Why some decision aids work and others do not

A computer based decision aid was designed which resulted in no improvement in expert operator performance. Four possible reasons are explored: 1. The model which the computer used to generate a solution may be inappropriate for the problem; a skilled operator's model may be better than a normative statistical model. Data is presented to show that this may be what happened with the aid mentioned above, and task models must be chosen carefully. 2. The aid did not present any information that was not already available to the operator as stored experience. Human operators may store more information than is available to the computer, so they may draw fast, accurate conclusions without further aid. 3. Certain types of aiding may augment rather than reduce operator biases. Computer generated decisions may assume a psychological validity beyond their actual validity, and operators may attribute greater reliability to equipment than is warranted. 4. The amount of information generated by the aid overwhelmed the operators with so many possibilities that most of them were ignored.