AI and Africa Symposium · University of Kansas · 17 September 2026
AI and the work
of the historian
Researching Islam and Muslims in West Africa.
Curating the sources through which we study them.
These slidesslides.frederickmadore.com
Abundance has a history
A familiar problem for historians: “scanning parties”, thousands of photographs, little time to read them.
Newly readable, still interpreted
- Multimodal LLMs extract text from printed and handwritten pages
- Draft transcriptions to check against the original page
- Small LLMs identify names, places and organisations
Which spellings and uncertainties survive?
Do two names identify the same person?
Which names and places get recorded?
A wrong transcription still reads well. Keep each transformation traceable to the document.
Reading printed Ewé
Togo Presse, 31 July 1976.
Loading the extraction…Model-generated transcription · scroll to inspect the output.
Where do the sources direct our attention?
Explore the places most frequently mentioned in the IWAC—and the documents that mention them.
What does a connection mean?
A co-mention is a starting point for investigating a relationship.
What should I read next?
AI-assisted sentiment analysis · 12,349 newspaper articles
Affordable access,
unequal control
Low entry costs, little local hardware.
Dependence on external providers.
Increasingly capable alternatives.
Greater control when run locally.
Small models can run on personal computers; larger ones need substantial infrastructure.
Open weights do not mean equal access to computing resources.
Whose sources become usable?
- DigitisationFaster processing still depends on preservation and scanning.
- LanguageFrench-language gains are only a starting point.
- ExpertiseThe people who know the sources remain essential.
Expanding what
historians can do
More accessible sources. New tools.
New starting points for reading.
Historical judgement throughout.
The goal isn’t simply to process more sources,
but to help more people work with them.