← AI Terminology
Swarm Intelligence
Swarm intelligence is a field of AI inspired by the collective behaviour of social insects (ants, bees, birds) — where simple agents following local rules produce complex, intelligent global behaviour without centralised control.
Algorithms include Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), and Bee Algorithms.
Algorithms include Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), and Bee Algorithms.
Why It Matters in AI
Swarm intelligence provides powerful optimisation algorithms for problems that are hard to solve analytically: routing (ACO has been used for TSP-class problems), hyperparameter optimisation, and swarm robotics. More relevantly for modern AI, the multi-agent AI paradigm draws heavily on swarm principles — many simple LLM agents coordinating without centralised control to solve complex problems. Robotic swarms (Amazon warehouse robots, drone fleets) demonstrate real-world swarm intelligence deployment at scale.
Key Points
| Aspect | Description |
|---|---|
| ACO | Virtual ants deposit pheromones on good paths — pheromone reinforcement finds optimal routes |
| PSO | Particles explore solution space, attract toward personal best + global best positions |
| Emergence | Global intelligent behaviour from simple local agent rules — no central coordinator needed |
| Stigmergy | Indirect coordination via environment modification — ants' pheromone trails are stigmergy |
| Robotic swarms | Amazon Kiva robots, drone swarms, AUV (underwater vehicle) fleets — real production deployments |
| LLM multi-agent | Parallel LLM agents with simple coordination rules exhibit swarm-like problem decomposition |
Simple Analogy
A colony of ants finding the shortest path to food: no ant has a map, no leader directs them, but by each ant following simple rules (follow pheromone trails, leave stronger trails on shorter paths), the colony collectively discovers the optimal route. Complex intelligence emerges from simple agents.
Common Usage Examples
- PSO:
pyswarms.single.GlobalBestPSO(n_particles=100, dimensions=5, options={"c1": 0.5, "c2": 0.3, "w": 0.9}) - ACO:
pip install ant-colony; AntColony(distances, n_ants=10, n_best=5).run()— TSP optimisation - Hyperparameter search: PSO over learning rate, batch size, model architecture — parallel evaluation
- Amazon Kiva: 750K+ robots with simple local collision avoidance + central WMS — swarm warehouse
- Multi-agent AI: 50 parallel Claude agents each handling one document chunk → coordinator synthesises
Summary
In short: Swarm intelligence harnesses collective simple-agent behaviour to solve complex problems without centralised control — inspiring optimisation algorithms, robotic swarms, and the multi-agent AI paradigm where emergent collective intelligence exceeds individual agent capability.