Human performance and strategy in a two-variable optimization task

A two-variable optimization was used as a laboratory simulation of an industrial process control task. Human performance, measured by the number of steps needed to reach the optimum, was compared with performance of a non-linear optimization over a range of information quality conditions, obtained both by altering the precision and reliability of feedback information. Strategy was measured by observing step size and direction, and also by a post experimental questionnaire. Human performance was superior to algorithm performance, especially with degraded feedback information. Human observed and reported strategies were highly correlated. The subject's strategy could be modelled as a two-stage optimization where elimination of unlikely areas was followed by construction of a mental map of the region around the optimum.