Knowledge-based graphical support in human-machine systems: an empirical example from network management

The experiment reported focuses on the role of confirmatory and disconformatory evidence during fault diagnosis of telecommunication networks. Subjects were trained to identify four fault types by reasoning about the qualitative values and rates of change of pairs of process variables graphically represented by line graphs. The process variable pairs consisted of only confirmatory evidence or a mixture of confirmatory and disconfirmatory evidence. Results showed that subjects initially reasoned more efficiently using confirmatory evidence alone, but eventually achieved equal performance with mixed confirmatory and disconfirmatory evidence. The authors discuss the results in terms of their implications for use of confirmatory and disconfirmatory evidence as decision aids for process control