The Invalidity of Validating Emotional Multi-Agent Systems Simulations

This paper discusses the issue of validating simulations that involve emotional agents. In disaster management scenaria, such as emergency evacuations, simulations are developed to demonstrate the behaviour of people acting under emotional pressure. It is implied that the models of agents in such simulations demand the integration of emotions and how these affect individuals. An issue present in any model development is to prove the validity of the model. More precisely, it is necessary to create experiments with models comparing their behaviour to known situations, e.g. real incidents that have been recorded in the past. Such comparisons usually involve calibrating model parameters with values so that model output maps to the real data acquired, as for example data from video footage. However, modelling emotional agents is an extremely challenging task because behaviour under the influence of emotions is affected by numerous factors, e.g. personality trait, emotion contagion, distortion of perception, noisy communication etc. Although there are various models that address these issues separately, there are very few models that integrate all of them. It is therefore invalid to calibrate partial emotional models against real data, and consequently, the similar to the real world behaviour obtained, would be a result of a careful selection of parameters. The present work aims at presenting arguments that believability should be used instead of validity, since the later will require a complete implementation of emotional agents to compare with complete real data.