Search Papers | Poster Sessions | All Posters
Poster C125 in Poster Session C - Friday, August 9, 2024, 11:15 am – 1:15 pm, Johnson Ice Rink
Exploring How Information Shapes Human Inference from Demonstrations
Inga Ibs1 (), Jaron Schäfer1, Constantin A. Rothkopf1; 1Technical University of Darmstadt
Humans are remarkably proficient in discerning objectives and task nuances from observing others' behavior. This skill enables people to efficiently learn solutions for tasks from demonstrations. However, one strategy may be to simply imitate the observed actions while an alternative strategy is to infer the goals implicit in the observed behavior. In this study, we designed a navigational task to investigate computational accounts of how humans decide between the two strategies. Our primary objective is to investigate how the information content of provided demonstrations relates to the different ways humans infer an agent's policy, based on either inferring the agent's objectives or determining the agent's actions based on similar situations. Our results challenge prior findings and provide new perspectives for future research.
Keywords: Inverse Reinforcement Learning Decision-Making Learning from Demonstrations