Analytic

Ostrom's principles come from commons where users "can substantially harm one another, but not situations in which participants can produce major external harm for others", in cases she says are "in no sense a random sample". Much frontier AI risk is harm to people outside the group using the system. So her evidence supports polycentric self-governance inside a collective whose members bear its costs, which is our own case, but does not by itself support polycentric governance of the world's AI.

Institutional design