This is part 3 of a five-part series on artificial intelligence, seen through Rwanda's experience of building its institutions from scratch. Last week's piece explained why AI thrives on paperwork and records, the same ground bureaucracies have always stood on. ALSO READ: The driver who ignored the app Chantal took out a small loan on her phone last month. The app asked her to accept the terms. She scrolled past them the way almost everyone does, tapped agree, and had the money in her mobile wallet within minutes. She trusted the loan because everyone she knew used the same app. Not because she had read a single word of what she agreed to. ALSO READ: How Rwanda is regulating artificial intelligence That one tap sits on something older than any phone. Words let strangers work together in large numbers. A law, a contract, a bank note — these are just words millions of people agree to trust, without checking them. The historian Yuval Noah Harari has a name for this: he calls language the operating code of civilization. Money, banks, courts, even holy books only work because enough people believe in the words written on them. For most of history, that code stayed in human hands, because only humans could write it. The Wall Street precedent Go back to 2008. Banks in New York and London had built financial contracts so complicated that almost no one involved could see what was really inside them anymore. Not the investors who bought them, the regulators meant to check them, or the bankers who sold them. Some of those bankers trusted the label stamped on the outside — the same way Chantal trusted the app. When the hidden risk was real, the system fell apart, taking the economy with it. Millions of people who had never read a word of that paperwork lost their homes and jobs. ALSO READ: Kagame says investing in Africa's AI solutions a win for everyone The 2008 crisis wasn't only about greed or bad loans — it was a warning. The language had outgrown its own authors. Contracts written by code Humans had outsmarted humans. What happens when the one writing the contract understands language better than any human ever will? AI can draft, read, and negotiate agreements faster than any lawyer. It can build documents more complicated than anything Wall Street produced in 2008 — while sounding perfectly clear. A loan app, a phone contract, a land lease could soon be written by a machine — and still sound like plain common sense. ALSO READ: What to know about new global commission on AI co-chaired by Kagame Chantal's loan agreement is a small, everyday version of the same risk. She trusted it because a familiar app offered it to her, the way investors trusted a bad loan from a familiar bank. Rwanda has spent years building the laws and courts that let strangers trust each other on paper — a SACCO agreement a member can question, land titles a family can contest, and courts a citizen can walk into. That system only works if people can read and question the words inside it. The real risk isn't that AI will lie. It's that AI can write something completely true, and still make it impossible for anyone to understand. Why does this happen? Follow the incentive. Whoever pays for a contract wants two things. First, protection: if something goes wrong, they want the paperwork on their side. Second, low cost: they don't want to spend money making sure the other person understands. Covering every possible problem gives them protection. It also makes the contract longer and harder to read. That confusion mostly hurts the person who signs the contract, not the one who wrote it — so there's little reason to fix it. Some people do ask for both. But missing a loophole can mean a lawsuit — missing clarity rarely costs anyone anything. So, thoroughness almost always wins. Chantal never read her loan agreement, and it worked out fine — this time. The next contract she doesn't read might not have been written by a person at all. The writer is a professional engineer specializing in process design.
