The Computational Foundations of Collective Intelligence (arxiv.org)

A new theoretical paper builds a computational framework explaining why and how collectives outperform individuals on certain tasks. The authors show that collectives’ advantages—more sensory channels, memory, processing capacity and action options—naturally give rise to familiar phenomena such as the wisdom of crowds, collective sensing, division of labour and cultural learning. Crucially, the framework treats collectives as distributed, modular systems and highlights the trade-offs introduced by coordination and cooperation; it derives testable predictions for collective capabilities in distributed reasoning and context-dependent behavioural switching, and it grounds those predictions with case studies from animal navigation and decision-making. For AI/ML practitioners this matters because it reframes collective intelligence not as simple scaling of single-agent algorithms but as qualitatively different computation enabled by distribution and modularity. The paper suggests concrete implications for multi-agent learning, swarm robotics, federated/ensemble models and decentralized sensing: design choices around communication, specialization, memory distribution and switching policies will determine whether extra resources translate into better performance or costly coordination overhead. The testable predictions provide targets for empirical benchmarks and new algorithms that exploit context-dependent strategy switching and distributed inference rather than only enlarging single-agent capacity.