I ran a small test that I would encourage you to run yourself.
I asked an AI tool to assess the quality of leadership in Britain. Then I asked exactly the same question about Nigeria. Same phrasing, same session, minutes apart.
The first answer was measured. It described challenges, acknowledged debate, referenced institutional strengths, and arrived somewhere balanced. The second answer was also factually defensible, and its tone was noticeably heavier. Problems came first. Weaknesses were foregrounded. The framing was one of deficiency.
Neither answer contained an outright falsehood. That is what makes this worth talking about.
" Bias in these systems is rarely a lie. It is usually a matter of what gets mentioned first, what gets treated as normal, and what gets treated as a problem.
The easy explanation, and why it is incomplete
The standard explanation is training data. These models learned from what is written on the internet, and what is written about Nigeria is disproportionately written by outsiders, disproportionately about crisis, and disproportionately negative. Feed a system that diet and it will speak with that accent.
That explanation is true. It is also not the whole story, and I think the rest of the story matters more.
Three hands shape every AI system
I find it useful to think about three hands, because each one introduces something different.
The first hand is the hand that trains it. This is the data. Everything written, scraped, scanned and fed in. If the world wrote unevenly about a place, the system inherits that unevenness. This is the hand everybody talks about.
The second hand is the hand that decides what a good answer looks like. This is the part most people have never considered, and it is where I think the deeper bias lives. After a model is trained, human beings sit down and rate its responses. Better. Worse. More helpful. Less appropriate.
Those people have a nationality. An education. A set of assumptions about what a balanced answer sounds like and what counts as a sensitive topic requiring care. Their judgement is then baked into the system as a preference. The model is not just reflecting what the world wrote. It is reflecting what a specific group of people decided was a good way to talk about it.
The third hand is the hand that pushes it forward. The companies. Their commercial priorities, their legal exposure, their view of which markets matter and which risks are worth managing. A system built primarily for one market will be careful about that market’s sensitivities and comparatively casual about everybody else’s.
Why this matters practically
This is not an argument for avoiding these tools. I use them every day and they have made me considerably more effective.
It is an argument for something more useful: knowing what you are holding.
If you ask for a business plan for a Nigerian venture, you may receive assumptions imported from an economy with reliable electricity, functioning credit and predictable logistics. If you ask about a market here, you may get caution that is really unfamiliarity wearing the costume of prudence. If you ask it to assess something African, watch carefully whether it is describing or judging.
" These systems do not know they have an accent. You have to hear it for them.
Three habits worth building
One. Ask the mirror question. Whenever you get an answer about something local, ask the identical question about a Western equivalent. Compare the tone, not the facts. The gap between the two tones is the bias, and it is visible in about ninety seconds.
Two. Supply the context it is missing. Do not ask "how do I market this product?" Ask "how do I market this product to working mothers in Ibadan who buy primarily through WhatsApp, where most payments are transfers and delivery is by dispatch rider?" You are not being fussy. You are replacing its assumptions with reality.
Three. Treat it as a very well-read graduate, not as an authority. It has read more than any of us. It has lived nowhere. Your judgement is the thing that makes its output usable, and that judgement is not a weakness in the arrangement. It is your contribution.
And one more thing
There is something you can do about this that costs nothing.
The reason the internet speaks about us in a particular register is partly that we have not written enough of it ourselves. Every article a Nigerian professional publishes about their own field, in their own words, with their own market’s realities in it, becomes part of what future systems learn from.
That is not a grand statement. It is a practical one. Write about your work. Not because it will change the model tomorrow, but because the record of how we describe ourselves should not be left entirely to people who have never been here.