Seonglae ChoInter-AI Communication, AI Social Simulation
Introducing variations to environments for one another, leading to non-stationary environments which can yield emergent behaviors.
A native approach for multi-agent communication is important, not just role-playing with formatted dialogue.
- Communication not have to be a Natural Language, Vector or Programming Language
- Like human being, Network Effect will give a huge leap for AI development
Clear role division is most important. It's best to consider the complexity and separability of tasks and only transfer necessary data between agents. Agents cannot handle multiple complex tasks at once and should be assigned one at a time. Additionally, returning results in JSON or Pydantic type formats provides significant Prompt Complexity improvements. One approach is to allow agents to generate according to Stream of consciousness initially, then have a separate agent for structuring and scheming the output.
Multi-agent AI Training Notion
Multi-agent AI Systems
intelligence explosion
AI intelligence explosion will not come from a single supermassive model, but from the social interactions between humans and many agents.
Agentic AI and the next intelligence explosion
For decades, the artificial intelligence (AI) “singularity” has been heralded as a single, titanic mind bootstrapping itself to godlike intelligence, consolidating all cognition into a cold silicon point. But this vision is almost certainly wrong in its most fundamental assumption. If AI development follows the path of previous major evolutionary transitions or “intelligence explosions,” our current step-change in computational intelligence will be plural, social, and deeply entangled with its forebears (us!).
https://www.science.org/doi/10.1126/science.aeg1895
Societies of Thought
arxiv.org
https://arxiv.org/pdf/2601.10825