MetaAgents and agent activation based on networks

[tpike3]

With AgentSet being firmly embedded and excellent features like groupby I think we can leverage these to allow the dynamic creation of meta agents fairly easily. I already started some testing and have a path forward but wanted to get thoughts to see if it is valid/ worth the effort.

Conceptual Foundation: Complex systems are adaptive networks. Using the brain as a metaphor, when we receive stimulus this activates groups of neurons to produce thoughts and behaviors. At any given time we have active and dormant neural networks.

Question: How can we provide features that effectively keep sets of linked agents, which given stimulus in the model (ideally endogenously) activates these networks?

Proposed Approach: Add a function to AgentSet that dynamically creates new agent types that are classes of these linked agents.

Slightly More than Pseudo Code:

def create_metaagent(self, new_agent_class_name, list_of_agents, **unique_attributes_behaviors):
      #Get agent types
      agent_types =(agent.__class__ for agent in list_of_agents)      
      #Create new class
      meta_agent_class = type(new_agent_class_name, 
                                      (agent_types,), 
                                       **list_of_belonging_agents_as_an_attribute and unique_attributes_behaviors)
      #Add to globals
      globals()[new_agent_class_name]   

Some optional features may be a kwarg ensuring levels named as ints but could represent individuals, families, neighborhoods or team, league etc. A kwarg on whether on not any individual agent could be in multiple groups at the same level. Potentially, if not agents are restricted to one meta agent at any level this would facilitate for multithreading in Python 3.13.

Curious on thoughts?

[EwoutH]

Very interesting topic! I once did a project in which households were very important, Households made car purchase and sharing decisions together. This concept seems to help/enable that.

One initial question: From my understanding, emergence in complex systems is fundamentally about patterns and behaviors arising naturally from the interactions of lower-level components, without being explicitly programmed or structured from above. This proposal seems to suggest that it can be build explicitly. Could you expand a little on what your vision is in this regard?

[tpike3]

@EwoutH Sorry for the delay, thanks for the question. I would say this allows for bottom up emergence and policy at multiple levels.

Generally, I would say complex systems often focus on bottom up emergence for a wide variety of scientific and historical reasons. However, to quote David Krakauer from SFI -- Complex systems are adaptive networks. So, societies emerge from individuals but then those governing organizations impose laws that shape individuals behavior. Similarly, entities within an ecosystem create an emergent ecosystem, but that ecosystem then shapes the fitness landscape on which entities within that ecosystem adapt. For me complex systems is about the flow of dynamics across networks and associated adaptions and subsequent flows.

The intent of this is to provide more options for both bottom up emergence, but also for the emergence of meta agents that can shape policies.

This paper is pretty rough as I did it several years ago, but it was my first foray into this topic.

https://arxiv.org/pdf/1904.08315

You can also see I didn't quite get sugarscape with traders right, but later for the Complexity Explorer tutorial @rht and I found the issues.

[EwoutH]

Interesting, thanks! Will dive into it.

[tpike3]

See #2561 for initial implementation -- to start this currently does not allow for agents to belong multiple meta-agents due to the exponential growth of potential combinations, however future iterations can involve such options