Abstract

Software agents are the result of the development in the fields of human computer interaction, artificial intelligent and cognitive science. Recently it was the focus of both computer science research and the software industry. It has become clear that agents can take many roles in helping people complete their tasks. But, agents in real life have not been perfect things that can satisfy everybody. Some problems related to agent's usability have arisen, such as, whether an agent can provide useful help without annoying its users? Whether an agent can be trusted? Building agents that will be satisfied with and trust is important issue in the agents' design. In this thesis, different approaches used in building agents and some current existing agents are introduced. Human-learning, machine-learning, and feedback learning are analysed. The thesis describes extensions made to an existing telephone agent. The enhanced telephone agent is presented by applying machine learning and feedback learning techniques. A demonstration is also given to show how the enhanced agent can make suggestions to its user unobtrusively with appropriate frequency and adjust its behaviour to support its user's daily work. Furthermore, a methodology for agents' development is derived from this case study. It is emphasised that through design and implementation to testing and maintenance, user's needs and user's satisfaction should always be the main concern of agent's designers. The thesis provides a frame for agents' development, which allows the construction of agents that meet these criteria.