# Stephanie Forrest - Echo

In collaboration with John Holland, Terry Jones, Peter Hraber, Andrew Kosoresow, and several ecologists, we are studying how genetic algorithms can be used in ecological modeling. Echo extends standard genetic algorithms in several interesting ways: (i) there is no explicit fitness function, (ii) individuals have local state, and (iii) the genetic representation is based on a higher-cardinality alphabet than binary strings. In Echo, fitness evaluation takes place implicitly. That is, individuals in the population (called {\em agents}) are allowed to make copies of themselves anytime they acquire enough resources'' to replicate their genome. Different resources are modeled by different letters of the alphabet (say, A, B, C, D), and genomes are constructed out of those same letters. However these resources can exist independently of the agent's genome, either free in the environment or stored internally by the agent. Agents acquire resources by interacting with other agents through trading relationships and combat. Echo thus relaxes the constraint that an explicit fitness function must return a numerical evaluation of each agent. This endogenous'' fitness function is much closer to the way fitness is assessed in natural settings. In addition to trade and combat, a third form of interaction between agents is mating.'' Mating provides opportunities for agents to exchange genetic material through crossover, thus creating hybrids. Mating, together with mutation, provides the mechanism for new types of agents to evolve.