The research of the unit concerns the development of both theories and methods for computer-based support of decision making under uncertainty. The characteristics for this type of decision making are that decisions have to be made on the basis of insufficient or uncertain information, and that the consequences of the decisions are usually subject to uncertainty.

This kind of research is part of Artificial Intelligence. However, rather than aiming at constructing systems that "think" as humans, the research of the unit aims at constructing systems that can advise humans to act rationally. By acting rationally we mean making decisions that maximizes the expected utility (the so-called em normative approach).

Normative decision support systems are based on models of the problem domain rather than on models of the decision maker's line of reasoning; the most widely used types of normative systems are known `as Bayesian networks and influence diagrams. A normative decision support system is characterized by:

• a (graphical) representation of causal relations between quantities,
• probabilities representing the strength of the causal relations,
• utility measures representing needs and preferences and
• an application of the principle of maximum expected utility for guiding the decision making.

The unit is engaged in the development of both theories and methods for doing modeling and inference in normative systems, along with the development of theories and methods for systems that adapt to specific users and contexts. Since the mid-80s, when a significant renewed interest in normative systems arose, the unit has had a central position in the world-wide development within this field.

The results of the research have been tested through the development of systems for e.g. medical diagnosis, troubleshooting in computer equipment and agricultural planning.