Reflections on Enhancing Access to Records through Cultural and Indigenous Intelligence at the Jesuit Historical Institute in Africa in Nairobi, Kenya
If you had walked into the Jesuit Historical Institute (JHIA) archives in Nairobi five years ago, you would have found orderly rows of boxes, careful handwritten finding aids, and a dedicated archivist who knew exactly where everything was, provided you asked in person. Today, this institution has begun increasing accessibility by digitising research outputs, and administrative files. But digitisation alone is not the revolution. The quiet revolution is happening where artificial intelligence meets the archive, and where AI is being asked not just to store the past, but to understand it, share it, and do so fairly.
Yet as we have discovered, we cannot simply import the AI models designed in Silicon Valley or London. Kenyan electricity is not always reliable, even though we have solar backup. Our budgets do not stretch to massive GPU clusters. And most importantly, our knowledge systems, indigenous naming conventions, oral traditions, faith‑based theological records, and community narratives, do not fit neatly into Western archival taxonomies. This reflection asks: how can the Jesuit Historical Institute in Africa design an AI‑driven archival system that is not only FAIR (Findable, Accessible, Interoperable, Reusable) but also culturally and intellectually embedded, computationally realistic, and genuine for social good?
In our institute, digital archive systems now exist. We have scanned thousands of records, stored electronic theses, and built multimedia documentation of institutional life. But look closer. Most systems are still keyword‑based rather than semantic. Metadata schemas vary wildly between departments. Indigenous knowledge, if it appears at all, is squeezed into categories that were never designed for it. And crucially, very few archival staff have been trained on AI or data science.
When we try to introduce AI, we hit three walls. First, energy and compute: generative AI needs stable power and expensive hardware. Cloud services shift costs to hard currency and create dependency. Second, data scarcity: AI thrives on large, structured datasets, but our archival materials are under‑digitised, and indigenous knowledge often exists in oral or undocumented forms. Third, interoperability: each department runs its own isolated system, using different formats and standards, so collaboration becomes nearly impossible.
But there is a fourth wall, less discussed but equally important: epistemic bias. When we use off‑the‑shelf AI models, they do not recognise Luo naming patterns, Kalenjin oral genealogies, or the theological categories of a Catholic mission archive. They are not malicious; they are simply trained on data that never included nor understands African dialects.
The FAIR principles are usually discussed in well‑resourced European or North American data centres. But we have learned that FAIR is even more urgent in low‑resource settings.
Findability does not require a supercomputer. Lightweight natural language processing tools running on a standard server can automatically generate subject tags and persistent identifiers. We do not need GPT‑4; we need good enough semantic tagging that works through intermittent power.
Accessibility for us means low‑bandwidth web portals, offline‑first mobile apps, and role‑based access that respect both privacy and cultural sensitivity. We have explored the possibility of implementingAI chat‑based retrieval on small local servers, not the cloud, and it works.
Interoperability is our greatest collective opportunity. If three faith‑based universities adopt shared RDF‑based metadata models and simple APIs, suddenly our separate archives become a shared knowledge ecosystem. That is not a technical fantasy; it is a policy choice.
Reusability demands clear licenses, provenance trails, and transparency. AI can help generate those trails automatically, making accountability a feature, not an afterthought.
The most exciting insight from our work is this: AI does not have to be culturally blind. We are now experimenting with what we call culturally grounded classification. Instead of imposing Western archival taxonomies, we are building systems that incorporate indigenous naming conventions, recognising oral traditions as legitimate archival records, and allow communities to define their own categories.
Training AI on local language datasets is no longer a distant dream. Recent projects in Kenya, collecting thousands of hours of speech in Kikuyu, Kalenjin, Dholuo, Maa, and Somali, prove that it is possible. We can do the same for archival materials. Imagine an archive where a user can search in Samburu proverbs and retrieve mission records, or where a theological student can trace indigenous concepts of hospitality through decades of community engagement reports.
Of course, this requires governance. We would have to establish an AI oversight committees, cultural sensitivity review boards, and data sovereignty policies rooted in Ubuntu, the philosophy that personhood is achieved through community. Technology without ethics is not innovation; it is risk.
We can deliberately choose not to chase large generative models. Instead, we can focus on “Small AI”: small language models, edge computing, algorithms optimised for low‑resource environments, and even solar‑supported backup systems. This is not a compromise. It is a strategic choice. Small AI is affordable, maintainable, and context‑appropriate. It runs on everyday laptops and smartphones. It can be repaired locally. And it is far less likely to introduce the ethical hazards of massive, opaque models.
In practical terms, this means our AI‑driven archives can support institutional research analytics, preserve histories that have been marginalised, retrieve evidence for policy work, and empower community researchers. One university has already reduced its document processing time from days to minutes using a real‑time pipeline designed for low‑bandwidth settings. That is social good, not speculative hype.
For JHIA, the archival mission is not merely administrative. It is for preserving memory and promoting historical knowledge. We hold records of spiritual heritage, missionary encounters, indigenous resilience, and community service. To lock those records in inaccessible formats or to rely on AI models that cannot respect their cultural meaning is to fail that mission.
Knowledge justice means that a grandmother’s oral testimony about land use in the 1870’s is as archivally valid as a vice‑chancellor’s annual report. It means that a student in a rural area with a basic smartphone can search across multiple university archives without a high‑end connection. It means that we build AI systems that serve our values, not the other way around.
We are not replicating the Global North’s AI infrastructures. We are pioneering something else: context‑sensitive, sustainable, culturally intelligent, and deeply committed to the common good. That is the archive we want to leave for the next generation.
By Philip Opiyo – Librarian | Jesuit Historical Institute in Africa (JHIA)
