AI through History, History through AI

I have just returned from the Eighth Conference on Digital Humanities and Digital History in Luxembourg, which this year had the theme "AI through History, History through AI".

A big thank you to Sarah Oberbichler and the Luxembourg Centre for Contemporary and Digital History (C²DH) at the University of Luxembourg for organising such an inspiring event. The papers critically examined the impact of AI on history: on research and method, on memory, and on whose past gets remembered.

I had the chance to present the MCP server I built for the Islam West Africa Collection (IWAC), and I learned a great deal from the other participants.

Ask a commercial AI chatbot about Islam in West Africa and you will most likely get a generic, stereotypical answer: the region is thin in the training data. IWAC itself is open access, and over the past year AI bots from the US and China have repeatedly accessed its servers to scrape the entire collection. Yet for all that traffic, the chatbots barely draw on the collection and rarely cite it.

The MCP server takes a different approach. You can ask for specific information in plain language, such as newspaper articles on Islam in Abidjan in the 1990s, Islamic publications on secularism in Togo and Benin, or academic references on Muslim women in Burkina Faso. Each query pulls from a different part of the collection with a different mix of filters (e.g. place, period, subject, country), and the AI determines which tools to use and how to combine them. The result is a much more grounded answer, with claims linked back to specific documents.

My slides are available, and you can test the MCP server yourself from my GitHub page; more details and the link are in the slides.