Evolutionary models typically rely on a single level of evolution for training a team of cooperating agents. I present a model that evolves at two levels·an ģorganizationalī level and the more traditional ģindividualī level. Each organization contains an embedded agent population that goes through a full evolutionary process every organizational time-step. The organizationēs genetic code is essentially a policy that specifies the training process for its embedded agents. It also defines the creation of a representative team that is compiled after each organizational time-step. An organizationēs fitness is based on the performance of this representative team. |
