The performance of highly active problem solving stratgies in novel task environments

Abstract

A series of simulation experiments evaluated the performance of seven different rule-based problem-solving strategies. Each of the strategies was based on a small set of decision rules that used performance feedback from prior actions to determine future behavior. Several environmental task factors were studied including feedback error, complexity, and system dynamics. The results showed that different strategies perform well in some environments but not in others. No one strategy performed well across the range of environments studied here. The implications for human decision-makers are that in order to be successful across a variety of tasks, a person must possess a broad repertoire of problem solving strategies and know when and how to apply them. In addition, two laboratory experiments were conducted with human subjects using the same task factors as in the simulation experiments. The findings lend support to a new theory on problem solving in novel task environments. In stable, positive, and non-declining environments, human decision-makers employed a two-stage approach to maximizing their payoff. Behavior in the first stage was characterized by bold actions that were used to explore the environment and gain a basic understanding of the payoff distribution. Approximately one third of the way through the task, subjects changed their problem solving strategy to a more systematic, small step approach similar to the way many of the rule-based simulated subjects behaved. Another interesting result was the inability of subjects to improve their performance in their second run. Relatively minor changes to the task from one run to the next were enough to block the ability to transfer knowledge from the first run to the second. Additionally, 12% of the runs in the laboratory experiment performed at a level that was less than or equal to what could have been achieved simply by choosing settings at random. These results suggest that subjects performing at this level could have saved a considerable amount of cognitive effort by taking random actions. Additional research is needed to evaluate new task factors, alternative problem-solving strategies, and gain a better understanding of the two-stage approach.