# CDM Seminar Series 2003-04

## Abstract:

### Preferences and Utility in Decision Theory

Michael McGeachie

Lab for Computer Science and Artificial Intelligence

MIT

Thursday Nov 13, 2:00pm

### From Preferences to Decision Making by way of Utility Functions

Decision making systems require methods for concisely representing and

efficiently reasoning with human preferences. This talk will cover

three stages of work achieving this goal. First we discuss ceteris

paribus (all-else equal) preference statements as one such

representation of natural and intuitive human preferences over outcomes

or goals. Although deduction in a logic of such preference statements

can compare the desirability of specific conditions or goals, many

decision-making methods require numerical measures of degrees of

desirability. Second, to permit ceteris paribus specifications of

preferences while providing quantitative comparisons, we present an

algorithm that compiles a set of qualitative ceteris paribus preferences

into an ordinal utility function. Constructing the utility function

can, in the worst case, take time exponential in the number of features.

Finally, we introduce common utility-independence conditions that reduce

the computational burden of constructing and evaluating the presented

utility function. Our heuristics use utility independence coupled with

constraint-based search to obtain efficient utility functions. As a

whole, our methods form a reasoning system that goes from abstract,

logical preferences to decision making by using concrete utility

functions. Our system is always correct and, under common conditions,

computationally efficient.