A utility-based theory of initiative in mixed-initiative systems

In this paper, we present a utility-based decision making process to be used by a system in determining when to take the initiative to interact with a user, in a mixed-initiative artificial intelligence system. The decision making is based on a calculation of the expected utility of various courses of action and the likelihood that the user will be an effective contributor of information, if an interaction were initiated. We demonstrate the model in the application of sports scheduling, but discuss its potential for other applications as well. In particular, we contrast with existing work on utility-based reasoning in environments of agent-agent interaction. Our overall conclusion is that there is value to adopting an explicit reasoning process about taking the initiative as part of the overall deliberation about problem solving in collaborative environments.