The role of models in managing the uncertainty of software-intensive systems

It is increasingly argued that uncertainty is an inescapable feature of the design and operational behaviour of software-intensive systems. This paper elaborates the role of models in managing such uncertainty, in relation to evidence and claims for dependability. Personal and group models are considered with regard to abstraction, consensus and corroboration. The paper focuses on the predictive property of models, arguing for the need for empirical validation of their trustworthiness through experimentation and observation. The impact on trustworthiness of human fallibility, formality of expression and expressiveness is discussed. The paper identifies two criteria for deciding the degree of trust to be placed in a model, and hence also for choosing between models, namely accuracy and informativeness. Finally, analogy and reuse are proposed as the only means by which empirical evidence can be established for models in software engineering.