Emergent Dangerous Capabilities in Army Ants

Cooperation between simple agents can give rise to complex emergent dangerous capabilities: Army ants are an impressive biological example of this. During their raids they form large bridges and scaffolds from their own bodies, which allow the colony to circumvent obstacles much larger than any individual. This coordination is not organized top-down but emerges in a self-organizing way from the purely local interactions between all individuals involved in the raid.

Write up the biological example concisely; back up with a simple agent-based simulation illustrating the central message (simple agents give rise to complex emergent dangerous capability).

Review

Answers 0

No answers yet

Discussion 4

Esben Kran

Implement very simple reinforcement learning agents with simple rules etc. and show that they can easily accomplish a large-scale, high-capability task. References to Sebastian Risi's self-organized critical systems.
Esben Kran
- Ant simulations
- 56 minutes of nerding out about simulating simple agents
Esben Kran
What dangerous capabilities (DC) emerge given multi-agent setup. Scenarios:
- Using AIs for work and steering a group properly
- Bot networks and cascading fake news or proliferation through multi-agent systems
- Multi-agent AI as a group against the world situation: Activists, terrorist organizations, companies, ant colonies
  - Central organizing principle, e.g. ideology, person, CEO, contract, network protocol, etc.
- "If all the ants were LLMs": Just brute force some experiments with ChatArena
Anonymous