2 Criteria for a Good Model of Agent Learning

2.6 Expressiveness of the Internal Language of Representation

In order that the agent's may be able to change the form as well as the parameterisation of its models of its world, the agent needs to be equipped with a language with which to do this. Such a language will have a syntax in the sense of a grammar which determines the sort of expressions that are allowed as candidate internal models. It will also have a semantics in the sense of some correspondence between these expressions and their referents in the agent's environment.

This internal modelling language has to be sufficiently expressive to include all relevant models that an agent might use. In particular it should be able to express simple models like: "if x goes up so will y" (direct relation), "the rate of change of x goes up with the amount of x" (exponential growth), "the average of x is increasing", "x is not effected by anything" (this may be a default assumption), "if P(z) is true then use model x otherwise model y", "x is constant", "x or y will happen (but I don't know which)".

Modelling Learning as Modelling - 23 FEB 98

[Next] [Previous] [Up] [Top] [Contents]

Generated with CERN WebMaker