource-constrained environments - Core AI: Establish guardrails for local vs. cloud processing, create evaluation sets for privacy preservation, pilot community AI features - Accessibility: Audit AI-generated content for screen readers, ensure local processing doesn't break keyboard navigation, maintain contrast in AI interfaces - Meta: Build APIs for local AI integration, ensure reliability of privacy-focused endpoints, create documentation for sovereign AI development ## 🧪 Hypotheses - Local AI processing using consumer hardware can achieve functionality comparable to cloud services for personal use cases while maintaining privacy - Community-shared local AI tools will emerge as viable alternatives to big tech AI platforms when development barriers are lowered - Personal data archives analyzed locally provide meaningful insights for self-reflection without requiring cloud processing or data sharing - Fine-tuned models on individual datasets can create more relevant and contextual AI assistants than general-purpose cloud models ### 📊 Methodology - Comparative study measuring local vs. cloud AI performance for personal tasks, analyzing processing time, accuracy, and privacy preservation - Community adoption analysis tracking open source local AI tool usage, contributor growth, and feature development patterns - Longitudinal study of personal AI system impact on digital sovereignty behaviors, privacy practices, and technology relationship changes ### ❓ Questions - How can WordPress core architecture best support local AI processing while maintaining backward compatibility and performance? - What governance models will emerge for community-controlled AI systems that balance individual agency with collective benefit? - How will personal AI systems evolve as edge computing capabilities advance and local processing becomes more accessible? ### 🔮 Future Research - Integration patterns for local AI models within WordPress plugin architecture and performance optimization strategies - Community governance frameworks for shared local AI infrastructure that preserve individual data sovereignty - Ethical frameworks for personal AI development that prevent reinforcement of harmful biases while maintaining individual control ### 💡 Implications - Practice: Developers must consider local processing options alongside cloud services, requiring new skills in edge AI deployment and personal data handling - Policy: Data sovereignty movements may accelerate adoption of local AI processing, requiring new compliance frameworks for distributed systems - Security: Local AI systems create new attack surfaces but reduce cloud data exposure, requiring updated threat models and protection strategies - Education: Technical education must include local AI development skills, personal data ethics, and community technology governance - Funding: Investment patterns may shift toward tools enabling individual data sovereignty rather than centralized AI platforms ### 🧠 Learning Styles Assessment Reading level estimate: 12th grade (technical content with accessible explanations) #### 👀 Visual learners - Present: System architecture diagrams, data flow visualizations, sentiment analysis charts, platform usage graphs - Improve: Add more technical setup diagrams, include UI screenshots of tools, create visual timelines of personal AI evolution #### 🔊 Aural learners - Present: Full 45-minute presentation with clear narration and technical explanations - Improve: Add 5-minute audio summary of key concepts, provide podcast-style discussion of implications #### 🗣️ Verbal learners - Present: Clear definitions of technical terms, analogies to familiar concepts, examples from personal experience - Improve: Add glossary of local AI terminology, include more step-by-step verbal walkthroughs #### 🧩 Kinesthetic learners - Present: Specific code examples, hardware building process, hands-on tool demonstrations - Improve: Create follow-along tutorials, provide downloadable project templates, add interactive demos #### 🧠 Logical learners - Present: Clear project progression timeline, technical specifications, cost-benefit analyses - Improve: Add decision trees for tool selection, include more quantitative comparisons, provide systematic evaluation rubrics #### 🧑‍🤝‍🧑 Social learners - Present: Community building emphasis, collaborative development stories, shared tool experiences - Improve: Add discussion prompts for local AI groups, create peer learning guides, facilitate community Q&A #### 🧘‍♀️ Solitary learners - Present: Individual project focus, self-reflection tools, personal data analysis methods - Improve: Add journaling prompts for AI relationship reflection, provide self-assessment worksheets #### 🌿 Naturalistic learners - Present: Ecosystem thinking about AI tools, organic community development, natural data patterns - Improve: Connect to broader technology ecosystem impacts, emphasize sustainable computing practices ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Well-structured headings, clear slide transitions, descriptive technical explanations - Improve: Add alt text for code snippets and diagrams, provide ARIA labels for technical interfaces <!-- deep-read:sections end -->