A survey of motivation frameworks for intelligent systems

AbstractThe ability to achieve oneʼs goals is a defining characteristic of intelligent behaviour. A great many existing theories, systems and research programmes address the problems associated with generating behaviour to achieve a goal; much fewer address the related problems of how and why goals should be generated in an intelligent artifact, and how a subset of all possible goals are selected as the focus of behaviour. It is research into these problems of motivation, which this article aims to stimulate. Building from the analysis of a scenario involving a futuristic household robot, we extend an existing account of motivation in intelligent systems to provide a framework for surveying relevant literature in AI and robotics. This framework guides us to look at the problems of encoding drives (how the needs of the system are represented), goal generation (how particular instances of goals are generated from the drives with reference to the current state), and goal selection (how the system determines which goal instances to act on). After surveying a variety of existing approaches in these terms, we build on the results of the survey to sketch a design for a new motive management framework which goes beyond the current state of the art