Adaptive Agents for Fit-for-Purpose Training

Abstract

Simulators and games provide contextually rich environments, enabling learners to experience the relations between actions, events and outcomes. In order to be effective, learning situations need to be tailored to the needs of the individual learner. Virtual characters (or agents) that, in real time, select, adapt, and exhibit the behavior that is exactly right for that learner, help to establish such fit-for-purpose training. This paper discusses principles for designing training with adaptive agents, and presents a framework for their autonomous and dynamic operation. A prerequisite for agents’ adaptation of behavior to be successful is that adjustments do not violate the consistency and believability of the character, and maintains the overall narrative of the scenario. For reasons of management and coordination, it is proposed not to assign control over adaptations to virtual character-agents themselves, but to a dedicated director agent. This director agent is not a virtual character in the gameplay, but operates in the background. It collects and manages information, makes decisions about adaptations and issues behavioral instructions to the virtual characters agents. The framework was used in a pilot study, employing a human facilitator that simulated a director agent, arranging the adaptive behavior of virtual characters in a game-based training of military tactical decision making. Effects of adaptive and non-adaptive agents in a training were compared. Adaptive agents had a positive influence on learning and performance, and an increased engagement and appreciation by learners. Additional research with more participants is needed to verify these preliminary findings.