Human search procedures and the use of expert systems

Abstract

Four experiments are reported. These experiments compare the effects of two different forms of machine advice on performance on a complex search task. The task involves determining which of a set of factories is responsible for polluting a river by testing the river for the presence or absence of various pollutants. Subjects either receive a suggestion as to which pollutant they should test for on each go, or following testing for a pollutant they are told which of the factories could possibly be polluting the river at that stage, or they receive a combination of the two forms of advice. The experiments show that suggesting to subjects which pollutant they should test for on each go is of no benefit, whereas providing a list of possible factory names after testing for each pollutant significantly benefits performance. Only subjects who receive both types of advice simultaneously, however, maintain an improved performance level when the advice is withdrawn.