Human Behavior Models and Performance Moderator Functions (PMFs) for Improving the Realism of Synthetic Bots, Crowds, Leaders

摘要

This tutorial will begin with a participative discussion of synthetic bots, avatars, and agents, including their usage to assist, train, and entertain people in simulators and in videogames. We will jointly examine the state of the practice in agent simulation and explore the need for greater realism in behavioral dimensions. What makes an agent believable? rational? emotionally appealing? entertaining? Humans are not perfectly rational agents, nor are they tireless automatons. In order to enhance the realism of synthetic avatars, one needs to include ways to model a potentially wide array of performance moderators such as heat/noise, injury, energy/fatigue, boredom and inattention, peer pressure, stress/panic, emotions, perceptual and cognitive errors, and the like. Many of these effects may be modeled via performance moderator functions (PMFs) that are simple functions indicating a stressor dosage vs. performance response. There are 100s of such PMFs in the behavioral literature, and the point of this tutorial will be to acquaint the attendee with the literature, and to help them learn how (1) to find the PMFs relevant to their scenario needs, (2) to assess the validity of given PMFs, (3) to model PMFs via hydraulic reservoir and other metaphors, (4) to integrate PMFs into their simulations and with other forms of agent modeling, (5) to attempt to experiment hands-on with a number of existing PMF models that affect physiology, endurance, reaction times, stress, emotion, attention, memory, and decision-making, and (6) to examine ways to validate the outcomes when PMFs are added to simulated bots and crowds.