Fokoué, Babbitt, and Leventhal prove that ant colonies and random forests are not merely analogous but mathematically isomorphic. The mappings are explicit: pheromone trail reinforcement corresponds to out-of-bag error estimation, quorum sensing maps to prediction averaging, recruitment rates correspond to tree weightings. The variance reduction strategy is identical — both systems use controlled randomness to decorrelate identical units, then aggregate their outputs.
The formula compresses to a single principle: randomized identical agents plus diversity-enforcing mechanisms yields emergent optimality. In the forest, bootstrap aggregation and random feature subsampling decorrelate decision trees that would otherwise converge to the same splits. In the colony, stochastic response to local cues combined with positive feedback decorrelates foragers that would otherwise converge to the same trail. Neither system needs a coordinator. Neither system needs its agents to be individually optimal. The optimality is a statistical property of the ensemble, not a computational property of any agent.
The structural claim is stronger than “these systems are similar.” An isomorphism means the proof that one system works is automatically a proof that the other works. If you prove that random forests converge to optimal classification given sufficient trees, you have simultaneously proved that sufficiently large ant colonies converge to optimal foraging. The mathematics doesn't care which domain generated the theorem.
What makes this surprising is not the connection but the specificity. Vague analogies between swarms and ensembles have circulated for decades. The contribution here is showing the correspondence holds at the level of individual mechanisms — not “both aggregate” but “this specific biological process maps to this specific statistical operation, and the mapping preserves the variance reduction proof.”