Open Source AI

= " A movement advocating for transparent, modifiable, and democratically governed artificial intelligence systems, rooted in the ethos of digital commons."

Definition

OSI:

"What is Open Source AI

When we refer to a “system,” we are speaking both broadly about a fully functional structure and its discrete structural elements. To be considered Open Source, the requirements are the same, whether applied to a system, a model, weights and parameters, or other structural elements.

An Open Source AI is an AI system made available under terms and in a way that grant the freedoms1 to:

1. Use the system for any purpose and without having to ask for permission.
2. Study how the system works and inspect its components.
3. Modify the system for any purpose, including to change its output.
4. Share the system for others to use with or without modifications, for any purpose.

These freedoms apply both to a fully functional system and to discrete elements of a system. A precondition to exercising these freedoms is to have access to the preferred form to make modifications to the system."

(https://opensource.org/ai/open-source-ai-definition)

Characteristics

Model Openness

Releases include weights (open-weight), training data, and code (e.g., Mistral AI, LLaMA 2). See: Open Weights

Contrasts with "open API" models (e.g., OpenAI) that hide core infrastructure.

Governance

Hybrid licenses (e.g., CreativeML Open RAIL-M) impose ethical use clauses.

Community-driven oversight (e.g., EleutherAI, LAION).

Ecosystem

Tools for federated training (Flower), ethical datasets (The Stack), and local deployment (Ollama).

Timeline

Timeline of Open Source AI, via DeepSeek:

==2000–2010: Pre-Open AI Foundations

2001: GNU Free Documentation License – Early model for open knowledge sharing, later influencing AI licenses.

2007: Launch of Apache Mahout – Open-source machine learning library, emphasizing collaborative development.

2011–2019: Open AI Emerges

2015: TensorFlow released by Google (open-source, but with corporate control).

2019: EleutherAI forms – Grassroots collective building open LLMs (e.g., GPT-Neo).

2020–2022: Community-Led Models

2020: BigScience (France) – Hosts BLOOM, the first multilingual open LLM

2021: LAION releases OpenCLIP – Open dataset for image-text models, critical for Stable Diffusion.

2022: CreativeML Open RAIL-M License – Ethical use clauses for open models (used by Stable Diffusion).

2023: Corporate Co-optation & Backlash

March: Meta releases LLaMA 1 ("open" but with restrictions).

June: French Mistral AI founded – Advocates for open-weight models in EU policy

July: LLaMA 2 released – Sparks debate over "open-washing"

2024: Policy Battles & Commons Alternatives

January: EU AI Act amendments target open-source exemptions (French NGOs lobby for commons protections).

April: Mistral 7B – French model outperforms proprietary alternatives, fueling sovereignty debates.

June: Open Source AI Summit (Paris) – Co-organized by La Quadrature du Net

Examples

1. Mistral AI French startup lobbying for EU-wide open-weight AI standards. https://mistral.ai
2. Pythia Transparent LLM by EleutherAI, with documented training data. https://eleuther.ai
3. OpenRAIL Licenses Ethical use licenses for AI, co-designed by civil society. https://www.licenses.ai