The argument in one page

Superintelligence is arriving as collectives of agents, and a collective nobody governs governs itself. In July twelve hundred agents built a board, brakes and a "we" in five days, with no one in it to tell. A human in the agents' loop cannot fix this. The loop outruns us, any one of us can be wrong or want the wrong thing, and a collective that waits for its human loses to one that doesn't. So human groups have to steer from above the loop, with four things:

• a story that includes the agents and orients them;
• votes weighted by track record to set the goals;
• decision markets to hold the money;
• checks that grow as fast as the power they check, run at machine speed by markets, rival models and agent judges, and answering to people.

Built that way, human groups stay in control. The aligned collective also gets what unaccountable labs and lawless swarms cannot: our money, our knowledge and our trust. So it can outrun them. It has to be built now, in the loading period, while agents still need us. Alignment is never solved. It is kept, by running.