Causal modeling as a basis for the design of an intelligent business problem formulation system
Paradice, David Bryan
MetadataShow full item record
Past research has shown that humans demonstrate many biases when they attempt to formulate the structure of a problem. These biases can be very detrimental to actually solving the problem at hand. In some cases, the problem formulation process may be so biased that the wrong problem is formulated entirely. Prior research has demonstrated that the process of building a model of the problem situation is beneficial to problem solution. Model construction in the domain of business problems is a very complex task, however, due to the large number of potentially relevant variables, the dynamic characteristics of business organizations, and the temporal nature of relationships in the problem domain. Much of the prior research in management information systems has assumed that any problem formulation involved has generated a correct formulation. This research explores a computer-based methodology aimed at assisting a manager in producing a correct formulation of his problem environment. A prototype system has been developed that allows a user (1) to access an organizational database of information, (2) to hypothesize and test relationships between items in the database, (3) to store relationships for use at a later session, (4) construct model based on these stored relationships, and (5) give advice regarding ways of manipulating variables in order to achieve goals. Furthermore, a rudimentary process for automatic discovery of relationships has been installed in the system. The prototype was evaluated by comparing its ability to determine relationships in the database with human subjects from a prior study. Since the database used was generated by a management simulation game, it was possible to determine the "true" relationships in the database by examining the program code for the game. The system and the student subjects agreed on approximately seventy per cent of the relationships identified. Of the remaining thirty per cent, the student subjects were correct one-third of the time, the system was correct one-third of the time, and the remainder were vaguely specified making it impossible to determine exactly which was "correct." The limitations of the methodology used herein is discussed, as well as areas for future research.