A new software agent `learning´ algorithm

Describes a software agent (Bradshaw 1997) learning algorithm. The philosophy this algorithm is based on is also outlined in the paper. The algorithm observes a user and, based on the user´s action, or lack of actions, will make a set of inference(s), and offer the user help/advice appropriate to the user´s current situation. Furthermore, the algorithm can take into account a user´s previous knowledge if this is made apparent by a user, while interacting with the system. The algorithm can be tailored in such a way that it can be of use to an agent in various virtual environments. The described learning algorithm has been implemented with the aim of helping novices get started with UNIX commands (Gilly, 1994). UNIX commands are typically very difficult for a novice to learn. The algorithm could be effectively modified to help novices learn other new interfaces, such as those found in command centres and control rooms. Such an algorithm would be a good tool in a training environment, as an addition to existing tools. To that end the paper considers: some of the existing related work on learning algorithms; a description of the new algorithm developed; and areas where the algorithm could help in a command/control room environment