For example, consider a multi-agent system M that, when run, displays a recurring pattern: an exponentially increasing number of inter-agent messages abruptly followed by an almost total communication collapse. This pattern recurs indefinitely. The algorithm I am seeking would find a simplification of M, call it M~, (or several alternative such simplifications) that has essentially the same communication properties through time as M. M~ would itself be a multi-agent system.
The algorithm should be applicable to ANY multi-agent system for ANY large-scale property.
Clearly computationally precise definitions will be needed for a multi-agent system, a simplification of a MAS, large-scale behaviour, etc.
One method of precisely defining a MAS is in terms of agents that are production systems as these are defined in computer science. But, of course, there are others.
Part of the motivation for this line of investigation is to find a means to examine the possibility that the large-scale behaviour of the human race (Homo sapiens sapiens) over the past 100,000 years or so entails that human individuals have certain cognitive characteristics which might include some or all of learning, imagination, plan creation and execution, a tendency towards cooperation, aggression, and a preference for risk taking.