We describe a new approach to the behaviour of 3D environments that supports the definition of physical processes and interactive phenomena. The work takes as a starting point the traditional event-based architecture that underlies most game engines. These systems discretise the environments' Physics by separating the objects' kinematics from the physical processes corresponding to objects interactions. This property has been used to insert a new behavioural layer, which implements AI-based simulation techniques. We introduce the rationale behind AI-based simulation and the techniques we use for qualitative Physics, as well as a new approach to world behaviour based on the induction of causal impressions. This is illustrated through several examples on a test environment. This approach has implications for the definition of complex world behaviour or non-standard physics, as required in creative applications.