After the community of agents solves a problem, the execution trace for each agent is evaluated and then stored in collective memory after a three step process:
The purpose of the first step is to reduce the amount of information that is stored in memory (and thus simplify the remaining steps and later adaptive effort). The second step is an attempt to remove inefficiencies in agent behavior. The third breaks the execution trace into useful pieces of information. Those pieces of information are then placed in collective memory in a manner that makes them retrievable under similar circumstances in future problem-solving situations.
Figure 2 gives streamlined execution traces for the heavy lifter (HL) and the hand-truck operator (HTO) after solving the problem of moving boxes BOX1, BOX2, and BOX4 from ROOM1 onto TRUCK1 using hand-truck HANDTR1.
The first and second boxes are loaded onto the truck in the first eight time steps: HL loads BOX4 on HANDTR1, then HL carries BOX2 to STREET1 while HTO pushes the HANDTR to the street, then HL loads the two boxes onto the truck. After this, HL goes back to ROOM1 and carries BOX1 to STREET1, intending to load it onto the truck. HTO has other ideas, though, so HTO asks HL to load BOX1 on HANDTR1. HL agrees (even though it is not very sensible) and the two agents repeat the LOAD/PUSH/UNLOAD sequence of actions.
In this episode, HL chooses to acquiesce to all requests from HTO even though it is not always more efficient to do so at the time. For this particular example, the inefficiency is removed when the trace is improved.