Learning to Notice: Adaptive Models of Human Operators
Emma Norling · 2007
Agent-based technologies have been used for a number of years to model human operators in complex simulated environments. The BDI agent framework has proven to be particularly suited to this sort of modelling, due to its "natural" composition of beliefs, goals and plans. However one of the weaknesses of the BDI agent model, and many other human operator models (agent-based or otherwise), is its inability to support agent learning. Human operators naturally adapt their behaviour over time, particularly to avoid repeating mistakes. This paper introduces an enhancement to the BDI framework which is based on a descriptive psychological model of decision making called "recognition-primed decision making." This enhancement allows the development of agents that adapt their behaviour in real-time, in the same manner as a person would, providing more realistic human operator models.