9 June 2011

Post date: Jun 9, 2011 3:48:35 PM

Discussion led by Elisabeth zu Erbach Schoenberg on good practice in the use of complexity science models that explain higher-level system properties by supposing they are made up of simple agents or components. Is it always necessary to identify macro-states in the higher-level system, i.e., to do classification? If we do such classification, is picking the best model at the lower level all about choosing the one that best matches the classification of real-system states over time? Does this view put too much stress on prediction: a very difficult goal for some complex systems. Or does the right coarse-grained level of classification mean that we can aim for prediction after all? What is the role of sensitivity analyses here? Are sensitivity analyses perhaps over-estimated: might it be more important to use the available computational resources to investigate structurally different models rather than many parameter-variants for one model?