When Agents Emerge from Agents: Introducing Multi-scale Viewpoints in Multi-agent Simulations