Model-based testing and validation on artificial intelligence systems
Gang Liu, Qun Liu, Wentao Zhang · 2007
In this paper, we discuss how viewing an artificial intelligence (AI) system as a model leads to certain criteria for testing methodologies. This includes a discussion of how certain mathematical techniques for testing AI systems can be used as criteria for determining the AI System's adequacy when no other models are available. We give an example of an error due to widespread rule interactions. Such errors are the keys to understanding why the independent rule assumption does not work, and therefore why AI systems must be modeled. We examine how testing can be applied both to individual system components as well as to the system as a whole. We also submit different criteria by which a set of test cases can be assembled and the problems in determining whether or not the performance of an AI system on a set of test cases is acceptable. In the end, the article shows the results of applying this model to a real case.