DESIGN ISSUES IN THE SIMULATION OF AN EMERGENCY CALL CENTRE

This paper presents current work which aims to develop a computer-based simulation of an emergency call centre near Paris. The call centre is a perfect example of a complex cooperative system (several agents interacting via artifacts to achieve a common goal). As a precursor to this project, work was conducted by cognitive ergonomists to study the decisions made by the agents involved, the cognitive processes undertaken by the agents, the sequence and content of the communications and the use of artifacts in the interactions. The aim of the simulation is to study the dynamics of the situation under various environmental factors. More specifically, the simulation will be used to explore how the mutual knowledge and beliefs held by each of the agents are affected by environmental factors such as the level of noise, spatial proximity and activity of the agents. An analysis of the emergency call centre revealed that the design and subsequent implementation would be suited to an object oriented approach. An initial design has been completed. This paper presents some issues important in the design and discusses why the design will be implemented using Swarm software. DESCRIPTION OF THE CASE The emergency control centre based on the outskirts of Paris serves a region of approximately one million people. The control room is responsible for sending fire engines (which in France provide some paramedic assistance), sending ambulances, and giving over the phone medical advice. The emergency team in the control room is composed of physicians, firemen, and specially trained nurses. From an initial call a member of the team has only a few minutes to assess the nature of the incident, make a diagnosis and decide on the most appropriate course of action. This process often involves cooperation with other team members (for example, to ask for advice or to inform others of a decision). Team members communicate either directly (face to face), via artifacts (e.g. telephone or software), or non-verbally. Frequently, several incoming calls are related and thus each member has to be aware of ongoing events. This awareness is facilitated by the close spatial proximity of the team which allows each member to overhear or observe what is happening. However, during busy periods, when the noise level rises and people are otherwise occupied, a member’s general awareness of the situation is threatened. Under these circumstances, the normal way of dealing with the situation or of communicating may change. From this brief description it is easy to see that the task of call management is complex. Numerous factors affect the task (decision making under time constraints, the management of resources, ensuring efficient communication and mutual awareness, adaptive behaviour due to changes in the environment, etc.) Since such factors are common to many domains the model of the situation can be generalised and the results of the long-term work can be applied to other areas. AIMS OF THE SIMULATION The principal aim is to simulate the cooperative interactions between the people in the emergency room. The first focus will be on the mode of communication (whether direct face to face communication or a telephone is used). From observing the daily activity of the control room it was noticed that when there are many emergency incidents to deal with the noise level rises. The investigation will look at how and why the mode of communication changes with the noise level in the emergency room. During quiet periods, when the emergency team is dealing with only a few incidents, people tend to communicate directly, face to face, rather than by telephone. The mode of communication allows others to overhear what is being said and to monitor the current state of affairs. Thus, agents in the control room have mutual knowledge and beliefs. This mutual awareness is useful when agents have to ‘pick up’ a query that they were not directly involved in from the beginning. This situation occurs frequently, for example, when several callers report an incident at different times to different members of the team. A member of the team will know exactly who is handling the incident and what action has been taken. This is possible because of the mutual knowledge and beliefs. The second focus of the simulation will investigate how an agent’s mutual knowledge and beliefs change as the noise level rises. A change in knowledge or beliefs will in turn affect how an agent deals with the query. Finally, the relationship between the spatial proximity of the agents and the noise level, mutual knowledge and beliefs will be analysed. WHAT IS REQUIRED IN THE DESIGN The requirements for simulating the human agent, the task and environment and the artifact will be addressed separately in the following sections. The discussion will focus on what is required when a system contains cooperating cognitive agents. Design considerations for the agent Previous work has shown that agent cooperation is extremely dependent on the notion of mutual knowledge (MK) and mutual beliefs (MB) (Pavard et al. 1990). MK is obtained as a result of explicit communication, as in direct face to face verbal exchange. MB, on the other hand, are derived from inferences made from implicit verbal communications or from observing the actions of another agent. In turn, the ability to acquire MK and MB is determined by the spatial proximity of the agents and the cognitive abilities for inference of each agent (Pavard et al. 1990). As Pavard and his colleagues pointed out, the proximity of the agents is not sufficient to acquire MK. Rather, the ability to see and hear other agents is also needed. Therefore, in addition to the sets of MK and MB that need to be represented, it is also necessary to represent an agent’s ability to see, hear and infer. With the exception of an agent’s inferential abilities, the aforementioned sets are dynamic and require revision over time. What an agent believes or knows will change as the environment changes. As a simulation progresses, simply adding extra MK and MB to the agent’s set can cause problems, since new information may contradict what is already known. This is a problem of monotonic reasoning. To be able to retract knowledge and beliefs in light of new information is the domain of non-monotonic reasoning and belief maintenance systems (Kelleher and Smith 1988; Dugdale 1996). Design considerations for the task and environment Design considerations relating to the task are closely linked to the environment. It has been observed that how a task is performed varies according to the environment. In a typical scenario, certain tasks can happily be attributed to certain agents. However, real life is not like this, and in busy situations agents often undertake tasks and use artifacts that they are not normally supposed to. The predefined protocol breaks down, due to changes in the environment. Thus, how a task is performed (the agents involved, the artifacts used, the sequencing of actions) is largely dictated by the environment. Since an agent cannot be responsible for a particular task in all circumstances, the task can be considered somewhat separate from the agent. Thus, the effect in the design is that tasks, whilst being ultimately performed by agents, must interact closely with the environment. The status of the environment could be derived from the state of agents and artifacts, number of interactions, etc. However, if one of the aims of the simulation is to observe what happens when the environment changes, it would make more sense to model the environment explicitly as a separate entity. In this way, characteristics of the environment, such as the level of noise, could be easily modelled and calculated. Associated with a task is the cognitive ability required to perform that task. Certain tasks may require planning or inferential abilities. For an agent to be able to perform a task, the cognitive abilities of the agent must be matched with the cognitive abilities required in the task. In design, this means modelling the cognitive abilities of an agent and the cognitive requirements of a task