On AI, power, and why Africa is building something it will never be allowed to own
The tower of babel has a server farm, where we carry someone else's message
MI’s 5 Minutes of Fame | Week 22
This week I am stepping sideways from Somalia. Not far, and you will see the connection soon enough, but I want to talk about artificial intelligence (aka Ai1), and I want to do it in a way that actually means something to people in our part of the world rather than the way it usually gets discussed, which is as a story about Silicon Valley genius and American tech billionaires.
Because here is the thing. The AI revolution is not being built for us. But it is, in important and troubling ways, being built by people who look a lot like us, who live in cities that look a lot like ours, who are being paid almost nothing for work that makes other people extraordinarily rich. That is a story worth understanding properly, and I will get to it.
But first, two things happened in the same week that I cannot stop thinking about. I should say, of course, that for many of us in our community, that week carried a significance that no political event could compete with. The Hajj and the Eid preparations, family obligations, and the particular kind of focus that the season demands. I understand if your mind was, rightly, elsewhere. Mine was too, for much of it. But these two things stayed.
Four days apart, and they could not have been more different. Donald Trump cancelled a planned executive order that would have set up a voluntary mechanism for AI developers to have their models reviewed by federal agencies before public release. He killed it after a phone call from a tech investor friend who told him it was too interventionist. Worth pausing on the word voluntary. Nobody was going to be forced to do anything. That was the threshold that proved too high.
Then Pope Leo XIV, the first American pope, released his first encyclical. The title is Magnifica Humanitas, which translates roughly as the magnificence of humanity, and the subject is artificial intelligence. He did not call for AI to be abandoned. He called for it to be governed: “those who control AI will impose their own moral vision, which will become the invisible infrastructure of these systems.” Without governance, he wrote, the moral vision of the people currently building these tools will simply become the default moral vision of the world. Right now that means the moral vision of Elon Musk and Mark Zuckerberg.
Two Americans. Same week. One of them a tech investor’s phone call away from any oversight at all, the other asking humanity to take seriously what it is unleashing. The Catholic philosopher Romano Guardini wrote decades ago that contemporary man has not been trained to use power well2. He was writing about nuclear weapons. The sentence has not aged badly.
The people who are actually building these systems will tell you, if you catch them honestly, that they are scared of what they are making. At a dinner in San Francisco convened for Australian AI policy officials, senior engineers and architects went around the table naming their p(doom), which is the industry shorthand for the probability that AI destroys humanity. None of them said zero. Some said fifty per cent. These are not people writing dystopian fiction. They are the people writing the code.
Ilya Sutskever, who was OpenAI’s chief scientist, reportedly told colleagues the plan was to build a bunker before releasing artificial general intelligence. AGI is the term for AI that matches or exceeds human cognitive ability across the board. The plan is to build the thing, release it to the world, and hide. I am not sure what to say about that except that it tells you something important about the gap between the stated mission and the actual priorities.
Karen Hao’s account of OpenAI’s internal history is worth reading if you want to understand how this happened. The organisation began as a non-profit with a genuine mission to develop AI safely and share the benefits with humanity. Over time, the pursuit of AGI became something closer to a religious conviction than a research programme. The “machine god”, as some insiders called it, became the point. Everything else, including the safety work and the public benefit mandate, became secondary to the race to get there first.
Now let me tell you about Kibera.
Kibera is one of the largest informal settlements in Africa, on the western edge of Nairobi. Unemployment runs at around fifty per cent. Clean water is unreliable. Sanitation is poor. It is also, for the past several years, one of the places where the AI systems now reshaping the global economy were cleaned and made safe for public consumption.
In late 2021, OpenAI sent tens of thousands of text snippets to a San Francisco-based outsourcing firm called Sama, which employed workers in Kenya. The text had been scraped from the internet and included descriptions of child sexual abuse, bestiality, torture, murder, and self-harm. The job was to read it, label it, and filter it out so that ChatGPT could present to the world as clean, helpful, and safe. Workers were paid less than two dollars an hour. A TIME investigation in January 2023 found that many of them developed post-traumatic stress disorder. They were not warned about the nature of the content before they started. They were told they were contributing to something important.
For years, a sign outside Sama’s Nairobi office read: “Samasource: THE SOUL OF AI.”
