Open, Transparent, and Responsible AI: Beyond Giant Black-Box Models
2025-10-01 , Artificial Intelligence

Most AI today runs on giant, opaque, black-box models — this talk explores how open source language models offer a sovereign, transparent, and responsible alternative.


Today, Artificial Intelligence is often associated with giant, closed, opaque, and centralized models. These systems function as black boxes: their training data is secret, their algorithms unverifiable, and their models unauditable. We don’t know where the data comes from, how it evolves, or who ultimately benefits.

But another path is possible.

In this talk, you will discover:
- What a truly open source language model is, from datasets to weights, with no “secret sauce”
- How such models can become digital commons, shared and auditable for the benefit of all
- How public resources like the Jean Zay supercomputer (GENCI) enable the creation of sovereign AI that is transparent and compliant with GDPR and the EU AI Act
- Why smaller, more efficient models are often a better choice than endlessly scaling up giant ones

By choosing Free and Open Source, we can build AI that serves the common good: transparent in its workings, sustainable in its resources, respectful of fundamental rights, and adapted to the real needs of citizens, researchers, administrations, businesses, and associations.

A passionate advocate for Free and Open Source Software, Benjamin is Business Development Manager for AI Solutions & Community Management at LINAGORA, where he turns open, privacy-respecting AI into practical products and helps grow vibrant contributor communities.

He also founded Ad Aures, creator of Castopod, a free and open source podcast-hosting platform built to empower creators and strengthen the open podcasting ecosystem.

Active in the Podcast 2.0 community, he contributes to the french radio show Libre à vous, a radio show produced by April (an association whose mission is to promote and defend Free/Libre Open Source Software), and hosts the podcast RdGP (Rien de Grave Patron).

Most recently, he joined WikiPortraits, a collective providing freely licensed, high-quality portraits for Wikipedia.

Andrzej Neugebauer. is the AI Program Director at LINAGORA and the creator of OpenRAG, a sovereign open-source framework for Retrieval-Augmented Generation. With over 15 years of experience, he has helped companies and institutions transform cutting-edge AI technologies into impactful, real-world solutions—ranging from knowledge graphs and intelligent search to healthcare innovation.

A specialist in natural language processing, machine learning, and AI systems architecture, Andrzej combines deep technical expertise with a strong commitment to openness and transparency. He has a solid academic background in theoretical mathematics, speaks English, French, German, and Polish, and enjoys working with diverse teams and communities to make advanced AI both reproducible and accessible.