A World for Swarms: Master Algorithms When No Single Mind Is in Charge

Repository abstract

Ants cross a gap by building a bridge out of their own bodies, and they do it with no member holding the plan. The question this article is built around is the design question that follows: what would a world have to provide for collectives alone to run it? Since Turing a line of researchers has proposed one principle that would account for intelligence in general — universal induction, the physical symbol system, AIXI, the Gödel machine, the free-energy principle, Domingos' Master Algorithm, reward-is-enough, autonomous machine intelligence, the thousand brains, the adjacent possible, multi-scale competency — bounded by the no-free-lunch theorems. Read as one question put to every proposal, each hands the same thing to a designer: the space of possibilities it works inside. For a collective acting in physical space, a large part of that space is the space itself. From fifty requirements for an open world we derive seven irreducible functions applied at five levels, audit the reduction against seventy-two named algorithms and sixty-three documented cases of intelligence, and then turn it around: five provisions — a writable surface, a quorum channel, complementarity by construction, containment of irreversibility, and transformation sites — move four of the seven functions out of the members and into the world. Five failure modes follow from the missing provisions, each with a measured signature in one thousand logged multi-agent episodes on a released procedural world.