The 'libraries as AI infrastructure' story isn't just European. African GLAM institutions are building linked data, authority control, and AI systems calibrated to their own knowledge traditions — not waiting for permission from European standards bodies.
Much of Africa's knowledge heritage is invisible to global discovery systems — not because it doesn't exist, but because metadata frameworks were built around European assumptions about 'work,' 'authorship,' and legitimate knowledge forms.
MARC records weren't designed for Gnawa ritual music, Swahili oratory, or Zimbabwe's performance traditions. When you can't describe something in the dominant framework, it doesn't appear in discovery systems. When it doesn't appear, AI systems don't know it exists. "This is the mechanism by which colonial knowledge hierarchies reproduce themselves in digital infrastructure."
Knowledge is primarily written. It has identifiable individual authors. It exists in stable fixed form. A griot's performance is not a 'work' in BIBFRAME. A praise poem that evolves with each telling has no single 'expression.' Forcing these into European categories distorts the thing being described.
Led by Ashleigh Harris (Uppsala) with Kenyatta University and University of Lagos. Wikibase-based multilingual metadata ontology for African genres in African languages. Starts from the premise: what if the ontology itself is wrong? Ontology design as cultural sovereignty.
Wikibase-based semantic authority control for Nigerian scholarly and cultural works. Links library records to Wikipedia and Wikidata. A Nigerian author moves from being a string to an identified entity in a linked graph. Addresses VIAF's thin African coverage.
Independently built AI tool in Marrakech. All heritage data stored in certified Moroccan national data centers — a deliberate data sovereignty choice. Rejects the 'clouded coloniality' of external infrastructure.
Lightweight AI tools using open-source Python and Hugging Face models trained on African datasets. Query response times: days → milliseconds. The cataloging work happening now is the foundation these AI tools run on.
European story: libraries built infrastructure before AI needed it (retrospective vindication). African story: authority files and AI being built simultaneously, with purpose visible from the start (construction).
When heritage data is processed in external cloud infrastructure under foreign legal frameworks — the extraction dynamic of colonialism reappearing in digital form. Linked data offers a partial answer through decentralization.
Building metadata frameworks from the epistemological ground up — not workarounds in existing frameworks. ALMEDA asks what linked data looks like when the knowledge it describes was never captured in the frameworks it was built to replace.
Library metadata as foundation for trustworthy AI — entity disambiguation, authority control, knowledge graphs.
Community governance of cultural heritage data — own infrastructure, own legal frameworks, own AI.
Galleries, Libraries, Archives, Museums building linked data calibrated to their own knowledge traditions.
Turning strings into connected entities — the disambiguation infrastructure AI systems depend on.
Audit your ontologies and authority files. What forms of knowledge are they built to describe — and what forms are they built to exclude?
Cite ALMEDA, SNAR, Dar Gnawa. Link from global systems. Federation, not extraction.
No single global ontology can describe all human knowledge. Support culturally-specific ontologies coexisting through linked data.
Direct funding to local infrastructure — not solutions that process African data in foreign cloud. Fund the builders, not the extractors.
VIAF (3/55 African countries), W3C, Dublin Core must include African institutions as equal participants — not recipients of exported standards.