Observations of Society
Main Menu Previous Topic Next Topic
The Effect of the Initial State
One possible objection to this model involves the initial state where societies begin. In particular, our decision to initialize the code with all neutral beliefs is suspect. Organizations and societies in the real world are rarely created with a "cultural blank state," so to speak. Here, we repeat the previous graph, except by starting each society with a code having fully random beliefs:
Clearly, the curves for moderate and high rates of code learning are relatively unaffected by the change. However, the curve corresponding to slow code learning has clearly shifted. It is now the case that slow socialization is preferable to fast socialization under all circumstances. It is also true that higher rates of code learning now perform better for all levels of socialization.
What causes the significant change in the slow code learning curve? We might speculate that for low rates of code learning, only a few belief components can change at a time. Thus, if the code starts with all neutral beliefs, only a few dimensions become positive or negative at a time. If the agents learn quickly, they are capable of quickly internalizing this change, so that once again, there is a certain stability between the agents and the code. The next time the code changes, it will learn from agents that have already internalized all the information the code holds. In effect, the code receives instant feedback from the changes that it makes. In a sense, this is a more "controlled" learning process. Under this set of conditions, quick learning agents have a certain extra advantage over slow learning ones.
The difference between the two graphs may not be interesting in itself, however it does point out the vulnerability of the results to the initial conditions we impose on the organization. It also brings to light the fact that under varying conditions, the process of learning can act in ways that appear very different from each other.
Previous Slide Next Slide