Preference acquisition through reconciliation of inconsistencies

Intelligent computer-based systems can be broadly divided into two categories--knowledge-based expert systems and preference-based decision-theoretic systems. The quality of their performance depends on their underlying knowledge bases and preference models respectively. Difficulties in acquiring these models make knowledge and preference acquisition a major focus of current research in artificial intelligence and decision theory.;This dissertation focuses on decision problems which require decision makers to make trade-offs among multiple conflicting objectives in selecting among available decision alternatives. Multiattribute utility theory provides a framework to specify trade-off preferences of the decision makers and to use these preferences in selecting the optimal decision. Traditional decision-theoretic preference-acquisition techniques are difficult to implement and rarely elicit the true preferences of the decision maker. This dissertation describes a new preference-acquisition technique for eliciting trade-off preferences of decision makers.;This dissertation makes two key contributions. First, it describes ACQUIRE--a new preference-acquisition technique--for decision-theoretic systems which evaluate competing decision alternatives and select the best decision alternative while fulfilling multiple conflicting objectives. ACQUIRE is based on reconciling inconsistencies between the preference model and the decision maker, similar to knowledge maintenance for knowledge-based systems. If the decision recommended by the preference model does not agree with the decision maker, she modifies the preference model interactively until the inconsistency is reconciled.;ACQUIRE is used to elicit the preferences of radiation oncologists for the objective evaluation of competing radiation treatment plans. The evaluation of radiation treatment plans requires radiation oncologists to make trade-offs between high radiation doses required for tumor eradication and low radiation doses required to prevent damage to healthy tissues. The second contribution of the dissertation is the development of a decision-theoretic system for evaluating competing radiation treatment plans for a patient undergoing radiotherapy. Multiattribute utility theory is used to construct the decision model. A decision-support system with an interactive graphical user interface is developed to implement ACQUIRE and the decision model.