Wikimania 2025: AI, Africa, and Open Knowledge

As Wikimania marked its 20th year in Nairobi, a keynote panel brought together leading voices to explore how artificial intelligence (AI) is reshaping open knowledge—especially across Africa—and how the Wikimedia movement can remain inclusive, impactful, and sustainable.

Moderator: Nadie Gunasina (Chief of Staff, Wikimedia Foundation)

Panelists:

• Dr. Joyce Nakatumba-Nabende (Makerere University) — Shared applied AI work in health, agriculture, and language technologies in Uganda, emphasizing intentional creation of local-language datasets and participatory evaluation with communities.
• Prof. Vukosi Marivate (University of Pretoria; ABSA Chair of Data Science; Lelapa AI; Deep Learning Indaba) — Highlighted small-but-mighty African language models (e.g., PuoBERTa), the use of Wikipedia text within speech data pipelines, and the importance of sustainable AI (compute, water, and licensing considerations).
• Dr. Job Mwaura (Wits Centre for Journalism / LMU / Strathmore) — Discussed AI’s growing role in Kenyan media and social justice, from newsroom automation to the ethical risks of Western-centric models.
• Selena Deckelmann (Chief Product & Technology Officer, Wikimedia Foundation) — Recounted Wikimedia’s long-standing use of machine learning, reaffirming principles of transparency, sustainability, and community-governed tooling (e.g., revert-risk, translation support).

Watch the session

Key takeaways

• Language inclusion is non-negotiable. For open knowledge to be truly open, African languages must be a first-class focus in dataset creation, evaluation, and tooling—on Wikimedia and beyond. Community review surfaced practical issues (e.g., punctuation, idioms, code-switching) that standard benchmarks often miss.
• Grassroots networks power progress. Communities like Masakhane, Data Science Africa, and the Deep Learning Indaba are catalyzing research, training, and cross-disciplinary collaboration, enabling reusable recipes (e.g., classification and metadata tagging pipelines for local-language news and Wikimedia content).
• Small models, real impact. Compact models such as PuoBERTa demonstrate that targeted, domain-aware approaches can deliver high value without supercomputer-scale resources—especially when paired with open datasets and fine-tuning for specific tasks.
• Responsible **and sustainable AI.** Beyond “responsible,” the panel stressed sustainability—from energy and water costs to the hidden labor of content moderation. The call: build problem-first, community-co-designed systems, and choose simpler methods when they solve the problem.
• Open knowledge as critical infrastructure. Wikimedia projects already underpin AI research and applications. The movement’s commitments to transparency, open tooling, and human-first access (e.g., APIs to reduce crawler strain) are central to long-term resilience.
• Licensing and equity. Emerging data/model licenses (e.g., community-benefit or regional-equity terms) can expand choices for contributors and institutions seeking to prevent extractive use while enabling research and public-interest applications.

Notable contributions from Prof. Marivate

• Wikipedia in the data pipeline: For the South African Next Voices speech initiative (multi-hundred hours per language across seven languages), Wikipedia sentences were among sources for scripted reads—now filterable by origin in the released datasets.
• Reusable recipes for Wikimedia: Demonstrated how local-language classifiers (e.g., news categorization) can enrich metadata and improve semantic linking for Wikimedia content—an approach transferable across languages and projects.
• Sustainable AI lens: Argued for a broader framing that accounts for compute, energy, water, labor, and governance—prioritizing domain-fit small models and community-validated use cases.

Why this matters for Wikimedia @20

At 20 years of Wikimania, the movement’s durability depends on aligning AI with its core values: inclusivity, impact, and sustainability. That means supporting language diversity, investing in community capacity, stewarding open infrastructure, and advancing licensing options that protect contributors while enabling public-interest innovation.

Credits

• Moderator: Nadie Gunasina (Wikimedia Foundation)
• Panelists: Dr. Joyce Nakatumba-Nabende, Prof. Vukosi Marivate, Dr. Job Mwaura, Selena Deckelmann
• Event: Wikimania 2025, Nairobi (Onsite)