4 Agent Models with defaults

Each of these should have a default value. Taken together, these default values effectively define some a priori information available to agents. They might know, for example, which are their decision variables and which are their goals. They might believe there is some relationship between them without knowing either what that relationship is or in what conditions any such relationship might hold. More precise default values will indicate more a priori knowledge held by agents. This a priori information may be implicitly encoded in the model structure described below in a natural way by the specification of the language in which the agent develops its models.

In many cases, the default value of the dependent-variables will be the set of target variables defined in the model of the environment and the default value of the independent-variables will be all of the decision-variables which the environment allows individual agents. The relationship will say that for any value or set of values of the decision variables, the target variables can take any (set of) feasible values. If, for example, the target is profit and the decision variable is price, then the initial (default) position is that for any feasible (i.e. non-negative) price, profit can take any rational value. The conditions of application will have by default a value indicating complete generality. Typically, this will be indicated by the simple value true since the conditions of application are either satisfied or not and, collectively, should return a boolean value.

The agents will, in general, have other a priori knowledge, including:

1. knowledge implicit in the grammar of its internal modelling language mentioned above;

2. maybe some explicit knowledge encoded as an initial model given to the agent (e.g. accounting rules);

3. knowledge encoded in a basic algorithm for learning (improving its models); and some goals.