Validated Probability Assessments in Strategic Settings

From Thinclabwiki

Jump to: navigation, search
What are Validated Probability Assessments in Strategic Settings ?
A snapshot of the flight simulator that is being used in the studies.

Humans, in generating probability estimations for their predictions, often lack justification. In realistic situations, such as military operations, judging uncertainty can be difficult. In addition to the complexities presented by these complex environments, humans suffer from cognitive biases, which are disparities between what is a stated belief and the actual belief of an individual.

People in strategic situations, such as UAV pilots, often assess their probabilities without any objective validation techniques. The goal of validated probability assessments in strategic settings is to establish field-valid probability assessment and update techniques by identifying potential loci of prediction errors and establishing mechanisms to counteract them.

Project Description
Three large studies are being held to assess the field-validity of the probability assessments.

We designed three experiments to evaluate cognitive biases and the mechanisms for correcting them. In each experiment, participants were asked to play a game in which they operated a virtual UAV in a simulated environment. The player was tasked with navigating a 4 by 4 grid in order to reach a goal state, given to the player at the start of each trial. In addition to reaching the goal state, a player was also required to avoid an enemy UAV that was patrolling the grid. They were given no knowledge of where the enemy was. Each player had 5 training rounds and 15 trial rounds. Success was incentivized with monetary payout, which varied from study to study.

Study 1: Players had full control over their UAVs. At each decision point, a player was asked to project their probability assessment for reaching the next sector. They were then given the option to take a random chance based on the probability assessment they provided to move without getting caught. For each decision point, if the player chose to make a move without the random event, the probability of winning the random event increases. Inversely, choosing the random event decreases the likelihood of winning the random event.

Study 2: Players had no control over their UAVs and followed a predefined trajectory. At each decision point, a player was asked to project their probability assessment for reaching the target state. Using proper and non-proper scoring rules (Brier and 0-1, respectively), players were incentivized for correct probability assessments. Another study was conducted similar to this, the only difference being that the players had full control over their UAVs.

Study 3: This study will investigate the impact of interventions such as filling questionnaires on the judgment and choice of subjects in our strategic settings.

Related Resources
Collaborating Institutions
Researchers
  • Prashant Doshi, (PI)
  • Adam Goodie, (Co-PI)
  • Dan Hall, (Co-PI)
  • Roi Ceren, Ph.D. student, CS
  • Matthew Meisel, Ph.D. student, Psychology
  • Amandah Falls, B.S. student, CSE
Personal tools