Operationalization of non-formal theories in operator-models Thomas Jrgensohn, Thomas B. Sheridan 3-354 MIT, Cambridge MA 02139, USA, thomasjmit.edu; sheridanmit.edu Abstract Quantitative and formal models of dynamic actions of humans in machine environments are usually based on formalisms that interrelate measurable or computable features of behaviour or set up relations of these features with time. Mainly engineers have set up such models for about 50 years as a tool for the development of traffic systems, control rooms of nuclear power or chemical plants, or aircraft. Problems in these models do not lie in the dynamics of action, but in the description of influences, which are cognitive, motivational, or emotional. This paper deals with the integration of such factors, which are normally described non-formally in psychological theories and models into quantitative and formal models of dynamic action explained for the modelling of motivational influences and mental models. The methodological problems are pointed out for the hypothetical example of driver assistant systems. 1 An example A typical example for formal modelling of driver behaviour is the modelling of the continuous steering of the vehicle. Such a model could be set up within the frame of developing a assisting system (e.g. obstacle assistant and predict the behaviour of the entire system). It is assumed, that the system supports the driver, but keeps him within the loop all the time. So, the driver can still be responsible for his actions. Figure. 1 Prediction of behaviour of vehicle A for sudden appearance of an obstacle B and traffic C To evaluate the system it is crucial to know whether an accident is avoided by the system in a statistical sense, or if there is the possibility that it can be harmful, for example by disturbing the driver in his actions. Since an accident here means contact with the obstacle, answering this question is only possible if the contact can be calculated independently of the situation. This requires knowledge of the vehicle position at the moment of the obstacle appearance (figure 1). As the driver can influence the system behaviour during the support phase, a prediction of the system behaviour requires a model of the operator i.e. the driver. Of course, the behaviour of the assisting system must also be computable. In the above example the driver together with the vehicle and the assisting system forms a continuous dynamic system. Let us assume that the continuous actions and reactions of the driver depend on the accuracy of the drivers situation judgment, whether he already was in the same situation before (so that he can rely on learned knowledge), whether he is timid, alert, thoughtful or hectic driver, whether he is hurried, distracted or in love. Let us call these influences mental factors. The accuracy of a model that does not account for mental factors therefore depends on the extent of the mentioned influences on the action. Mental influences are normally defined non-formally. The focus of this paper is the consideration of these non- formal factors into formal-dynamic driver models. Central is the modelling of motivation influences, because the non-formal character of motives is obvious. Based on this approach the formal modelling of mental models or mental representations will be investigated. 2 Classification of Model Types Before we embark on the central issue just pointed out, let us clarify some terminology and basic characteristics of models in general. In this paper the main distinctive feature of models is whether they are formal or non-formal. Formal modelling is characterized by a fixation of attributes of the modelled system to free elements of formalism and at the same time changing these attributes according to the rules of the formalism. As a typical feature of formal models the behaviour of the modelled object becomes known only after computation (after several steps of the formalism). For example, a model of steering behaviour by means of differential equations needs the solution of the equations using real numbers to get an idea what really happens. Of course this solution could be done purely mentally. Therefore, the difference between formal and non-formal cannot be set equal to computer based versus mental. Non-formal models are models that are not formal that means one of the main features, the fixation or the change on the basis of fixed rules, is absent. However, models can be fixated and yet be non-formal. A typical example for fixated but non-formal models are descriptive models. These are made up usually of a number of non-formal sentences about a subject and together form a model of a subject. They are fixated by the medium language, but cannot be modified according to rules. Distinctive feature of non-formal models is not necessarily the absence of formulae. Linguistic expressions translated into formulae are also