Conclusions from Learning Experiments


The results on learning deeper agent models show that the UMDL SMS benefits from the existence of agents that have the capability of keeping deeper models and of looking out for their best interests. Agents are encouraged to model other because this can bring them higher profit. However, there is also a computational cost associated with modeling, along with a decreasing return as other agents also start to build models. This means that that the SMS will likely evolve to a point where some agents build models, some of the time. This is a great scenario because it gives us a very robust system (i.e. one that can not be sabotaged by deviant agents), while using few resources for this purpose and distributing the resources it uses among the agents.

Jose M. Vidal
Tue Sep 30 14:35:40 EDT 1997