Journal on Policy and Complex Systems
When some agents change their behavior , this alters the system for the other agents . A new context “ forces ” agents to alter their rule models as the context determines what goals , interests , and behaviors are appropriate or fit . Adaptive agents are always trying to “ fit ” with their context . When their internal rule models fit their context , the agents are successful . When their rules do not fit , the agents are not successful . System change results when innovation on the part of a subset of agents throws the system rules into flux and other agents then adapt their rule models and therefore their actions and interactions . [ ... ] The system rules , produced by agent actions and interactions , do more than constrain potential actions ; they become incorporated , through the evaluation process , into the agents ’ rule models [ emphasis added ]. ( Hoffmann , 2006 , p . 98 )
As it can be inferred from this dense block quote , adaptation is a process that shapes the behavior of agents , meta-agents , and systems . Using the grammar defined in this section , it can be said that whereas agents are the generators of the corpus , adaptation is the grammatical rule under which the generators combine to make a meaningful sentence .
To understand how adaptation works , it is useful to borrow from computer science and look at how adaptive agents are programmed in ABM and other computational models . Agents are adaptive if they can perform three procedures : performance system , credit assignment , and rule discovery ( Holland , 1995 , p . 42 ). These procedures , which happen naturally and often unconsciously in humans , are useful for the scope of this paper to describe how adaptation occurs at large .
The performance system procedure has already been introduced in the previous section in the form of internal attributes and rules of behavior . To recap , a performance system “ specifies the agent ’ s capabilities at a fixed point in time ” and what an agent “ could do in the absence of any further adaptation ” ( Holland , 1995 , p . 88 ). In the words of Hoffmann , the internal attributes and rules of behavior :
Represent the agent ’ s internal ( or subjective ) understanding of the world ( the larger system ) around them . They allow the agents to perceive and define their situation , predict the consequences of action , and act . In most applications of adaptive agents , the rules are behavioral , but they can also represent identities , interests , and goals . ( Hoffmann , 2006 , p . 98 )
The second procedure , credit assignment , requires the introduction of the concept of fitness in order to be understood . Fitness is a concept that in mathematical genetics is used to represent the “ ability of an organism to produce successful offspring ” ( Holland , 1995 , p . 65 ). In complex systems theory and computer science , fitness is instead
124