2 The effects of complexity on modelling by agents

2.3 Ideal rationality and inadequate information

Let us increase the prediction complexity further by now only giving it inadequate (i.e. noisy and insufficient) information about its environment. Let us assume that the agent can posit deliberately imprecise models, i.e. it can include room for inexact predictions using some mechanism like error terms in its language of representation*1. Now the most appropriate model an agent can infer will typically be, at least somewhat, imprecise, so that as well as accuracy the agent also has to take into account the specificity of its models. Since the agent has insufficient information it has no way of certainly distinguishing noisy data from very complex behaviour so there will be an inevitable trade-off between the accuracy and specificity of candidate models. Although some [19] have argued that particular trade-offs can be principled, in general the nature of this trade-off will depend on the goals of the agent (e.g. its tolerance to risk).

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

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

Generated with CERN WebMaker