Sierra Leone Is Writing Its First AI Strategy. The Minister Asked a Better Question Than Most Rich Countries Do.

Sierra Leone is writing its first ever National Artificial Intelligence Strategy. That sentence will be reported as "small West African country embraces AI," and the reporting will miss what is actually interesting about it.

What is interesting is the question the ministry decided to ask.

What is actually happening

The Ministry of Communication, Technology and Innovation, working with the World Bank Group, is developing the country's first National AI Strategy. It sits under the 50 million dollar Sierra Leone Digital Transformation Project, and it is being built on top of an AI Readiness Diagnostic organised around three pillars:

• Compute, meaning infrastructure.
• Capacity, meaning skills and talent.
• Context, meaning policy and regulation.

The stated priority sectors are education, healthcare and governance, and the ministry has also formalised a partnership with Qhala, a Nairobi-based firm, to help co-develop inclusive and ethical AI policy and build capacity across ministries, departments and agencies.

Those three pillars are a sensible frame. Compute, skills and rules really are the binding constraints, in that order, for most countries in Sierra Leone's position.

The framing is genuinely better than most

Here is the line from the Minister of Communication, Technology and Innovation, Salima Bah, that made me pay attention:

We are not asking how to fit AI into our systems; we are asking which national challenges AI can help us solve today.

Read that against how national AI strategies usually read. The standard version starts from the technology and works backwards to find a use for it: we must adopt AI, we must not be left behind, therefore here are the sectors we will insert it into. That produces pilots that impress visitors and change nothing.

Bah's framing inverts it. Start from the national problem, then ask whether AI is the right instrument. That is the correct order, it is the order this publication keeps arguing for, and plenty of far richer countries have not managed it. Credit where it is due.

Now the constraint that will decide whether the strategy survives contact with the country.

The number the strategy has to survive

At the end of 2025, Sierra Leone had roughly 1.85 million internet users, about 20.8% of the population. Around 7.02 million people, 79.2%, were offline.

There were about 8.94 million active cellular mobile connections, equivalent to 101% of the population, but that figure flatters the picture badly: many of those connections are voice and SMS only, and do not carry internet at all. Fixed broadband is minimal outside Freetown.

What that means for each pillar, concretely

Compute. The instinct will be to chase a data centre. Resist it. On this base the useful question is not "where does the model train" but "what can run cheaply, in batch, on infrastructure that already exists." Scheduled jobs that produce a result somebody can act on beat always-on services that assume connectivity nobody has. If the strategy funds one thing, fund cheap reliable inference and the data pipelines that feed it, not a flagship building.

Capacity. Policy literacy matters, but the scarcer skill is people who can take a model from a notebook into something that runs. A generation trained to write AI policy but not to deploy anything will produce excellent documents and no systems. Capacity should be measured in shipped things, not workshops attended.

Context. This is where the strategy can do what nothing else can: decide who owns Sierra Leonean data, on what terms it leaves the country, and what a citizen is entitled to know when a public decision is made with a model. That is genuinely a policy job, and getting it right early is cheaper than fixing it later.

What this looks like when it is real

I can be concrete here, because I built one.

The crop yield model I developed for Sierra Leone is exactly the shape of thing the "which national challenge" question should produce. It runs on national crop statistics and free satellite rainfall data. It needs no frontier compute and no data centre. It runs as a scheduled batch job, because rainfall arrives monthly and planting decisions move in weeks, not seconds. Its output is an early warning that a person deciding when to plant can act on.

It also failed honestly before it worked: trained on the crop statistics alone, it could not beat a naive baseline, which told us something a policymaker needed to hear more than any accuracy score would have. The data the country currently collects does not carry enough signal. That is a data-collection finding, not a modelling one, and it is precisely the kind of thing a readiness diagnostic should be surfacing.

That is what "AI for a national challenge" looks like in practice. Unglamorous, cheap, offline-tolerant, and useful. Not a chatbot.

The question I would like answered

A readiness diagnostic is only as good as who it asked. I would like to know, plainly, who was consulted: which Sierra Leonean practitioners, which researchers working on Sierra Leonean data, which people outside Freetown. Not because I assume the answer is bad, but because it is the single best predictor of whether the strategy describes the country or describes a workshop.

I would happily contribute. So would others. The talent exists, some of it in the diaspora, some of it at home, and a strategy that routes around it will be poorer for no good reason.

What would make me wrong

This is early. A diagnostic is not a strategy, and a draft strategy is not an implemented one, so judging it now would be unfair; I am reacting to a framing and a set of pillars, not a finished document. The full draft diagnostic is not public as far as I can find, so I cannot assess its actual recommendations, only the public description. Connectivity figures also move, and mobile internet in particular is growing, so the 20.8% will not stay 20.8%. And a fair counterpoint to my own emphasis: building for the connected fifth first is not automatically wrong, since that is where institutions and revenue are, and it may be the honest sequencing. I just want it said out loud rather than assumed.

Key takeaways

1. Sierra Leone is building its first National AI Strategy with the World Bank, under the 50 million dollar digital transformation project, on a Compute, Capacity, Context diagnostic.
2. The framing is problem-first, asking which national challenges AI can solve rather than how to adopt AI. That is better than most strategies manage.
3. About 79% of the country is offline. Roughly 7 million people. That is the constraint everything else has to respect.
4. On that base, the right AI is mobile-first, low-bandwidth, batch and offline-tolerant. Not always-on cloud services.
5. Ask who was consulted. A diagnostic reflects the room it was written in.

If you work in or on Sierra Leone: what national problem would you point this strategy at first?

---

Sources: Ministry of Communication, Technology and Innovation (mocti.gov.sl) on the development of Sierra Leone's first National AI Strategy with the World Bank Group, the AI Readiness Diagnostic and its Compute, Capacity and Context pillars, and the Qhala partnership; reporting by Ecofin Agency, Telecompaper and TechAfrica News on the readiness assessment and the 50 million dollar Sierra Leone Digital Transformation Project; connectivity figures from DataReportal, Digital 2026: Sierra Leone (1.85 million internet users, 20.8% penetration, 8.94 million mobile connections at end-2025). My crop yield model and its null result: arXiv

.13959.