The DIVERSIFAIR project has officially reached its final milestone. On 19 May in Brussels, we hosted our closing conference, bringing together industry leaders, policymakers, researchers, and civil society advocates. Our goal was to reflect on our shared journey, showcase the concrete tools we’ve built, and chart the path forward for intersectional fairness in Artificial Intelligence. If you weren’t able to join us live, here is a summary of the core insights and highlights from across our sessions.
Key Insights
MC-ed by Tessa Bruijne (TNO), the conference opened with a vital reality check regarding the nature of technology: algorithmic bias is not an accidental glitch or a rare malfunction (Ilina Georgieva, TNO). It is structural to how AI systems are built. To move forward, our approach to building technology must transform from the ground up, acknowledging these flaws so we can proactively dismantle them.
The first keynote, by Gemma Galdón-Clavell (Eticas) cut straight to the heart of accountability, arguing that moving from the feeling that something is wrong to doing something about it requires the quantification of harm. AI models naturally optimise for the majority, creating a structural trap that systematically penalises minority groups. Because the impact of these systems is often invisible to the developers themselves, output auditing at scale is essential. We don’t need to decode every line of a black-box system to make a difference; we simply need to measure and regulate its real-world impact.
During our first panel, dedicated to the industry sector, “AI is Not Neutral,” the conversation shifted toward implementation and education (Moderator: Adebola Olomo – Women in AI, ZentAI Labs, Celeste Olayinka -Studio3Launchpad, Fet Özyürek – Turing College, Dr Laura Caroli – Independent, Women in AI). True AI literacy means involving everyone, not just engineers. Panelists noted that achieving algorithmic fairness requires a cultural shift: the future of AI governance demands more leaders and fewer technicians working alongside active, in-process frameworks to test fairness dynamically in live applications.
The Policy Panel “Intersections of power” brought together industry, policy, and critical legal scholarship to discuss governance (Moderator: Ilina Georgieva, Oussama Errazi – Sopra Steria, Monika Zalneriute – Vilnius University, Jeroen Delfos – European Commission AI Office). While the risk-based approach of the EU AI Act is a massive step forward, the panel highlighted its limitations. To prevent real-world harm, the legal framework must be supplemented by adaptive governance and aggressive grassroots action to enforce compliance and prohibit harmful AI applications.
Dr Alessandra Sala (Women in AI) closed the day with a sobering reminder regarding diversity: the tech gender gap has remained essentially unchanged for the last 50 years. We are still a long way from achieving 50% parity. The closing call to action was clear: the time for discussing these gaps has passed, and the ecosystem now requires immediate, concrete action on the ground.
Inside the Workshops: Showcasing DIVERSIFAIR resources
Beyond the main stage discussions, the conference featured interactive parallel sessions where we proudly showcased the tangible resources developed over the course of the project. Attendees could explore the AI Ethics course, developed by Turing College, which has already successfully trained 300+ professionals on the core principles of intersectional fairness (session facilitated by Fet Özyürek (Turing College). We also deep-dived into the Fair AI Scrum Handbook, developed by TNO, a practical framework that is already actively in use and helping to train tech professionals in agile environments (session facilitated by Quirine Smit, Eliza Hobo, Nina Van Liebergen (TNO)). Finally, we celebrated the launch of Eticas’s Community-led AI Auditing Guide, with Gemma Galdón-Clavell representatives from the A11 Initiative in Serbia (Gorica Nikolin) and the European Disability Forum (Kave Noori). This crucial resource was introduced based on the findings from real-world audits. Together, these three resources transform abstract ethical concepts into operational workflows for developers, educators, and community advocates alike.
Thank You
We would like to thank the eight consortium partners, our brilliant speakers, the conference attendees and all those who participated in the development of the DIVERSIFAIR resources. The project has reached its conclusion, but the ecosystem we’ve built together will continue to push for a fairer, more inclusive AI future.
