AI is learning your craft. Own what it learns.

Manifesto

Who owns what AI learns from you?

1. 01

   AI learns from people.

   Every model you have ever used was shaped by people. People wrote the books, the code and the conversations it read. People graded its answers, again and again, until it got them right.

   That second kind of work is now an industry. The labs pay experts by the hour to grade and fix their models' answers, and the companies that supply those experts are worth billions. The rest of us do the same work for free: every thumbs-up, every correction, every suggestion we accept.

2. 02

   Paid once, owned by someone else.

   The people who teach AI are paid once, or not at all. What they teach becomes someone else's asset, and it keeps earning long after they've been paid. Often, it becomes their replacement.

   That isn't a law of nature. It's a choice about who owns what AI learns.

3. 03

   What a machine can check, AI learns. What only people can judge, it gets worse at.

   The newest models are superhuman at puzzles a computer can mark. But the rewards that make them brilliant at code flatten their writing, and the labs will keep choosing what they can measure, because that is where the money is.

   A good sentence, a faithful translation, a safe triage note, a fair contract, in Czech or Swahili or Tagalog: that judgment lives in people. It is the scarcest thing in AI, and it is ours.

4. 04

   The last time this happened, people invented two things.

   In 1844, twenty-eight working people in Rochdale, England, many of them weavers, opened a shop and agreed that its surplus would go back to the people who bought there, in proportion to what they bought. The modern cooperative was born.

   In 1847, a Paris songwriter named Ernest Bourget took a café owner to court for performing Bourget's songs without paying, and won. Three years later, Bourget and two other songwriters founded the society that became SACEM, which still pays musicians when their work is played.

   Both were invented by people a new economy was leaving behind. Both still exist. We need both again.

5. 05

   So every craft gets a guild.

   A guild is a craft in a language: writers in English, translators of Czech, nurses in Manila. Its members write the exam for their own craft, from their real work, and grade every model blind, frontier or open, from any country.

   Their grades and fixes train the guild's own judge, and the judge trains the guild's apprentice: a small specialist model that does the craft well, costs little to run, and belongs to the guild. When the apprentice works, the people who taught it are paid. Every contribution is on a public ledger, in their name.

6. 06

   Small models make fair pay possible.

   Nobody can say what share of a giant general model is yours. But an apprentice trained on three thousand translators' corrections has a short list of authors, and the ledger names every one of them. The more AI splits into specialists, the more exactly the people behind it can be paid.

   And a guild's judgment outlives any model. Models go stale in months. The judgment keeps growing, and it can teach the next open model in days.

7. 07

   It belongs to no country.

   The race to build the most capable AI is run by a few companies in two countries. They train for English, and for whatever a machine can check. That is exactly where they are weakest: other languages, local law, local care, local culture. It's where apprentices win first, and where most of the world has been left watching from the stands.

   One member, one vote, whether you live in Lagos, Lima, Lyon or Los Angeles. Seats on the council for every region, so no country can outvote the rest. Guilds founded by the people who speak their language.

8. 08

   A ladder, not a cliff.

   AI will let newcomers do work that used to take years to learn. That can push experienced people off a cliff, or it can build a ladder.

   In a guild, it's a ladder. A newcomer works with the apprentice, the people who taught it are paid for that use, and as the newcomer gets good, their corrections start teaching it too.

9. 09

   Owning it is not about getting rich.

   We will never sell a token. There is no dividend to wait for and no share price to watch. You are paid for work you do.

   What you own is better than a payout. It's a guarantee: an AI that can't be turned against the people who taught it, because nobody above them profits from turning it. And it can never be sold off, not even by a future vote of members chasing a windfall.

10. 10

    Judge us by what we publish.

    Every contribution on the ledger. Every exam result in public. Every member vote in public.

    That includes who holds this today. For now it is run by a company, because a company can sign contracts and pay people, and a cooperative that doesn't exist yet can't. Our founding documents will set the number of members, and the date, at which it passes to a cooperative of its members. If we ever break these promises, you'll be able to see it, and you'll be able to leave with everything that's yours.

11. 11

    It starts with your judgment.

    The first cooperatives started with a few neighbors and a decision not to wait for someone else. This one starts with a task from your work, and two answers to it.

    Pick the better one. Fix it. Put your name on it.