When OpenAI was asked about this, a spokesperson confirmed the arrangement and described the workers’ contributions as “immensely valuable.” The company’s mission, the statement explained, is to ensure that artificial general intelligence benefits all of humanity.
All of humanity. Take a moment with that.
Sama eventually cancelled its contracts with Meta and OpenAI. It no longer recruits content moderators. The workers it left behind, some of them dealing with lasting psychological damage from years of reading the worst material the internet produces, received no meaningful compensation. The AI systems they helped build are now worth hundreds of billions of dollars. One journalist who visited Nairobi to report on this described data labelling as a significant portion of the Kenyan tech workforce. She said that almost everyone she met had either done the work themselves or knew someone who had. Leaving the airport, you drive straight past the office where it happened.
Meanwhile, the platforms built partly on this kind of labour are restructuring the information economy in ways that hit the Global South hard. Google’s original promise, made by co-founder Larry Page in 2004, was that the company existed to get you out of Google and to the right place as fast as possible. That philosophy is now dead. The company’s AI systems answer questions without requiring users to leave the page at all. Traffic to news websites from Google search fell by a third in 2025. Industry projections suggest a further forty-three per cent decline over the next three years.
For publishers in Africa, the Middle East, and diaspora communities who spent years building digital audiences on the assumption that Google would keep sending readers their way, this is not a disruption. It is a structural elimination. The economic model that made independent journalism viable in the digital era is being dissolved, and the people who built those audiences have no seat at the table where these decisions are being made.
At Meta, Zuckerberg laid off eight thousand people in a single week as part of a pivot toward AI infrastructure. His plan is to invest up to a hundred and forty-five billion dollars in AI in 2026 alone. He told employees that humans were, in the circumstances, a casualty of the shift in priorities. This is the same company that relied on African workers for years to moderate its platforms under conditions that repeatedly attracted legal action.
The geopolitical conversation about AI rarely includes Africa except as a footnote. The dominant frame is US versus China. When Air Force One stopped to pick up Nvidia’s chief executive Jensen Huang in Alaska on the way to Beijing, it was a statement about where power now sits: in chips, in semiconductors, in the capacity to train large models at scale. What began as a trade dispute about soybeans and tariffs has become a contest over the technologies that will determine military and economic dominance for the next generation.
Africa is not absent from this contest. It is present in the most extractable way. Labour, data, and infrastructure, provided cheaply, with minimal regulatory protection, and with no share of the value created. Andrew Leigh, an Australian government minister, gave a speech recently titled “The Economics of Human Extinction” in which he argued that the risk posed by superintelligent AI is categorically different from other economic risks because extinction permanently forecloses all future human welfare, innovation, and potential. He is right about the category. But the foreclosure of welfare and potential is not a future event for the workers in Kibera. It is already the present tense.
Back in Australia, new research by education consultants Ben Jensen and Emily Pearson surveyed more than four thousand teachers and school leaders in New South Wales. They found that AI has moved into schools much faster than any plan to manage its consequences. Three quarters of secondary teachers report students using AI to complete assessments. The researchers call the result a mirage of false mastery: students producing polished-looking work that reflects no actual thinking or learning. The phrase stuck with me because it applies well beyond the classroom. The AI revolution as currently constituted risks becoming exactly that, a mirage of false progress, where the outputs look like transformation but the underlying structure keeps the same people accumulating value and the same people absorbing risk.
Leo’s encyclical is not a counsel of despair. It does not call for AI to be shut down or abandoned. It calls for it to be governed, disarmed in the sense of being made accountable, and built in the interests of humanity rather than the interests of the people who happen to control the infrastructure. Those are different things, and the distance between them is where most of the harm is currently occurring.
Guardini was writing about nuclear power when he said that contemporary man has not been trained to use power well. But he was really writing about a pattern of human behaviour that keeps recurring: we build things we cannot govern, we distribute the risks to people who had no say in the decision, and we call the profits a sign of progress. The slave trade was progress for some people. The structural adjustment programmes of the 1980s were progress for some people. The question each generation has to answer is whether it can see the pattern while it is still happening rather than fifty years after the fact.
The soul of AI, the sign said, was made in Kibera. The profits are banked in San Francisco. That is a political problem. Political problems have political solutions, but only when enough people understand clearly what they are looking at.
That is what this week’s edition is for. Until next week. Comments and feedback as always welcome.
References and Sources