Manual Optimization of Ill-Structured Problems

This study investigates the question, "Are human operators good optimizers in ill-structured industrial problems compared to a random-automatic optimizer?" Different problem situations are investigated. Changes in the performance and behavior of human operators are viewed with these situational changes relative to changes in the random-automatic method. The human operators observed in this study greatly outperformed the random-automatic method in limited trials when there were a greater number of variables to be optimized and very few irregularities in the reported payoff to the optimizer. Noise in the reported payoff seriously reduced the performance of these operators. Additional trials for searching did not appear to aid in obtaining significantly better performance but did cause behavioral changes. Irregularities due to interactions in the function being optimized also reduced performance.