A Generic Personal Assistant Agent Model for Support in Demanding Tasks

Human task performance varies over time, depending on the personal characteristics of the human, the task and the environment. High effectiveness and efficiency levels are of particular importance when tasks of an organisation are demanding and consequences of performance impairment are high. Furthermore, as a longer term aim, the human should remain healthy during the processes of task execution. We propose a personal agent that provides automated adaptive assistance to ensure proper task execution. In the past, a variety of intelligent personal assistants have been proposed to support humans during the execution of tasks (e.g. [1]). Such personal assistants usually include models that represent the state of the human and his or her tasks at particular time points (e.g., a model for the cognitive load evaluation of the human), which can be used to determine when intervention is needed. However, these assistants usually consider only some aspects of human behaviour and the tasks. As a result, a specific personal assistant may react only to particular events and states of the environment. To provide a more proper form of assistance it may be important that a larger set of models is available from which an appropriate instance can be chosen depending on circumstances. Thus, the assistant will be able to adapt its behaviour to the circumstances. This paper presents a generic design for a multi-agent system architecture for adaptive personal support during task execution. Besides being adaptive to the task, the proposed personal assistant tries to keep the operator in-the-loop, in the sense that it only intervenes when a possible problem is detected. As automated systems may reduce situation awareness (the ‘picture’ of the environment) and increase cognitive load (e.g. [2]), such an adaptive assistant provides better human support than non-adaptive systems. To realize its adaptiveness, the personal assistant agent has two modes of functioning: the selfmaintenance mode and the monitoring and guidance mode. In the self-maintenance mode the personal assistant agent is able to reason about the model specifications required to perform its tasks and to achieve its goals. Some of the agent’s goals are related to the quality of the task execution, others concern the operator’s well-being. The personal assistant agent