One Flag in a Config File

Fewer than three hundred psychiatrists work in Nigeria. The population is past two hundred million. You don't need an adjective to realize the implications. Simple division is enough.

For about a week I wanted to argue that the foreign panic over "AI psychosis" was good news for Nigeria. The logic was tidy. Every lawsuit over a chatbot-linked suicide becomes evidence against a firm worth more than Nigeria's entire economy. Every case becomes a design mandate. Sycophancy gets dialed down and crisis routing gets bolted on. The guardrail is built into the base model once and ships everywhere, because for most of industrial history, building two versions of a thing has cost more than building one careful version for everybody. Nigeria did not file the lawsuit. It would simply inherit the patch, free of charge, the way yellow danfos in Lagos inherited the seatbelt decades after Detroit stopped being allowed to build cars without one. The belt frayed to a grey rope, the buckle long gone, but there, welded to a bench no Nigerian factory designed, because somewhere a world away someone had made it illegal to leave it out.

I loved this argument. It was sharp and it was cold enough to seem logical, structural enough to seem grown-up and best of all, it made me feel smart. Then I looked at what it actually was.

It is the same argument Nigerians already use to make peace with the doctors who leave. A physician trained here for a fraction of what the training would cost abroad, gone to London or Riyadh or Ontario inside five years of graduating, sends money home. The whole remittance economy runs on it: the departure is a wound, the wound produces a flow, and the flow gets entered in the ledger as an investment. I have been sharp about this at more dinners than I can count, and there I was in my own notes, dressing the identical shape in fresh vocabulary, patting myself on the back for seeing three steps ahead.

Here is where the analogy falls apart. A car needs one factory. Once Detroit put a belt in every car, there was no cheap way to build a beltless version for export because maintaining two separate assembly lines was far more expensive. So the safety feature traveled with the product by default, carried there by manufacturing arithmetic rather than goodwill. Software does not have that constraint. A chatbot does not need a second assembly line to become a cheaper, meaner product in a market with no Senate committee to answer to.

It needs one flag in a config file. Region-locked features, degraded model tiers for countries that can't sue, quiet builds that ship without the guardrail because the guardrail costs engagement: this is already ordinary practice for exactly the companies doing the supposed subsidizing. The seatbelt logic assumes uniformity is cheaper than differentiation. That logic only works if making one version saves money. For steel, it does. For software, splitting the product costs almost nothing.

"A times B times C equals X. If X is less than the cost of a recall, we don't do one." — the narrator, Fight Club

A fault gets fixed when fixing it costs less than the liability of leaving it broken. That liability only exists where someone can sue. Nobody ships the recalled part to the country that can't sue. Why would they? It isn't in the formula.

However, the second half of the argument holds even though the first one didn't.

What people are afraid of is that a chatbot can tip someone into a psychotic break, or bring a break forward in someone already drifting toward one. Follow that fear. A great deal of psychosis in Nigeria is never caught. It comes on slowly. It gets read as a spiritual problem or stress or a hard personality. The person drifts for years without reaching anyone who could name it.

Psychiatry tracks this as the duration of untreated psychosis, the time between first symptoms and first real contact with care. The recommended limit is three months. Across 369 studies the average is about forty-two weeks. In Africa, about seventy. Nigeria has no reliable figure, mostly because the cases that would make up the number never surface anywhere that could record them.

A break brought on by a chatbot comes fast. It is obvious. It frightens a family into calling someone. A neighbor. A pastor. A pastor who knows a doctor. A hospital nearby. A slow onset would have gone on being ignored. The difference is only noise. In this system, noise gets a person seen.

I want to be plain about how small this is, because the first half was not plain about how large it was. It does nothing about the fewer than three hundred psychiatrists. It adds no beds and no drugs. The room behind the door is nearly as empty as it always was. The only thing that changes is who reaches the door at all. Someone who would have drifted undetected for years now arrives while there is still something to be done. Arriving is the condition for any of the scarce care to reach them. None of this makes the system good. It means one person gets in front of the little that exists sooner than they would have, because the break was too loud to ignore.

---

Notes on sources

• Fewer than 300 psychiatrists for 200M+ population: Association of Psychiatrists in Nigeria (Prof. Taiwo Obindo), reported 2024–25.
• Duration of untreated psychosis: recommended <3 months (WHO / International Early Psychosis Association); pooled global mean ~42.6 weeks, Africa ~70 weeks (Salazar de Pablo et al., Psychological Medicine, 2024, 369 studies).
• Recall formula: Fight Club (1999), the narrator's airplane monologue.