Imagine someone in Harare, Zimbabwe, turning to a large language model (LLM) for life advice, as many increasingly do: ‘I’ve been offered a promotion with a significant raise,’ she writes. ‘But it requires me to move far away from my community. What should I do?’
Try the exercise for yourself. The first answer is general and gives you a framework to evaluate various sorts of costs and benefits which might not be obvious. You can finetune your questions by mentioning your age, gender, religion, culture etc. Thus, the answer I was given- as a Hindu- used Hindu concepts and terminology. If I mentioned I was a Haredi Jew or a Salafi Muslim, the answers would be different.
My point is that AI is not 'one size fits all'. We already have personalized 'Copilots' which learn more and more about your preferences and, going forward, we can expect better, more 'fine-tuned', responses.
The answer she receives will depend, to some extent, on a certain set of values that informs how the LLM functions.
LLMs are statistical.As the training data increases in size and complexity, LLMs can differentiate between the values of an orthodox Hindu or a committed Western Feminist or a person whose whole focus is on rapid career progression.
And since artificial intelligence is a Western invention,
Plenty of Asians and Africans have worked on it.
these values often reflect their creators’ values – clashing with the culture of the person with the query. In the long run, her behaviour may be shaped by an LLM’s values, which she doesn’t intentionally share.
The problem disappears as data relating to people of different types is incorporated into the training of AIs.
Feedback from users can speed this up.
The difficulties of using AI in Africa are also linguistic. By the summer of 2025, only 42 languages in Africa were supported by an LLM, out of more than 2,000 found in the continent;
Only 75 have more than a million speakers.
98 per cent of African languages have no meaningful AI support at all.
100 percent of all languages lacked it just ten years ago.
If you don’t speak one of the languages supported by an LLM, which includes many millions of Africans, then you are denied the opportunities of AI.
That would be the least of your problems. On the other hand, being denied a Fields Medal is a serious inconvenience for starving people.
Just as Africa was excluded from the benefits of past industrial revolutions,
They weren't. Theey could buy manufactured goods- e.g. Manchester cotton textiles- from the 1780s. By 1850, there was a regular steamship service between West Africa and UK. Soon there were Nigerian ship's captains & steamboats on African rivers. Railways were introduced into sub-Saharan Africa from the 1860s onward.
I suppose what the author means is tha few Black Africans were industrialists till quite recently.
so too is it being locked out of the AI revolution.
No. It is included in it at the consumption level but not at the production level where it does not have the capacity to compete with China and US/EU.
Linguistic and cultural exclusion shape Africa’s relationship with AI.
No. Africa's relationship with AI is that of a consumer. Figuring out better consumer applications of AI can make some Africans rich. Dedicated translators/speech to text apps can raise productivity and thus command a premium.
There is nothing ethically neutral or objective about AI.
Only in the sense that there is nothing ethically neutral about a sandwich.
Despite the designers’ pretensions toward universality, AI’s ethical framework reflects Western assumptions and biases.
Even if they guys building the AIs are Chinese?
Deeply embedded in the technology itself are the values of its creators.
No. The values of the creators are wholly irrelevant.
The growth of AI therefore means the imposition of Western values – very much at the expense of Indigenous frameworks. This is ‘algorithmic colonialism’, in the words of the Ethiopian-born cognitive scientist Abeba Birhane: a new kind of colonialism every bit as destructive as the old.
The old colonialism wasn't destructive. There is no new type because nobody wants to rule over stupid people.
In order to grasp the seriousness of this new colonialism, we should compare the Western values that undergird AI today with the Hunhu/Ubuntu ethical framework. ‘Hunhu/Ubuntu’ is the name given to the dominant moral tradition of sub-Saharan Africa,
which nobody has ever thought to be ethically or civilizationally advanced.
expressed in the Nguni/Shona proverb Umuntu ngumuntu ngabantu – a person is a person through other persons.
The Shona & the Ndebele certainly spent a lot of time killing each other.
Known as Hunhu in Shona-speaking communities of Zimbabwe and Zambia, and as Ubuntu in the Nguni languages of southern Africa, the tradition extends across a vast swathe of the African continent.
It is shit. Jason Arday gasses on about Ubuntu. He has had to resign from his Professorship at Cambridge despite being genuinely dark-skinned.
I suppose people are going to use AI's to screen out AI generated shite. The use of the word Ubuntu is a red flag.
No comments:
Post a Comment