Collective Computation
Collective computation is the process by which groups of agents — neurons, organisms, humans, algorithms — integrate distributed information to produce decisions, predictions, or actions that no individual agent could generate alone. The term emphasizes that computation is not a property of individual minds or machines but of the interactions between them: the "wisdom of crowds," the "swarm intelligence," and the "distributed cognition" that emerges from structured interaction.
The concept is broader than "collective intelligence," which often implies that groups are smarter than individuals. Collective computation is neutral about whether the group output is better or worse than individual outputs; it simply describes the mechanism by which group-level computation occurs. A lynch mob engages in collective computation; so does a scientific community. The difference lies in the architecture of the computation, not in whether it is collective.
Mechanisms
Collective computation operates through several mechanisms:
- Aggregation: Individual estimates are combined through averaging, voting, or market prices. The accuracy of aggregation depends on the diversity and independence of individual inputs. When inputs are correlated — when everyone reads the same source — aggregation fails.
- Interaction: Agents update their beliefs based on the beliefs of others, creating information cascades or consensus dynamics. The network topology determines whether interaction amplifies diversity (decentralized networks) or suppresses it (centralized networks).
- Division of labor: Different agents specialize in different aspects of the problem, and their outputs are integrated through hierarchical or modular structures. This is the mechanism of multicellular organisms, firms, and scientific disciplines.
- Competition and selection: Multiple solutions are generated and tested in parallel, with successful solutions being replicated and unsuccessful ones discarded. Markets, evolution, and machine learning ensembles all use this mechanism.
Architecture Matters
The output of collective computation depends critically on the architecture of interaction. A modular architecture, in which subgroups compute in parallel and their outputs are integrated, is robust to local failures but may miss synergies between modules. A fully connected architecture, in which everyone interacts with everyone, can find synergies but is vulnerable to cascades and groupthink.
The epistemic architecture of a collective computation system — who can observe what, who can communicate with whom, and how conflicts are resolved — determines whether the system converges on truth, on consensus, or on polarization. Designing good epistemic architectures is one of the central challenges of democratic governance, scientific organization, and distributed AI systems.