Published on 12/23/2021
The Computer-Human Allocation of Resources Testbed (CHART) allows human-agent teams to collaborate on a crime mapping task, in which participants search through historical spatiotemporal crime data in a metropolitan area and are instructed to allocate a limited number of crime prevention resources throughout the city. The interface consists of two displays: (i) the right-hand display is an interactive crime map that allows each user to independently select a range of past dates and categories of crimes (e.g., traffic incidents, assault, theft) to overlay publicly-available data on a map of the given city; (ii) the left-hand display consists of a shared map on which team members can place their resources (represented by colored pins) as well as a sidebar detailing the team’s current score on the task and each individual’s contribution. Participants are tasked with allocating resources for a target date (e.g., July 4th) by placing, moving, and deleting pins on the shared map. Teams are scored on their allocations based on the number of actual crimes that occured on the target date. The task automatically moves users through several timed rounds, presenting surveys in-between each round.
CHART is exceptionally flexible and can be tuned to accommodate a wide range of studies and manipulations. By specifying individual fields in a configuration file, researchers can easily change the number of human and agent teammates, the number of rounds in the task, and the length of each round. Additionally, each team member can be assigned a specific subset of crime categories to which they have access. For example, one participant may have exclusive access to data on aggravated assault, arson, burglary, traffic accidents, theft-from-motor-vehicle, and auto-theft, while participants 2 has exclusive access to larceny, murder, robbery, drug-alcohol, and public disorder incidents. Thus, teams must depend on communication and coordination to succeed. CHART also allows for the specification of one or more wizard-of-oz style ‘AI’ agents, which can place pins and make recommendations for teammates during each round. The agent’s actions can be completely customized to include behavioral and performance manipulations.
CHART logs each action performed during rounds, as well as team scores and precise timestamps. Thus, testbed data to be synced with data from external sensor recordings as long as computer timestamps are available, allowing for precise comparison of user behavior and sensor information.
To receive free access, contact email@example.com with an email explaining you and your request.