This study explores challenges of system engineers with respect to the multidisciplinary development of an intelligent document composition system called Probabilistic Document Model (PDM). Implementing intelligent entities that automate human-like decision and performance requires behavior model of subject matter experts (SMEs) whose knowledge could be unfamiliar to system engineers. To include SMEs knowledge in PDM machine learning algorithm, professional graphic designers participate in development by creating templates that demonstrate their design aesthetics on page composition. By comparing such multidisciplinary activity of PDM team (Group 2) and more natural design activity of professional graphic designers in a work setting (Group 1), we examine how successfully PDM algorithm echoes professional graphic designers' knowledge in action. The result revealed that Group 1 designers made the decision through planning, pagination, and evaluation. However, Group 2 designers activity was limited to pagination because their involvement was guided by engineers interpretation on SMEs knowledge. Among 24 design heuristics demonstrated by Group 1 designers, 37.5% were supported by PDM. Some simple heuristics (33.33%) were not included in PDM because engineers did not perceive such heuristics, while other sophisticated heuristics (29.17%) were not implemented due to the difficulty to transfer abstract design concept to explicit code. The loss of such heuristics implies the high cost of communication barrier that engineers need to hurdle in the development of intelligent system.
PDF
download
878987797
028-85220240
