The role of instruction and verbalization in improving performance on complex search tasks

Abstract
This paper examines methods of improving human search performance on a diagnostic task where it does not help to provide computer suggestions about the next enquiry to make. In three experiments it was found (a) that verbal instruction in optimal procedures was ineffective in changing actual performance, although it changed answers to verbal tests of knowledge; (b) that requiring people to say aloud the reasons for each action was ineffective in changing either performance or verbal tests of knowledge; but if people were given both verbal instructions and the requirement to justify each action aloud, performance was improved; (c) this successful training method changed performance not merely on the specific task that was trained, but also on a superficially different search task in which the same general procedures were optimal. These findings suggest that human decision processes change if key information is temporarily activated at the time it is needed, but not if it is merely learned at an irrelevant time. Such a process also explains the beneficial effect of interfaces that provide explanation or the results of inference at key points in the task.