4 Towards dealing with the complexity of modelling agents - modelling modelling

4.3 The modelling language

The collection of all possible model forms can be considered as a language in its broadest sense. This might correspond to a natural or formal language, but might also be something like a range of possible real-valued vectors (as in some connectionist models). What is clear is that the modelling language can critically effect the nature and "success" of the modelling process. For example if the agent has a relatively inexpressive language that can only capture some of the real possibilities then it can be genuinely surprised by observations from the environment (as distinct from merely experiencing what it considered to be a low probability event). Here it literally can not model certain combinations of possibilities, even if it can model the whole possibility space.

For example, in chaotic processes one can be forced to models of ever greater size as one approaches the critical point at onset of chaos if one restricts oneself to using finite automata but this is simply handlable when using [4]. Despite its importance in learning processes and calls for it to be further studied [20], very little is known about the effect of different modelling languages.

From Complexity to Agent Modelling and Back Again - Bruce Edmonds - 15 MAY 97

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

Generated with CERN WebMaker