A Hybrid Learning Model of Tactical Decision Making
Abstract
This report is a summary of work done on a project to develop a hybrid learning architecture for decision making tasks that involve generation, revision and evaluation of explanatory hypotheses in the context of the CIC (Combat Information Center) tactical decision making task. The project studied how people acquire and use statistical information. It specifically investigated the acquisition of prior and conditional probabilities, how the order of evidence affects probabilistic decision making (i.e.. order effects), and how experience affects order effects. The experimental results were modeled using a hybrid symbolic-connectionist cognitive model that used Soar as the symbolic component and a modified version of Echo as the connectionist component. Although subjects can accurately acquire probabilities through experience with a tactical decision making task, order effects are still present without extended training. However, with extended training (i.e., experience on the task) order effects disappear. In addition, changing situations (such as from hostile to peaceful) can negatively affect decision making, but has less of an effect than predicted by normative models. The hybrid symbolic-connectionist cognitive model produced under this grant can predict the amount of training needed to eliminate order effects and to adequately prepare a person for a different situation.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 18, 2000
- Accession Number
- ADA373014
Entities
People
- Jiajie Zhang
- Todd R. Johnson
Organizations
- University of Texas Health Science Center at Houston