An Intelligent Human-Computer Interface for Provision of On-Line Help

Some user interfaces, such as that of Unix, aredifficult for novices to use, and this paper suggestsa possible solution to such problems. The results ofa study of Unix users enabled the development of ataxonomy of error types so that users' errors can beclassified. This information is encapsulated asproduction rules within a knowledge base and forms thebasis for the design and development of an intelligentinterface to Unix. The prototype makes inferencesabout users' mental models and uses these to selectappropriate tutorial advice. Performance of users ofthe prototype intelligent interface was compared withthat of users of the usual Unix interface. Theprototype users were found to make fewer errors,exhibit fewer misconceptions and take less time tocomplete a standard set of tasks.