Improving System Safety through Agent-Supported User/System Interfaces: Effects of Operator Behavior Model Ch. Santoni (*), J. M. Mercantini (*), M. F.Q.V. Turnell (**), T.I. Ören (***) (*) LSIS, Université Paul Cézanne, Marseille, France charles.santoni@lsis.org - jean-marc.mercantini@lsis.org (**) Universidade Federal de Campina Grande, Brazil fatima@dee.ufcg.edu.br (***) M&SNet: OC-MISS / SITE, University of Ottawa, Canada oren@site.uottawa.ca Abstract The article addresses improvement of safety in industrial control systems such as electric power distribution systems, through better user/system interfaces. The concepts are equally applicable to health care systems. The focus of improvement is on the operator behavior model. The representation of the knowledge about the user cognitive behavior when facing risk and critical situations is the object of a project. This project aims to produce a model of the cognitive user behavior into the requirement analysis phase of the user/system interface development. The agent support is advocated for implementation. 1. Introduction Any software consists of two parts: The main part which is the solution generator for the problem and an interface. The interface also consists of two parts: A front-end interface and a back-end interface. The front-end interface can facilitate the provision of the input, pre-processes and filters inputs from users, sensors, and/or from other software modules including software agents. The back-end interface can post-process the output of the main part of the software and provides inputs to users, actuators, and/or to other software modules including software agents. The intelligent interfaces including knowledge-based and agent-supported interfaces add several types of cognitive abilities to both front-end and back-end interfaces. The systematization of user/system interface issues can help to be complete in the user/system interface studies. For example, a taxonomy of adaptive user interfaces is given by Malinowski et al. [1993]. A systematic study of the user/system interface quality principles is developed by Ören and Yilmaz [2005]. The quality of the user/system interface can be systematically studied by taking into account user/system interface quality principles which make it possible to design as well as evaluate and compare the user/system interfaces according to the preferences of the user. They are presented in four groups: 1. The usability principles related with users (least training, minimum memory load, simplicity, and familiarity) or related with problems (separation of concerns and functionality). 2. The communicativeness principles related with users (restrained relationship with user), related with formulation and solving of problems (informativeness, perception, explanation ability), and related with display (expressiveness). 3. The reliability principles related with users (access reliability), related with usages (predictability, consistency, and safety), and related with computerization (built-in quality assurance). 4. The evolvability principles related with users (adaptability, customizability, learning ability) and related with software product (maintainability and portability). An important quality of the human/computer interfaces is reliability and safety which is vitally important issue in health care and delivery as well as in industrial control [Leveson and Turner (1993]. From the point of view of safety in operation, the higher degree of perfection and performance of industrial machinery and tools has transferred to the human operators the responsibility of almost all of the failures that happen during the interaction between them. This is the result of differences in work pace and in representation languages, which lead into misunderstandings, responsible for the majority of the accidents reported. Although the design of human interfaces acknowledges the importance of the human error and its effects on the performance and safety of both the users and the system, the application of this knowledge into the interface design remains a subject of investigation.