Specialization versus Re-Specialization: Effects of Hebbian Learning in a Dynamic Environment

Specializing on a subset of tasks available within a system allows agents to more efficiently fulfill system demands. When demands change, agents need to Re-Specialize. Since Re-Specialization inherently requires undoing some prior Specialization, the opposing effort often results in agents settling on a worse task allocation than after Specialization, even when presented with similar demands. In this work, we demonstrate these task allocation differences by looking at how well demands are fulfilled, as well as how much task switching is happening within the system. We analyze what causes the observed differences and discuss potential approaches to improving Re-Specialization in the future.