Argues that, although expert systems--which are supposed to give users all the advantages of consulting with human experts--can be useful for medical diagnosis, where tests tend to be reliable, they can be hazardous in such areas as psychological testing, where test reliability is difficult to measure. (MBR)