Artificial intelligence as writing: knowledge-based hypertext systems as a medium for communication

This thesis is an exploration of a new metaphor for artificial intelligence (AI). Traditionally, the computer within AI has been viewed as an agent, one with which the user engages in a conversation. More recently certain researchers have proposed the notion that artificial intelligence (and indeed computing in general) can be more appropriately seen as a form of writing. Initially this thesis reviews the literature in this area, and aspects of AI which support the approach. Features of writing are then described which show parallels with AI. This then allows us to take lessons from the history and development of both traditional writing and the new computer-based writing systems to inform the design of a new type of artificial intelligence system. A design based on these features, called Running Texts is presented through a number of small examples. Issues that arise from these and possible future developments, based on the implementation are then discussed. A rationale for users choosing to learn a system such as Running Texts is proposed, as benefits from the psychological and social implications of writing can be applied to AI systems, when they are seen as writing. The same parallels point out potential problems, and suggest new ways to see the relation between AI and thought.