Toward Achieving Optimal Decisions Using Human Preference Modeling
Speaker
Abstract
| We should be happiest when our decisions maximize our preferences. However, people can be highly suboptimal decision makers, engaging in behavior that is in direct conflict with their stated preferences. I will discuss how suboptimality arises from erroneous assumptions people make about the structure of the environment and from an inaccurate understanding of the value of options. Overcoming some of this suboptimality may be possible using agents that learn human preferences, model human decisions, and provide individualized feedback to users about the relative value of options. I will present a proposal for such a system that involves using state-of-the-art machine learning methods to elicit preferences in the dutch flower market, a large scale supply chain problem involving a large sequence of time critical decisions. I will explain how the preferences we elicit may be used to provide feedback with the potential to improve decision-making. |
| Contact information: |
| Wolf Ketter |
