Keywords: Utility, Reinforcement Learning, Preference-Based Reinforcement Learning
Abstract: Utility is a frequently used concept to model worth and value. t is intrinsically linked with human preferences, thereby offering a potential explanation for human behavior. Many efforts across various fields have been made to modeling and learning about utility. In this essay, we will first introduce Expected Utility Theory (EUT), an important theory for understanding and modeling preferences and decision-making in economics. Then we will discuss some computational methods in RL to represent and learn human utility. Finally, we concludes that modeling utility can be a promising way towards AGI.
Submission Number: 173
Loading