The results of this study indicate that novice computer user s tend to prefer systems which exercise low to moderate level s of power over those systems which initiate action on thei r own . We argue that individuals want to maintain a certai n degree of control over the decisions which influence their lives . Systems which assist individuals by providin g information and recommending courses of action ar e preferred to those which are empowered to initiate action without seeking approval and confirmation from the user . In the medical domain, individuals tend to prefer the system wit h the lowest level of power -- those which provide the user wit h the greatest degree of control . This finding may generaliz e to other systems whose application could have a seriou s consequences on an individual's life or well being . This research represents the first step in identifying the types of expert systems individuals prefer. Our interest has focused on individuals who are novice users . Degree o f computer experience may moderate the relationships reported here . Future research should focus on identifyin g those domains for which individuals do and do not wan t expert systems developed . For domains where exper t systems appear useful, we need to identify the appropriat e power level for those systems, not only from a compute r science perspective, but also from the perspective of th e user . Only by being responsive to the needs and desires of users can we hope to develop systems which truly maximize the potential for human-computer interaction . Reference s Buchanan, B . G . & Shortliffe, E . (1984) . Rule-Based Expert Systems. Reading, MA : Addison-Wesley Publishing . Coovert, M . D . & Goldstein, M . (1980). Locus of control as predictor of user' attitude toward computers . Psychologica l Reports, 47, 1167-117 3 Freedman, R . S . (1987) . Guest editor's introduction . IEE E EXPERT, 2, 13 . Phares, E. (1976) . Locus of Control in Personality . Morristown, NJ : General Learning Press . Rich, E . (1981) . Artificial Intelligence . New YIork : Vantag e Address Correspondence to: Dr . Michael D. Coovert, Department of Psychology, University of South Florida, Tampa, Fl . 33620-8200 EVALUATION OF MENTAL MODEL S AND META MODELS THROUGH INTERACTION S BETWEEN USERS AND HELPERS ABOU T SOFTWARE USAGE PROBLEM S PANTO OEHDASHT I support computer use . The research question is whether a human helper provides appropriate help for the use r (problem solved in user ' s mind) if the helper understand s the problem and the software from the user's view i n addition to his/her own . A user's understanding of a software, user ' s mental model, and a helper ' s understanding of a user's understanding of that software , helper's meta model, are compared through softwar e characteristics : objects, syntax, semantics, and actions . Based on these, a set of twenty software terms wer e selected to represent repeated problems with three softwar e applications . Each user rated the terms, on a five poin t scale, for relevancy to the problem and the solution (mental model) . Users provided five additional terms t o individualize the terms to their problems . Helpers, rated the same terms based on their own understanding as well a s their understanding of users ' understanding (Meta model) . The subjects also talked about their thoughts on th e problem and the given help for qualitative analysis . Conversations between users and helpers were audio taped and later transcribed and analyzed for types of problems , problem solving approaches, and evidence of learning . Preliminary results of the term ratings show that from thos e interactions where the helper understood user's menta l model (i .e ., helper meta model matched user menta l model), about 89% of the problems were rated completel y solved by the users . When there was a clear mismatch (i .e. the helper did not understand the user's mental model) , 67% of the problems still were solved . These percentage s indicate that, when encountering small problems wit h software applications, most problems can be solve d whether the helper fully understands the user's menta l model or not . It seems unnecessary for the helper to full y understand the user in order for problems to get solved . O n the other hand, when a helper understands the user, it i s highly likely that the problem will get solved . Th e implications of these results are encouraging for design of automated help systems . It shows that a pre-designed help system has a chance o f success in helping a user even though the system will no t be specific to the user's mental model . On the other hand , it also shows that a static help system should provide ver y specific answers to various types of problems user s encounter . Based on the interactions between huma n helpers and users, one possible conclusion is that "an automated help system has to be detailed, but can still b e static " . The qualitative analysis of the results will furthe r concentrate on the specific differences between th e interactions and the types of problems users encountered . *Parto Dehdashti is now at NeuroComp, Inc ., Bal a Cynwyd, Pa . This research investigates the concept of mental models b y concentrating on the conversations between student users who as a result of encountering a problem on a software ask for assistance from student helpers who are assigned to 33 SIGCHI Bulletin October 1988 Volume 20, Number 2

/lp/association-for-computing-machinery/evaluation-of-mental-models-and-meta-models-through-interactions-Qqqs9SdWpM