Applications of Human-AI Collaboration: Insights from Theory and Practice
Permanent URI for this collection
1 - 3 of 3
ItemWhen Learning Turns To Surveillance – Using Pedagogical Agents in Organizations( 2023-01-03)Workplace learning is often used to train employees systematically. New in this context is workplace learning with the help of a pedagogical agent (PA). Following Actions Design Research (ADR), this paper describes organizational training for telephone service using such PA. To develop the training, existing employee telephone service problems were analyzed, and the content of the learning program was determined based on this analysis. Subsequently, a PA was developed, implemented, and used in three municipalities. The evaluation of the learning outcome shows promising results but also yields some challenges: even though the employees improved in various aspects of the learning, they also developed a perception of surveillance. This research concludes with the formulation of design principles and suggestions for the organizational embedding of a PA in a workplace setting.
ItemThe Design of an Ostensible Human Teammate( 2023-01-03)Reliance on computer-mediated teaming has exploded in recent years, making research on how teammates calibrate their behavior critical. Here, we offer a simplistic, viable method to model human behavior for use in subsequent research investigating coordination among partners. We collected human performance data in a multiple object tracking task and a communications task to serve as the basis of our agent performance in multiple tasks. We demonstrate our model in real-time by drawing from existing research involving probabilistic models of detecting critical events and sample from a parametric log normal model of human response times to mimic human behavior. We endow our agent with team-based etiquette through a hesitancy to intervene, a parameter sampled from a uniform distribution, and manipulated agent performance through parametric shifts to detection and the log normal distribution that represents agent response times. The present work does not offer hypotheses as we did not conduct an experiment. Rather, we derive and provide a validation of an agent modeled from human performance parameters in two tasks for future team-level research with ad hoc partners.