Binary Decision Automata Modelling Stress in the Workplace

Matthew Page and Daniel Ashlock
Submitted to CEC 2013

Abstract PDF eprint

This study builds on previous work modeling stress in the workplace. It incorporates a new and more sophisticated agent representation called a binary decision automata. Agent training uses inaccurate mimetic behaviour to adopt the successful behaviour of highly productive mentors. There are three tasks an agent can undertake; rest, a base job and a special project. The relative worth of these tasks vary stochastically week-to-week representing the changing priorities of management. Stress is accumulated through working long hours and impacts performance of the agent by decreasing productivity. Covert drug use is implemented into the model through the incorporation of a few individuals with much higher stress tolerance than the base agents. Binary decision automata have substantially greater learning capabilities, reflected in the increased productivity and lower overall monthly firings compared to previous research that used a simple string representation for agents. Moreover, with the inclusion of cover drug use amongst agents, the binary decision automata have the capabilities to learn effective behaviour and adapt to the challenging demands of the high performing drug agent mentors. This is in sharp contradistinction to the string agents.